{ "bizCode": "MNNTest", "extraInfo": { "version": "3.1.3" }, "oplists": [ { "main_type": "Input", "main": { "dims": [ 1, 1, 5120 ], "dtype": "DT_FLOAT", "dformat": "NCHW" }, "name": "input_ids", "outputIndexes": [ 2 ], "type": "Input", "defaultDimentionFormat": "NHWC" }, { "main_type": "Input", "main": { "dims": [ 1, 1, -1, -1 ], "dtype": "DT_FLOAT", "dformat": "NCHW" }, "name": "attention_mask", "outputIndexes": [ 129 ], "type": "Input", "defaultDimentionFormat": "NHWC" }, { "main_type": "Input", "main": { "dims": [ 1, -1 ], "dtype": "DT_INT32", "dformat": "NCHW" }, "name": "position_ids", "outputIndexes": [ 36 ], "type": "Input", "defaultDimentionFormat": "NHWC" }, { "main_type": "Input", "main": { "dims": [ 64, 2, 1, -1, 8, 128 ], "dtype": "DT_FLOAT", "dformat": "NCHW" }, "name": "past_key_values", "outputIndexes": [ 0 ], "type": "Input", "defaultDimentionFormat": "NHWC" }, { "main_type": "Input", "main": { "dims": [ 1 ], "dtype": "DT_INT32", "dformat": "NCHW" }, "name": "logits_index", "outputIndexes": [ 5999 ], "type": "Input", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 0 ], "main_type": "Reshape", "main": { "dims": [ -1 ], "dimType": "NCHW" }, "name": "presents", "outputIndexes": [ 1 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 3 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 1, -1, 5120 ] }, "name": "/Constant_output_0", "outputIndexes": [ 3 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_output_0", "outputIndexes": [ 4 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm/Mul_1_output_0", "inputIndexes": [ 4 ], "outputIndexes": [ 5 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 0, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5 ], "outputIndexes": [ 6003 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6003 ], "outputIndexes": [ 6004 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/q_proj/Linear", "inputIndexes": [ 6004 ], "outputIndexes": [ 6005 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 40960, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6005 ], "outputIndexes": [ 6006 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6006 ], "outputIndexes": [ 6 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5 ], "main_type": "NONE", "name": "/Shape_output_0", "outputIndexes": [ 7 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 7 ], "main_type": "NONE", "name": "Shape52", "outputIndexes": [ 8 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 7 ], "main_type": "NONE", "name": "Rank54", "outputIndexes": [ 9 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 9, 9 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp56", "outputIndexes": [ 10 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 10 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze57", "outputIndexes": [ 11 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 1 ] }, "name": "Const9", "outputIndexes": [ 12 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 10, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp59", "outputIndexes": [ 13 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 13 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze60", "outputIndexes": [ 14 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 1 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 1 ] }, "name": "Unsqueeze62", "outputIndexes": [ 15 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 8, 11, 14, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice63", "outputIndexes": [ 16 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 16 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze64", "outputIndexes": [ 17 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 17, 17 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp66", "outputIndexes": [ 18 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 7, 18, 10 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_1_output_0", "outputIndexes": [ 19 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 1 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 0 ] }, "name": "/rotary/Constant_4_output_0", "outputIndexes": [ 20 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 19, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_output_0", "outputIndexes": [ 21 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 10 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze76", "outputIndexes": [ 22 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 13 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze79", "outputIndexes": [ 23 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 8, 22, 23, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice82", "outputIndexes": [ 24 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 24 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze83", "outputIndexes": [ 25 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 25 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp84", "outputIndexes": [ 26 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 26, 25 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp85", "outputIndexes": [ 27 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 7, 27, 10 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_2_output_0", "outputIndexes": [ 28 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 28, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1_output_0", "outputIndexes": [ 29 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 1 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 64 ] }, "name": "/Constant_3_output_0", "outputIndexes": [ 30 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 1 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 128 ] }, "name": "/Constant_4_output_0", "outputIndexes": [ 31 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 21, 29, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_output_0", "outputIndexes": [ 32 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 6, 32 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_1_output_0", "outputIndexes": [ 33 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 33 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm/Cast_output_0", "outputIndexes": [ 34 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm/Mul_1_output_0", "inputIndexes": [ 34 ], "outputIndexes": [ 35 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 23633942, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 36 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/rotary/Cast_output_0", "outputIndexes": [ 37 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 2 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ -1, 1 ] }, "name": "/rotary/Constant_1_output_0", "outputIndexes": [ 38 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 37, 38 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/rotary/Reshape_output_0", "outputIndexes": [ 39 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 64 ], "dataFormat": "NCHW", "dataType": "DT_FLOAT", "float32s": [ 1.0, 0.805842, 0.649382, 0.523299, 0.421697, 0.339821, 0.273842, 0.220673, 0.177828, 0.143301, 0.115478, 0.093057, 0.074989, 0.06043, 0.048697, 0.039242, 0.031623, 0.025483, 0.020535, 0.016548, 0.013335, 0.010746, 0.00866, 0.006978, 0.005623, 0.004532, 0.003652, 0.002943, 0.002371, 0.001911, 0.00154, 0.001241, 0.001, 0.000806, 0.000649, 0.000523, 0.000422, 0.00034, 0.000274, 0.000221, 0.000178, 0.000143, 0.000115, 9.3e-05, 7.5e-05, 6e-05, 4.9e-05, 3.9e-05, 3.2e-05, 2.5e-05, 2.1e-05, 1.7e-05, 1.3e-05, 1.1e-05, 9e-06, 7e-06, 6e-06, 5e-06, 4e-06, 3e-06, 2e-06, 2e-06, 2e-06, 1e-06 ] }, "name": "/rotary/Constant_2_output_0", "outputIndexes": [ 40 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 39, 40 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/rotary/Mul_output_0", "outputIndexes": [ 41 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 41 ], "main_type": "UnaryOp", "main": { "opType": "COS", "T": "DT_FLOAT" }, "name": "/rotary/Cos_output_0", "outputIndexes": [ 42 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 42, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/rotary/Unsqueeze_output_0", "outputIndexes": [ 43 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 41 ], "main_type": "UnaryOp", "main": { "opType": "SIN", "T": "DT_FLOAT" }, "name": "/rotary/Sin_output_0", "outputIndexes": [ 44 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 44, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/rotary/Unsqueeze_1_output_0", "outputIndexes": [ 45 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 43, 45 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/rotary/Concat_output_0", "outputIndexes": [ 46 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 46, 46 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/rotary/Concat_1_output_0", "outputIndexes": [ 47 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 1 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 2 ] }, "name": "/rotary/Constant_7_output_0", "outputIndexes": [ 48 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 47, 48 ], "main_type": "SqueezeParam", "main": {}, "name": "/rotary/Unsqueeze_2_output_0", "outputIndexes": [ 49 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 49, 15 ], "main_type": "SqueezeParam", "main": {}, "name": "/rotary/Unsqueeze_3_output_0", "outputIndexes": [ 50 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 50 ], "main_type": "NONE", "name": "Shape116", "outputIndexes": [ 51 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 50 ], "main_type": "NONE", "name": "Rank118", "outputIndexes": [ 52 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 52, 52 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp120", "outputIndexes": [ 53 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 53 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze121", "outputIndexes": [ 54 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 53, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp123", "outputIndexes": [ 55 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 55 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze124", "outputIndexes": [ 56 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 51, 54, 56, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice127", "outputIndexes": [ 57 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 57 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze128", "outputIndexes": [ 58 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 58, 58 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp130", "outputIndexes": [ 59 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 50, 59, 53 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_3_output_0", "outputIndexes": [ 60 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 35, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_output_0", "outputIndexes": [ 61 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 35 ], "main_type": "NONE", "name": "/Shape_2_output_0", "outputIndexes": [ 62 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "main_type": "Blob", "main": { "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 3 ] }, "name": "/Constant_10_output_0", "outputIndexes": [ 63 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 62 ], "main_type": "NONE", "name": "Shape258", "outputIndexes": [ 64 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 62 ], "main_type": "NONE", "name": "Rank260", "outputIndexes": [ 65 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 65, 65 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp262", "outputIndexes": [ 66 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 66 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze263", "outputIndexes": [ 67 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 66, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp265", "outputIndexes": [ 68 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 68 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze266", "outputIndexes": [ 69 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 64, 67, 69, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice269", "outputIndexes": [ 70 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 70 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze270", "outputIndexes": [ 71 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 71 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp271", "outputIndexes": [ 72 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 72, 71 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp272", "outputIndexes": [ 73 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 62, 73, 66 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_5_output_0", "outputIndexes": [ 74 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 2 ] }, "name": "/rotary/Constant_6_output_0", "outputIndexes": [ 75 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 74, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_output_0", "outputIndexes": [ 76 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 76, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_7_output_0", "outputIndexes": [ 77 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 1 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 1073741824 ] }, "name": "/Constant_19_output_0", "outputIndexes": [ 78 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 1 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 3 ] }, "name": "/Constant_13_output_0", "outputIndexes": [ 79 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 35, 77, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_1_output_0", "outputIndexes": [ 80 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 80 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_output_0", "outputIndexes": [ 81 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 76, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_6_output_0", "outputIndexes": [ 82 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 35, 20, 82, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_output_0", "outputIndexes": [ 83 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 81, 83 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_3_output_0", "outputIndexes": [ 84 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 53 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze167", "outputIndexes": [ 85 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 55 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze170", "outputIndexes": [ 86 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 51, 85, 86, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice173", "outputIndexes": [ 87 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 87 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze174", "outputIndexes": [ 88 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 88 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp175", "outputIndexes": [ 89 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 89, 88 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp176", "outputIndexes": [ 90 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 50, 90, 53 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_4_output_0", "outputIndexes": [ 91 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 84, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_1_output_0", "outputIndexes": [ 92 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 61, 92 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_output_0", "outputIndexes": [ 93 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5 ], "outputIndexes": [ 6007 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6007 ], "outputIndexes": [ 6008 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/k_proj/Linear", "inputIndexes": [ 6008 ], "outputIndexes": [ 6009 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 23634966, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6009 ], "outputIndexes": [ 6010 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6010 ], "outputIndexes": [ 94 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 19, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_2_output_0", "outputIndexes": [ 95 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 28, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_3_output_0", "outputIndexes": [ 96 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "main_type": "Blob", "main": { "dims": [ 1 ], "dataFormat": "NCHW", "dataType": "DT_INT32", "int32s": [ 8 ] }, "name": "/Constant_6_output_0", "outputIndexes": [ 97 ], "type": "Const", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 95, 96, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_1_output_0", "outputIndexes": [ 98 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 94, 98 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_2_output_0", "outputIndexes": [ 99 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 99 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm/Cast_output_0", "outputIndexes": [ 100 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm/Mul_1_output_0", "inputIndexes": [ 100 ], "outputIndexes": [ 101 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 26584108, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 101, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_2_output_0", "outputIndexes": [ 102 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 101 ], "main_type": "NONE", "name": "/Shape_3_output_0", "outputIndexes": [ 103 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 103 ], "main_type": "NONE", "name": "Shape135", "outputIndexes": [ 104 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 103 ], "main_type": "NONE", "name": "Rank137", "outputIndexes": [ 105 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 105, 105 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp139", "outputIndexes": [ 106 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 106 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze140", "outputIndexes": [ 107 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 106, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp142", "outputIndexes": [ 108 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 108 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze143", "outputIndexes": [ 109 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 104, 107, 109, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice146", "outputIndexes": [ 110 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 110 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze147", "outputIndexes": [ 111 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 111 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp148", "outputIndexes": [ 112 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 112, 111 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp149", "outputIndexes": [ 113 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 103, 113, 106 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_6_output_0", "outputIndexes": [ 114 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 114, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_1_output_0", "outputIndexes": [ 115 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 115, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_9_output_0", "outputIndexes": [ 116 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 101, 116, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_3_output_0", "outputIndexes": [ 117 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 117 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_1_output_0", "outputIndexes": [ 118 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 115, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_8_output_0", "outputIndexes": [ 119 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 101, 20, 119, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_2_output_0", "outputIndexes": [ 120 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 118, 120 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_4_output_0", "outputIndexes": [ 121 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 121, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_3_output_0", "outputIndexes": [ 122 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 102, 122 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_1_output_0", "outputIndexes": [ 123 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5 ], "outputIndexes": [ 6011 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6011 ], "outputIndexes": [ 6012 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/v_proj/Linear", "inputIndexes": [ 6012 ], "outputIndexes": [ 6013 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 26585132, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6013 ], "outputIndexes": [ 6014 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6014 ], "outputIndexes": [ 124 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 19, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_4_output_0", "outputIndexes": [ 125 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 28, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_5_output_0", "outputIndexes": [ 126 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 125, 126, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_2_output_0", "outputIndexes": [ 127 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 124, 127 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_3_output_0", "outputIndexes": [ 128 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 93, 123, 128, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_8_output_0", "outputIndexes": [ 130 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 130 ], "outputIndexes": [ 6015 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6015 ], "outputIndexes": [ 6016 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/o_proj/Linear", "inputIndexes": [ 6016 ], "outputIndexes": [ 6017 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 29534274, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6017 ], "outputIndexes": [ 6018 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6018 ], "outputIndexes": [ 131 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4, 131 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_3_output_0", "outputIndexes": [ 132 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm/Mul_1_output_0", "inputIndexes": [ 132 ], "outputIndexes": [ 133 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 53127256, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 133 ], "outputIndexes": [ 6019 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6019 ], "outputIndexes": [ 6020 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/gate_proj/Linear", "inputIndexes": [ 6020 ], "outputIndexes": [ 6021 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 53168216, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6021 ], "outputIndexes": [ 6022 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6022 ], "outputIndexes": [ 134 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 134 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn/Mul_output_0", "outputIndexes": [ 135 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 133 ], "outputIndexes": [ 6023 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6023 ], "outputIndexes": [ 6024 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/up_proj/Linear", "inputIndexes": [ 6024 ], "outputIndexes": [ 6025 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 126896238, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6025 ], "outputIndexes": [ 6026 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6026 ], "outputIndexes": [ 136 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 135, 136 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp/Mul_output_0", "outputIndexes": [ 137 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 137 ], "outputIndexes": [ 6027 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6027 ], "outputIndexes": [ 6028 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/down_proj/Linear", "inputIndexes": [ 6028 ], "outputIndexes": [ 6029 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 200624260, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6029 ], "outputIndexes": [ 6030 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.0/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6030 ], "outputIndexes": [ 138 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 132, 138 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_4_output_0", "outputIndexes": [ 139 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 139, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_9_output_0", "outputIndexes": [ 140 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_1/Mul_1_output_0", "inputIndexes": [ 140 ], "outputIndexes": [ 141 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 274352282, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 141 ], "outputIndexes": [ 6031 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6031 ], "outputIndexes": [ 6032 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/q_proj/Linear", "inputIndexes": [ 6032 ], "outputIndexes": [ 6033 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 274393242, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6033 ], "outputIndexes": [ 6034 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6034 ], "outputIndexes": [ 142 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 141 ], "main_type": "NONE", "name": "/Shape_14_output_0", "outputIndexes": [ 143 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 143 ], "main_type": "NONE", "name": "Shape502", "outputIndexes": [ 144 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 143 ], "main_type": "NONE", "name": "Rank504", "outputIndexes": [ 145 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 145, 145 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp506", "outputIndexes": [ 146 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 146 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze507", "outputIndexes": [ 147 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 146, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp509", "outputIndexes": [ 148 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 148 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze510", "outputIndexes": [ 149 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 144, 147, 149, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice513", "outputIndexes": [ 150 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 150 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze514", "outputIndexes": [ 151 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 151, 151 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp516", "outputIndexes": [ 152 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 143, 152, 146 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_18_output_0", "outputIndexes": [ 153 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 153, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_32_output_0", "outputIndexes": [ 154 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 146 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze524", "outputIndexes": [ 155 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 148 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze527", "outputIndexes": [ 156 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 144, 155, 156, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice530", "outputIndexes": [ 157 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 157 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze531", "outputIndexes": [ 158 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 158 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp532", "outputIndexes": [ 159 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 159, 158 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp533", "outputIndexes": [ 160 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 143, 160, 146 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_19_output_0", "outputIndexes": [ 161 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 161, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_33_output_0", "outputIndexes": [ 162 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 154, 162, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_13_output_0", "outputIndexes": [ 163 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 142, 163 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_10_output_0", "outputIndexes": [ 164 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 164 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_1/Cast_output_0", "outputIndexes": [ 165 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_1/Mul_1_output_0", "inputIndexes": [ 165 ], "outputIndexes": [ 166 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 297986224, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 166, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_8_output_0", "outputIndexes": [ 167 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 166 ], "main_type": "NONE", "name": "/Shape_16_output_0", "outputIndexes": [ 168 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 168 ], "main_type": "NONE", "name": "Shape651", "outputIndexes": [ 169 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 168 ], "main_type": "NONE", "name": "Rank653", "outputIndexes": [ 170 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 170, 170 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp655", "outputIndexes": [ 171 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 171 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze656", "outputIndexes": [ 172 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 171, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp658", "outputIndexes": [ 173 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 173 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze659", "outputIndexes": [ 174 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 169, 172, 174, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice662", "outputIndexes": [ 175 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 175 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze663", "outputIndexes": [ 176 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 176 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp664", "outputIndexes": [ 177 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 177, 176 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp665", "outputIndexes": [ 178 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 168, 178, 171 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_20_output_0", "outputIndexes": [ 179 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 179, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_3_output_0", "outputIndexes": [ 180 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 180, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_39_output_0", "outputIndexes": [ 181 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 166, 181, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_5_output_0", "outputIndexes": [ 182 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 182 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_2_output_0", "outputIndexes": [ 183 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 180, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_38_output_0", "outputIndexes": [ 184 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 166, 20, 184, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_4_output_0", "outputIndexes": [ 185 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 183, 185 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_16_output_0", "outputIndexes": [ 186 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 186, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_9_output_0", "outputIndexes": [ 187 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 167, 187 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_5_output_0", "outputIndexes": [ 188 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 141 ], "outputIndexes": [ 6035 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6035 ], "outputIndexes": [ 6036 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/k_proj/Linear", "inputIndexes": [ 6036 ], "outputIndexes": [ 6037 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 297987248, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6037 ], "outputIndexes": [ 6038 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6038 ], "outputIndexes": [ 189 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 153, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_34_output_0", "outputIndexes": [ 190 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 161, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_35_output_0", "outputIndexes": [ 191 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 190, 191, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_14_output_0", "outputIndexes": [ 192 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 189, 192 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_11_output_0", "outputIndexes": [ 193 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 193 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_1/Cast_output_0", "outputIndexes": [ 194 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_1/Mul_1_output_0", "inputIndexes": [ 194 ], "outputIndexes": [ 195 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 300936390, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 195, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_10_output_0", "outputIndexes": [ 196 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 195 ], "main_type": "NONE", "name": "/Shape_17_output_0", "outputIndexes": [ 197 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 197 ], "main_type": "NONE", "name": "Shape548", "outputIndexes": [ 198 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 197 ], "main_type": "NONE", "name": "Rank550", "outputIndexes": [ 199 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 199, 199 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp552", "outputIndexes": [ 200 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 200 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze553", "outputIndexes": [ 201 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 200, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp555", "outputIndexes": [ 202 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 202 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze556", "outputIndexes": [ 203 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 198, 201, 203, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice559", "outputIndexes": [ 204 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 204 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze560", "outputIndexes": [ 205 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 205 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp561", "outputIndexes": [ 206 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 206, 205 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp562", "outputIndexes": [ 207 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 197, 207, 200 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_21_output_0", "outputIndexes": [ 208 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 208, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_4_output_0", "outputIndexes": [ 209 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 209, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_41_output_0", "outputIndexes": [ 210 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 195, 210, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_7_output_0", "outputIndexes": [ 211 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 211 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_3_output_0", "outputIndexes": [ 212 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 209, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_40_output_0", "outputIndexes": [ 213 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 195, 20, 213, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_6_output_0", "outputIndexes": [ 214 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 212, 214 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_17_output_0", "outputIndexes": [ 215 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 215, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_11_output_0", "outputIndexes": [ 216 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 196, 216 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_6_output_0", "outputIndexes": [ 217 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 141 ], "outputIndexes": [ 6039 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6039 ], "outputIndexes": [ 6040 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/v_proj/Linear", "inputIndexes": [ 6040 ], "outputIndexes": [ 6041 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 300937414, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6041 ], "outputIndexes": [ 6042 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6042 ], "outputIndexes": [ 218 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 153, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_36_output_0", "outputIndexes": [ 219 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 161, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_37_output_0", "outputIndexes": [ 220 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 219, 220, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_15_output_0", "outputIndexes": [ 221 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 218, 221 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_12_output_0", "outputIndexes": [ 222 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 188, 217, 222, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_17_output_0", "outputIndexes": [ 223 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 223 ], "outputIndexes": [ 6043 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6043 ], "outputIndexes": [ 6044 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/o_proj/Linear", "inputIndexes": [ 6044 ], "outputIndexes": [ 6045 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 303886556, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6045 ], "outputIndexes": [ 6046 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6046 ], "outputIndexes": [ 224 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 140, 224 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_8_output_0", "outputIndexes": [ 225 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_1/Mul_1_output_0", "inputIndexes": [ 225 ], "outputIndexes": [ 226 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 327479538, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 226 ], "outputIndexes": [ 6047 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6047 ], "outputIndexes": [ 6048 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/gate_proj/Linear", "inputIndexes": [ 6048 ], "outputIndexes": [ 6049 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 327520498, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6049 ], "outputIndexes": [ 6050 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6050 ], "outputIndexes": [ 227 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 227 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_1/Mul_output_0", "outputIndexes": [ 228 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 226 ], "outputIndexes": [ 6051 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6051 ], "outputIndexes": [ 6052 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/up_proj/Linear", "inputIndexes": [ 6052 ], "outputIndexes": [ 6053 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 401248520, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6053 ], "outputIndexes": [ 6054 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6054 ], "outputIndexes": [ 229 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 228, 229 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_1/Mul_output_0", "outputIndexes": [ 230 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 230 ], "outputIndexes": [ 6055 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6055 ], "outputIndexes": [ 6056 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/down_proj/Linear", "inputIndexes": [ 6056 ], "outputIndexes": [ 6057 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 474976542, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6057 ], "outputIndexes": [ 6058 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.1/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6058 ], "outputIndexes": [ 231 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 225, 231 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_9_output_0", "outputIndexes": [ 232 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 232, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_18_output_0", "outputIndexes": [ 233 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_2/Mul_1_output_0", "inputIndexes": [ 233 ], "outputIndexes": [ 234 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 548704564, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 234 ], "outputIndexes": [ 6059 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6059 ], "outputIndexes": [ 6060 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/q_proj/Linear", "inputIndexes": [ 6060 ], "outputIndexes": [ 6061 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 548745524, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6061 ], "outputIndexes": [ 6062 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6062 ], "outputIndexes": [ 235 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 234 ], "main_type": "NONE", "name": "/Shape_28_output_0", "outputIndexes": [ 236 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 236 ], "main_type": "NONE", "name": "Shape890", "outputIndexes": [ 237 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 236 ], "main_type": "NONE", "name": "Rank892", "outputIndexes": [ 238 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 238, 238 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp894", "outputIndexes": [ 239 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 239 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze895", "outputIndexes": [ 240 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 239, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp897", "outputIndexes": [ 241 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 241 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze898", "outputIndexes": [ 242 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 237, 240, 242, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice901", "outputIndexes": [ 243 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 243 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze902", "outputIndexes": [ 244 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 244, 244 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp904", "outputIndexes": [ 245 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 236, 245, 239 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_33_output_0", "outputIndexes": [ 246 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 246, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_64_output_0", "outputIndexes": [ 247 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 239 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze912", "outputIndexes": [ 248 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 241 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze915", "outputIndexes": [ 249 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 237, 248, 249, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice918", "outputIndexes": [ 250 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 250 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze919", "outputIndexes": [ 251 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 251 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp920", "outputIndexes": [ 252 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 252, 251 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp921", "outputIndexes": [ 253 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 236, 253, 239 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_34_output_0", "outputIndexes": [ 254 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 254, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_65_output_0", "outputIndexes": [ 255 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 247, 255, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_26_output_0", "outputIndexes": [ 256 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 235, 256 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_19_output_0", "outputIndexes": [ 257 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 257 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_2/Cast_output_0", "outputIndexes": [ 258 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_2/Mul_1_output_0", "inputIndexes": [ 258 ], "outputIndexes": [ 259 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 572338506, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 259, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_16_output_0", "outputIndexes": [ 260 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 259 ], "main_type": "NONE", "name": "/Shape_30_output_0", "outputIndexes": [ 261 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 261 ], "main_type": "NONE", "name": "Shape1039", "outputIndexes": [ 262 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 261 ], "main_type": "NONE", "name": "Rank1041", "outputIndexes": [ 263 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 263, 263 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1043", "outputIndexes": [ 264 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 264 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1044", "outputIndexes": [ 265 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 264, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1046", "outputIndexes": [ 266 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 266 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1047", "outputIndexes": [ 267 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 262, 265, 267, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1050", "outputIndexes": [ 268 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 268 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1051", "outputIndexes": [ 269 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 269 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1052", "outputIndexes": [ 270 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 270, 269 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1053", "outputIndexes": [ 271 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 261, 271, 264 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_35_output_0", "outputIndexes": [ 272 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 272, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_6_output_0", "outputIndexes": [ 273 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 273, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_71_output_0", "outputIndexes": [ 274 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 259, 274, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_9_output_0", "outputIndexes": [ 275 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 275 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_4_output_0", "outputIndexes": [ 276 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 273, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_70_output_0", "outputIndexes": [ 277 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 259, 20, 277, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_8_output_0", "outputIndexes": [ 278 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 276, 278 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_29_output_0", "outputIndexes": [ 279 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 279, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_17_output_0", "outputIndexes": [ 280 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 260, 280 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_10_output_0", "outputIndexes": [ 281 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 234 ], "outputIndexes": [ 6063 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6063 ], "outputIndexes": [ 6064 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/k_proj/Linear", "inputIndexes": [ 6064 ], "outputIndexes": [ 6065 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 572339530, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6065 ], "outputIndexes": [ 6066 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6066 ], "outputIndexes": [ 282 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 246, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_66_output_0", "outputIndexes": [ 283 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 254, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_67_output_0", "outputIndexes": [ 284 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 283, 284, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_27_output_0", "outputIndexes": [ 285 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 282, 285 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_20_output_0", "outputIndexes": [ 286 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 286 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_2/Cast_output_0", "outputIndexes": [ 287 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_2/Mul_1_output_0", "inputIndexes": [ 287 ], "outputIndexes": [ 288 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 575288672, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 288, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_18_output_0", "outputIndexes": [ 289 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 288 ], "main_type": "NONE", "name": "/Shape_31_output_0", "outputIndexes": [ 290 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 290 ], "main_type": "NONE", "name": "Shape936", "outputIndexes": [ 291 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 290 ], "main_type": "NONE", "name": "Rank938", "outputIndexes": [ 292 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 292, 292 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp940", "outputIndexes": [ 293 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 293 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze941", "outputIndexes": [ 294 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 293, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp943", "outputIndexes": [ 295 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 295 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze944", "outputIndexes": [ 296 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 291, 294, 296, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice947", "outputIndexes": [ 297 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 297 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze948", "outputIndexes": [ 298 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 298 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp949", "outputIndexes": [ 299 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 299, 298 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp950", "outputIndexes": [ 300 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 290, 300, 293 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_36_output_0", "outputIndexes": [ 301 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 301, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_7_output_0", "outputIndexes": [ 302 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 302, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_73_output_0", "outputIndexes": [ 303 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 288, 303, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_11_output_0", "outputIndexes": [ 304 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 304 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_5_output_0", "outputIndexes": [ 305 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 302, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_72_output_0", "outputIndexes": [ 306 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 288, 20, 306, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_10_output_0", "outputIndexes": [ 307 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 305, 307 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_30_output_0", "outputIndexes": [ 308 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 308, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_19_output_0", "outputIndexes": [ 309 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 289, 309 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_11_output_0", "outputIndexes": [ 310 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 234 ], "outputIndexes": [ 6067 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6067 ], "outputIndexes": [ 6068 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/v_proj/Linear", "inputIndexes": [ 6068 ], "outputIndexes": [ 6069 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 575289696, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6069 ], "outputIndexes": [ 6070 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6070 ], "outputIndexes": [ 311 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 246, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_68_output_0", "outputIndexes": [ 312 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 254, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_69_output_0", "outputIndexes": [ 313 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 312, 313, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_28_output_0", "outputIndexes": [ 314 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 311, 314 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_21_output_0", "outputIndexes": [ 315 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 281, 310, 315, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_26_output_0", "outputIndexes": [ 316 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 316 ], "outputIndexes": [ 6071 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6071 ], "outputIndexes": [ 6072 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/o_proj/Linear", "inputIndexes": [ 6072 ], "outputIndexes": [ 6073 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 578238838, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6073 ], "outputIndexes": [ 6074 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6074 ], "outputIndexes": [ 317 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 233, 317 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_13_output_0", "outputIndexes": [ 318 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_2/Mul_1_output_0", "inputIndexes": [ 318 ], "outputIndexes": [ 319 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 601831820, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 319 ], "outputIndexes": [ 6075 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6075 ], "outputIndexes": [ 6076 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/gate_proj/Linear", "inputIndexes": [ 6076 ], "outputIndexes": [ 6077 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 601872780, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6077 ], "outputIndexes": [ 6078 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6078 ], "outputIndexes": [ 320 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 320 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_2/Mul_output_0", "outputIndexes": [ 321 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 319 ], "outputIndexes": [ 6079 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6079 ], "outputIndexes": [ 6080 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/up_proj/Linear", "inputIndexes": [ 6080 ], "outputIndexes": [ 6081 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 675600802, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6081 ], "outputIndexes": [ 6082 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6082 ], "outputIndexes": [ 322 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 321, 322 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_2/Mul_output_0", "outputIndexes": [ 323 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 323 ], "outputIndexes": [ 6083 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6083 ], "outputIndexes": [ 6084 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/down_proj/Linear", "inputIndexes": [ 6084 ], "outputIndexes": [ 6085 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 749328824, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6085 ], "outputIndexes": [ 6086 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.2/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6086 ], "outputIndexes": [ 324 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 318, 324 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_14_output_0", "outputIndexes": [ 325 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 325, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_27_output_0", "outputIndexes": [ 326 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_3/Mul_1_output_0", "inputIndexes": [ 326 ], "outputIndexes": [ 327 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 823056846, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 327 ], "outputIndexes": [ 6087 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6087 ], "outputIndexes": [ 6088 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/q_proj/Linear", "inputIndexes": [ 6088 ], "outputIndexes": [ 6089 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 823097806, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6089 ], "outputIndexes": [ 6090 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6090 ], "outputIndexes": [ 328 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 327 ], "main_type": "NONE", "name": "/Shape_42_output_0", "outputIndexes": [ 329 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 329 ], "main_type": "NONE", "name": "Shape1278", "outputIndexes": [ 330 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 329 ], "main_type": "NONE", "name": "Rank1280", "outputIndexes": [ 331 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 331, 331 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1282", "outputIndexes": [ 332 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 332 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1283", "outputIndexes": [ 333 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 332, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1285", "outputIndexes": [ 334 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 334 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1286", "outputIndexes": [ 335 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 330, 333, 335, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1289", "outputIndexes": [ 336 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 336 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1290", "outputIndexes": [ 337 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 337, 337 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1292", "outputIndexes": [ 338 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 329, 338, 332 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_48_output_0", "outputIndexes": [ 339 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 339, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_96_output_0", "outputIndexes": [ 340 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 332 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1300", "outputIndexes": [ 341 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 334 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1303", "outputIndexes": [ 342 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 330, 341, 342, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1306", "outputIndexes": [ 343 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 343 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1307", "outputIndexes": [ 344 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 344 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1308", "outputIndexes": [ 345 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 345, 344 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1309", "outputIndexes": [ 346 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 329, 346, 332 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_49_output_0", "outputIndexes": [ 347 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 347, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_97_output_0", "outputIndexes": [ 348 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 340, 348, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_39_output_0", "outputIndexes": [ 349 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 328, 349 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_28_output_0", "outputIndexes": [ 350 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 350 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_3/Cast_output_0", "outputIndexes": [ 351 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_3/Mul_1_output_0", "inputIndexes": [ 351 ], "outputIndexes": [ 352 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 846690788, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 352, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_24_output_0", "outputIndexes": [ 353 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 352 ], "main_type": "NONE", "name": "/Shape_44_output_0", "outputIndexes": [ 354 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 354 ], "main_type": "NONE", "name": "Shape1428", "outputIndexes": [ 355 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 354 ], "main_type": "NONE", "name": "Rank1430", "outputIndexes": [ 356 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 356, 356 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1432", "outputIndexes": [ 357 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 357 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1433", "outputIndexes": [ 358 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 357, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1435", "outputIndexes": [ 359 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 359 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1436", "outputIndexes": [ 360 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 355, 358, 360, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1439", "outputIndexes": [ 361 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 361 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1440", "outputIndexes": [ 362 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 362 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1441", "outputIndexes": [ 363 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 363, 362 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1442", "outputIndexes": [ 364 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 354, 364, 357 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_50_output_0", "outputIndexes": [ 365 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 365, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_9_output_0", "outputIndexes": [ 366 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 366, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_103_output_0", "outputIndexes": [ 367 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 352, 367, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_13_output_0", "outputIndexes": [ 368 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 368 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_6_output_0", "outputIndexes": [ 369 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 366, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_102_output_0", "outputIndexes": [ 370 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 352, 20, 370, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_12_output_0", "outputIndexes": [ 371 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 369, 371 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_42_output_0", "outputIndexes": [ 372 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 372, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_25_output_0", "outputIndexes": [ 373 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 353, 373 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_15_output_0", "outputIndexes": [ 374 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 327 ], "outputIndexes": [ 6091 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6091 ], "outputIndexes": [ 6092 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/k_proj/Linear", "inputIndexes": [ 6092 ], "outputIndexes": [ 6093 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 846691812, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6093 ], "outputIndexes": [ 6094 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6094 ], "outputIndexes": [ 375 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 339, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_98_output_0", "outputIndexes": [ 376 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 347, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_99_output_0", "outputIndexes": [ 377 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 376, 377, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_40_output_0", "outputIndexes": [ 378 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 375, 378 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_29_output_0", "outputIndexes": [ 379 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 379 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_3/Cast_output_0", "outputIndexes": [ 380 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_3/Mul_1_output_0", "inputIndexes": [ 380 ], "outputIndexes": [ 381 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 849640954, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 381, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_26_output_0", "outputIndexes": [ 382 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 381 ], "main_type": "NONE", "name": "/Shape_45_output_0", "outputIndexes": [ 383 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 383 ], "main_type": "NONE", "name": "Shape1324", "outputIndexes": [ 384 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 383 ], "main_type": "NONE", "name": "Rank1326", "outputIndexes": [ 385 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 385, 385 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1328", "outputIndexes": [ 386 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 386 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1329", "outputIndexes": [ 387 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 386, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1331", "outputIndexes": [ 388 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 388 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1332", "outputIndexes": [ 389 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 384, 387, 389, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1335", "outputIndexes": [ 390 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 390 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1336", "outputIndexes": [ 391 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 391 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1337", "outputIndexes": [ 392 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 392, 391 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1338", "outputIndexes": [ 393 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 383, 393, 386 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_51_output_0", "outputIndexes": [ 394 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 394, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_10_output_0", "outputIndexes": [ 395 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 395, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_105_output_0", "outputIndexes": [ 396 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 381, 396, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_15_output_0", "outputIndexes": [ 397 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 397 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_7_output_0", "outputIndexes": [ 398 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 395, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_104_output_0", "outputIndexes": [ 399 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 381, 20, 399, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_14_output_0", "outputIndexes": [ 400 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 398, 400 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_43_output_0", "outputIndexes": [ 401 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 401, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_27_output_0", "outputIndexes": [ 402 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 382, 402 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_16_output_0", "outputIndexes": [ 403 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 327 ], "outputIndexes": [ 6095 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6095 ], "outputIndexes": [ 6096 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/v_proj/Linear", "inputIndexes": [ 6096 ], "outputIndexes": [ 6097 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 849641978, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6097 ], "outputIndexes": [ 6098 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6098 ], "outputIndexes": [ 404 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 339, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_100_output_0", "outputIndexes": [ 405 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 347, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_101_output_0", "outputIndexes": [ 406 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 405, 406, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_41_output_0", "outputIndexes": [ 407 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 404, 407 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_30_output_0", "outputIndexes": [ 408 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 374, 403, 408, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_35_output_0", "outputIndexes": [ 409 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 409 ], "outputIndexes": [ 6099 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6099 ], "outputIndexes": [ 6100 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/o_proj/Linear", "inputIndexes": [ 6100 ], "outputIndexes": [ 6101 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 852591120, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6101 ], "outputIndexes": [ 6102 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6102 ], "outputIndexes": [ 410 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 326, 410 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_18_output_0", "outputIndexes": [ 411 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_3/Mul_1_output_0", "inputIndexes": [ 411 ], "outputIndexes": [ 412 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 876184102, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 412 ], "outputIndexes": [ 6103 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6103 ], "outputIndexes": [ 6104 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/gate_proj/Linear", "inputIndexes": [ 6104 ], "outputIndexes": [ 6105 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 876225062, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6105 ], "outputIndexes": [ 6106 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6106 ], "outputIndexes": [ 413 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 413 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_3/Mul_output_0", "outputIndexes": [ 414 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 412 ], "outputIndexes": [ 6107 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6107 ], "outputIndexes": [ 6108 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/up_proj/Linear", "inputIndexes": [ 6108 ], "outputIndexes": [ 6109 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 949953084, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6109 ], "outputIndexes": [ 6110 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6110 ], "outputIndexes": [ 415 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 414, 415 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_3/Mul_output_0", "outputIndexes": [ 416 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 416 ], "outputIndexes": [ 6111 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6111 ], "outputIndexes": [ 6112 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/down_proj/Linear", "inputIndexes": [ 6112 ], "outputIndexes": [ 6113 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 1023681106, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6113 ], "outputIndexes": [ 6114 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.3/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6114 ], "outputIndexes": [ 417 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 411, 417 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_19_output_0", "outputIndexes": [ 418 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 418, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_36_output_0", "outputIndexes": [ 419 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_4/Mul_1_output_0", "inputIndexes": [ 419 ], "outputIndexes": [ 420 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 1097409128, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 420 ], "outputIndexes": [ 6115 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6115 ], "outputIndexes": [ 6116 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/q_proj/Linear", "inputIndexes": [ 6116 ], "outputIndexes": [ 6117 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 1097450088, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6117 ], "outputIndexes": [ 6118 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6118 ], "outputIndexes": [ 421 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 420 ], "main_type": "NONE", "name": "/Shape_56_output_0", "outputIndexes": [ 422 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 422 ], "main_type": "NONE", "name": "Shape1667", "outputIndexes": [ 423 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 422 ], "main_type": "NONE", "name": "Rank1669", "outputIndexes": [ 424 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 424, 424 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1671", "outputIndexes": [ 425 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 425 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1672", "outputIndexes": [ 426 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 425, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1674", "outputIndexes": [ 427 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 427 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1675", "outputIndexes": [ 428 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 423, 426, 428, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1678", "outputIndexes": [ 429 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 429 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1679", "outputIndexes": [ 430 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 430, 430 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1681", "outputIndexes": [ 431 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 422, 431, 425 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_63_output_0", "outputIndexes": [ 432 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 432, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_128_output_0", "outputIndexes": [ 433 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 425 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1689", "outputIndexes": [ 434 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 427 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1692", "outputIndexes": [ 435 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 423, 434, 435, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1695", "outputIndexes": [ 436 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 436 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1696", "outputIndexes": [ 437 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 437 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1697", "outputIndexes": [ 438 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 438, 437 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1698", "outputIndexes": [ 439 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 422, 439, 425 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_64_output_0", "outputIndexes": [ 440 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 440, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_129_output_0", "outputIndexes": [ 441 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 433, 441, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_52_output_0", "outputIndexes": [ 442 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 421, 442 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_37_output_0", "outputIndexes": [ 443 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 443 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_4/Cast_output_0", "outputIndexes": [ 444 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_4/Mul_1_output_0", "inputIndexes": [ 444 ], "outputIndexes": [ 445 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 1121043070, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 445, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_32_output_0", "outputIndexes": [ 446 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 445 ], "main_type": "NONE", "name": "/Shape_58_output_0", "outputIndexes": [ 447 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 447 ], "main_type": "NONE", "name": "Shape1817", "outputIndexes": [ 448 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 447 ], "main_type": "NONE", "name": "Rank1819", "outputIndexes": [ 449 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 449, 449 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1821", "outputIndexes": [ 450 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 450 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1822", "outputIndexes": [ 451 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 450, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1824", "outputIndexes": [ 452 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 452 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1825", "outputIndexes": [ 453 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 448, 451, 453, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1828", "outputIndexes": [ 454 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 454 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1829", "outputIndexes": [ 455 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 455 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1830", "outputIndexes": [ 456 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 456, 455 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1831", "outputIndexes": [ 457 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 447, 457, 450 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_65_output_0", "outputIndexes": [ 458 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 458, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_12_output_0", "outputIndexes": [ 459 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 459, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_135_output_0", "outputIndexes": [ 460 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 445, 460, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_17_output_0", "outputIndexes": [ 461 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 461 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_8_output_0", "outputIndexes": [ 462 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 459, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_134_output_0", "outputIndexes": [ 463 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 445, 20, 463, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_16_output_0", "outputIndexes": [ 464 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 462, 464 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_55_output_0", "outputIndexes": [ 465 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 465, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_33_output_0", "outputIndexes": [ 466 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 446, 466 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_20_output_0", "outputIndexes": [ 467 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 420 ], "outputIndexes": [ 6119 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6119 ], "outputIndexes": [ 6120 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/k_proj/Linear", "inputIndexes": [ 6120 ], "outputIndexes": [ 6121 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 1121044094, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6121 ], "outputIndexes": [ 6122 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6122 ], "outputIndexes": [ 468 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 432, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_130_output_0", "outputIndexes": [ 469 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 440, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_131_output_0", "outputIndexes": [ 470 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 469, 470, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_53_output_0", "outputIndexes": [ 471 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 468, 471 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_38_output_0", "outputIndexes": [ 472 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 472 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_4/Cast_output_0", "outputIndexes": [ 473 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_4/Mul_1_output_0", "inputIndexes": [ 473 ], "outputIndexes": [ 474 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 1123993236, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 474, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_34_output_0", "outputIndexes": [ 475 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 474 ], "main_type": "NONE", "name": "/Shape_59_output_0", "outputIndexes": [ 476 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 476 ], "main_type": "NONE", "name": "Shape1713", "outputIndexes": [ 477 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 476 ], "main_type": "NONE", "name": "Rank1715", "outputIndexes": [ 478 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 478, 478 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1717", "outputIndexes": [ 479 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 479 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1718", "outputIndexes": [ 480 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 479, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1720", "outputIndexes": [ 481 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 481 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze1721", "outputIndexes": [ 482 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 477, 480, 482, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice1724", "outputIndexes": [ 483 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 483 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze1725", "outputIndexes": [ 484 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 484 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1726", "outputIndexes": [ 485 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 485, 484 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp1727", "outputIndexes": [ 486 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 476, 486, 479 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_66_output_0", "outputIndexes": [ 487 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 487, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_13_output_0", "outputIndexes": [ 488 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 488, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_137_output_0", "outputIndexes": [ 489 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 474, 489, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_19_output_0", "outputIndexes": [ 490 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 490 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_9_output_0", "outputIndexes": [ 491 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 488, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_136_output_0", "outputIndexes": [ 492 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 474, 20, 492, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_18_output_0", "outputIndexes": [ 493 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 491, 493 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_56_output_0", "outputIndexes": [ 494 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 494, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_35_output_0", "outputIndexes": [ 495 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 475, 495 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_21_output_0", "outputIndexes": [ 496 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 420 ], "outputIndexes": [ 6123 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6123 ], "outputIndexes": [ 6124 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/v_proj/Linear", "inputIndexes": [ 6124 ], "outputIndexes": [ 6125 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 1123994260, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6125 ], "outputIndexes": [ 6126 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6126 ], "outputIndexes": [ 497 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 432, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_132_output_0", "outputIndexes": [ 498 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 440, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_133_output_0", "outputIndexes": [ 499 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 498, 499, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_54_output_0", "outputIndexes": [ 500 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 497, 500 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_39_output_0", "outputIndexes": [ 501 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 467, 496, 501, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_44_output_0", "outputIndexes": [ 502 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 502 ], "outputIndexes": [ 6127 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6127 ], "outputIndexes": [ 6128 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/o_proj/Linear", "inputIndexes": [ 6128 ], "outputIndexes": [ 6129 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 1126943402, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6129 ], "outputIndexes": [ 6130 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6130 ], "outputIndexes": [ 503 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 419, 503 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_23_output_0", "outputIndexes": [ 504 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_4/Mul_1_output_0", "inputIndexes": [ 504 ], "outputIndexes": [ 505 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 1150536384, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 505 ], "outputIndexes": [ 6131 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6131 ], "outputIndexes": [ 6132 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/gate_proj/Linear", "inputIndexes": [ 6132 ], "outputIndexes": [ 6133 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 1150577344, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6133 ], "outputIndexes": [ 6134 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6134 ], "outputIndexes": [ 506 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 506 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_4/Mul_output_0", "outputIndexes": [ 507 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 505 ], "outputIndexes": [ 6135 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6135 ], "outputIndexes": [ 6136 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/up_proj/Linear", "inputIndexes": [ 6136 ], "outputIndexes": [ 6137 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 1224305366, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6137 ], "outputIndexes": [ 6138 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6138 ], "outputIndexes": [ 508 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 507, 508 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_4/Mul_output_0", "outputIndexes": [ 509 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 509 ], "outputIndexes": [ 6139 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6139 ], "outputIndexes": [ 6140 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/down_proj/Linear", "inputIndexes": [ 6140 ], "outputIndexes": [ 6141 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 1298033388, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6141 ], "outputIndexes": [ 6142 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.4/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6142 ], "outputIndexes": [ 510 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 504, 510 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_24_output_0", "outputIndexes": [ 511 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 511, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_45_output_0", "outputIndexes": [ 512 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_5/Mul_1_output_0", "inputIndexes": [ 512 ], "outputIndexes": [ 513 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 1371761410, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 513 ], "outputIndexes": [ 6143 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6143 ], "outputIndexes": [ 6144 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/q_proj/Linear", "inputIndexes": [ 6144 ], "outputIndexes": [ 6145 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 1371802370, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6145 ], "outputIndexes": [ 6146 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6146 ], "outputIndexes": [ 514 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 513 ], "main_type": "NONE", "name": "/Shape_70_output_0", "outputIndexes": [ 515 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 515 ], "main_type": "NONE", "name": "Shape2056", "outputIndexes": [ 516 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 515 ], "main_type": "NONE", "name": "Rank2058", "outputIndexes": [ 517 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 517, 517 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2060", "outputIndexes": [ 518 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 518 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2061", "outputIndexes": [ 519 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 518, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2063", "outputIndexes": [ 520 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 520 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2064", "outputIndexes": [ 521 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 516, 519, 521, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2067", "outputIndexes": [ 522 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 522 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2068", "outputIndexes": [ 523 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 523, 523 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2070", "outputIndexes": [ 524 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 515, 524, 518 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_78_output_0", "outputIndexes": [ 525 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 525, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_160_output_0", "outputIndexes": [ 526 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 518 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2078", "outputIndexes": [ 527 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 520 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2081", "outputIndexes": [ 528 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 516, 527, 528, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2084", "outputIndexes": [ 529 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 529 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2085", "outputIndexes": [ 530 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 530 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2086", "outputIndexes": [ 531 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 531, 530 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2087", "outputIndexes": [ 532 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 515, 532, 518 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_79_output_0", "outputIndexes": [ 533 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 533, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_161_output_0", "outputIndexes": [ 534 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 526, 534, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_65_output_0", "outputIndexes": [ 535 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 514, 535 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_46_output_0", "outputIndexes": [ 536 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 536 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_5/Cast_output_0", "outputIndexes": [ 537 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_5/Mul_1_output_0", "inputIndexes": [ 537 ], "outputIndexes": [ 538 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 1395395352, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 538, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_40_output_0", "outputIndexes": [ 539 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 538 ], "main_type": "NONE", "name": "/Shape_72_output_0", "outputIndexes": [ 540 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 540 ], "main_type": "NONE", "name": "Shape2206", "outputIndexes": [ 541 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 540 ], "main_type": "NONE", "name": "Rank2208", "outputIndexes": [ 542 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 542, 542 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2210", "outputIndexes": [ 543 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 543 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2211", "outputIndexes": [ 544 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 543, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2213", "outputIndexes": [ 545 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 545 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2214", "outputIndexes": [ 546 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 541, 544, 546, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2217", "outputIndexes": [ 547 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 547 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2218", "outputIndexes": [ 548 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 548 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2219", "outputIndexes": [ 549 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 549, 548 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2220", "outputIndexes": [ 550 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 540, 550, 543 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_80_output_0", "outputIndexes": [ 551 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 551, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_15_output_0", "outputIndexes": [ 552 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 552, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_167_output_0", "outputIndexes": [ 553 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 538, 553, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_21_output_0", "outputIndexes": [ 554 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 554 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_10_output_0", "outputIndexes": [ 555 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 552, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_166_output_0", "outputIndexes": [ 556 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 538, 20, 556, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_20_output_0", "outputIndexes": [ 557 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 555, 557 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_68_output_0", "outputIndexes": [ 558 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 558, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_41_output_0", "outputIndexes": [ 559 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 539, 559 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_25_output_0", "outputIndexes": [ 560 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 513 ], "outputIndexes": [ 6147 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6147 ], "outputIndexes": [ 6148 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/k_proj/Linear", "inputIndexes": [ 6148 ], "outputIndexes": [ 6149 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 1395396376, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6149 ], "outputIndexes": [ 6150 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6150 ], "outputIndexes": [ 561 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 525, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_162_output_0", "outputIndexes": [ 562 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 533, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_163_output_0", "outputIndexes": [ 563 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 562, 563, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_66_output_0", "outputIndexes": [ 564 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 561, 564 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_47_output_0", "outputIndexes": [ 565 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 565 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_5/Cast_output_0", "outputIndexes": [ 566 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_5/Mul_1_output_0", "inputIndexes": [ 566 ], "outputIndexes": [ 567 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 1398345518, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 567, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_42_output_0", "outputIndexes": [ 568 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 567 ], "main_type": "NONE", "name": "/Shape_73_output_0", "outputIndexes": [ 569 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 569 ], "main_type": "NONE", "name": "Shape2102", "outputIndexes": [ 570 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 569 ], "main_type": "NONE", "name": "Rank2104", "outputIndexes": [ 571 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 571, 571 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2106", "outputIndexes": [ 572 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 572 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2107", "outputIndexes": [ 573 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 572, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2109", "outputIndexes": [ 574 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 574 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2110", "outputIndexes": [ 575 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 570, 573, 575, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2113", "outputIndexes": [ 576 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 576 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2114", "outputIndexes": [ 577 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 577 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2115", "outputIndexes": [ 578 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 578, 577 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2116", "outputIndexes": [ 579 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 569, 579, 572 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_81_output_0", "outputIndexes": [ 580 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 580, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_16_output_0", "outputIndexes": [ 581 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 581, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_169_output_0", "outputIndexes": [ 582 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 567, 582, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_23_output_0", "outputIndexes": [ 583 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 583 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_11_output_0", "outputIndexes": [ 584 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 581, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_168_output_0", "outputIndexes": [ 585 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 567, 20, 585, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_22_output_0", "outputIndexes": [ 586 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 584, 586 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_69_output_0", "outputIndexes": [ 587 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 587, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_43_output_0", "outputIndexes": [ 588 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 568, 588 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_26_output_0", "outputIndexes": [ 589 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 513 ], "outputIndexes": [ 6151 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6151 ], "outputIndexes": [ 6152 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/v_proj/Linear", "inputIndexes": [ 6152 ], "outputIndexes": [ 6153 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 1398346542, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6153 ], "outputIndexes": [ 6154 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6154 ], "outputIndexes": [ 590 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 525, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_164_output_0", "outputIndexes": [ 591 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 533, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_165_output_0", "outputIndexes": [ 592 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 591, 592, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_67_output_0", "outputIndexes": [ 593 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 590, 593 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_48_output_0", "outputIndexes": [ 594 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 560, 589, 594, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_53_output_0", "outputIndexes": [ 595 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 595 ], "outputIndexes": [ 6155 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6155 ], "outputIndexes": [ 6156 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/o_proj/Linear", "inputIndexes": [ 6156 ], "outputIndexes": [ 6157 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 1401295684, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6157 ], "outputIndexes": [ 6158 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6158 ], "outputIndexes": [ 596 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 512, 596 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_28_output_0", "outputIndexes": [ 597 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_5/Mul_1_output_0", "inputIndexes": [ 597 ], "outputIndexes": [ 598 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 1424888666, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 598 ], "outputIndexes": [ 6159 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6159 ], "outputIndexes": [ 6160 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/gate_proj/Linear", "inputIndexes": [ 6160 ], "outputIndexes": [ 6161 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 1424929626, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6161 ], "outputIndexes": [ 6162 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6162 ], "outputIndexes": [ 599 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 599 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_5/Mul_output_0", "outputIndexes": [ 600 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 598 ], "outputIndexes": [ 6163 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6163 ], "outputIndexes": [ 6164 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/up_proj/Linear", "inputIndexes": [ 6164 ], "outputIndexes": [ 6165 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 1498657648, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6165 ], "outputIndexes": [ 6166 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6166 ], "outputIndexes": [ 601 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 600, 601 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_5/Mul_output_0", "outputIndexes": [ 602 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 602 ], "outputIndexes": [ 6167 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6167 ], "outputIndexes": [ 6168 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/down_proj/Linear", "inputIndexes": [ 6168 ], "outputIndexes": [ 6169 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 1572385670, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6169 ], "outputIndexes": [ 6170 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.5/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6170 ], "outputIndexes": [ 603 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 597, 603 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_29_output_0", "outputIndexes": [ 604 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 604, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_54_output_0", "outputIndexes": [ 605 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_6/Mul_1_output_0", "inputIndexes": [ 605 ], "outputIndexes": [ 606 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 1646113692, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 606 ], "outputIndexes": [ 6171 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6171 ], "outputIndexes": [ 6172 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/q_proj/Linear", "inputIndexes": [ 6172 ], "outputIndexes": [ 6173 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 1646154652, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6173 ], "outputIndexes": [ 6174 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6174 ], "outputIndexes": [ 607 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 606 ], "main_type": "NONE", "name": "/Shape_84_output_0", "outputIndexes": [ 608 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 608 ], "main_type": "NONE", "name": "Shape2445", "outputIndexes": [ 609 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 608 ], "main_type": "NONE", "name": "Rank2447", "outputIndexes": [ 610 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 610, 610 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2449", "outputIndexes": [ 611 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 611 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2450", "outputIndexes": [ 612 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 611, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2452", "outputIndexes": [ 613 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 613 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2453", "outputIndexes": [ 614 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 609, 612, 614, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2456", "outputIndexes": [ 615 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 615 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2457", "outputIndexes": [ 616 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 616, 616 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2459", "outputIndexes": [ 617 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 608, 617, 611 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_93_output_0", "outputIndexes": [ 618 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 618, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_192_output_0", "outputIndexes": [ 619 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 611 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2467", "outputIndexes": [ 620 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 613 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2470", "outputIndexes": [ 621 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 609, 620, 621, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2473", "outputIndexes": [ 622 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 622 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2474", "outputIndexes": [ 623 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 623 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2475", "outputIndexes": [ 624 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 624, 623 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2476", "outputIndexes": [ 625 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 608, 625, 611 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_94_output_0", "outputIndexes": [ 626 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 626, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_193_output_0", "outputIndexes": [ 627 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 619, 627, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_78_output_0", "outputIndexes": [ 628 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 607, 628 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_55_output_0", "outputIndexes": [ 629 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 629 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_6/Cast_output_0", "outputIndexes": [ 630 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_6/Mul_1_output_0", "inputIndexes": [ 630 ], "outputIndexes": [ 631 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 1669747634, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 631, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_48_output_0", "outputIndexes": [ 632 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 631 ], "main_type": "NONE", "name": "/Shape_86_output_0", "outputIndexes": [ 633 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 633 ], "main_type": "NONE", "name": "Shape2595", "outputIndexes": [ 634 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 633 ], "main_type": "NONE", "name": "Rank2597", "outputIndexes": [ 635 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 635, 635 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2599", "outputIndexes": [ 636 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 636 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2600", "outputIndexes": [ 637 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 636, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2602", "outputIndexes": [ 638 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 638 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2603", "outputIndexes": [ 639 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 634, 637, 639, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2606", "outputIndexes": [ 640 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 640 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2607", "outputIndexes": [ 641 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 641 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2608", "outputIndexes": [ 642 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 642, 641 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2609", "outputIndexes": [ 643 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 633, 643, 636 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_95_output_0", "outputIndexes": [ 644 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 644, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_18_output_0", "outputIndexes": [ 645 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 645, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_199_output_0", "outputIndexes": [ 646 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 631, 646, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_25_output_0", "outputIndexes": [ 647 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 647 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_12_output_0", "outputIndexes": [ 648 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 645, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_198_output_0", "outputIndexes": [ 649 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 631, 20, 649, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_24_output_0", "outputIndexes": [ 650 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 648, 650 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_81_output_0", "outputIndexes": [ 651 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 651, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_49_output_0", "outputIndexes": [ 652 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 632, 652 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_30_output_0", "outputIndexes": [ 653 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 606 ], "outputIndexes": [ 6175 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6175 ], "outputIndexes": [ 6176 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/k_proj/Linear", "inputIndexes": [ 6176 ], "outputIndexes": [ 6177 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 1669748658, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6177 ], "outputIndexes": [ 6178 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6178 ], "outputIndexes": [ 654 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 618, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_194_output_0", "outputIndexes": [ 655 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 626, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_195_output_0", "outputIndexes": [ 656 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 655, 656, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_79_output_0", "outputIndexes": [ 657 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 654, 657 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_56_output_0", "outputIndexes": [ 658 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 658 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_6/Cast_output_0", "outputIndexes": [ 659 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_6/Mul_1_output_0", "inputIndexes": [ 659 ], "outputIndexes": [ 660 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 1672697800, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 660, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_50_output_0", "outputIndexes": [ 661 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 660 ], "main_type": "NONE", "name": "/Shape_87_output_0", "outputIndexes": [ 662 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 662 ], "main_type": "NONE", "name": "Shape2491", "outputIndexes": [ 663 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 662 ], "main_type": "NONE", "name": "Rank2493", "outputIndexes": [ 664 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 664, 664 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2495", "outputIndexes": [ 665 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 665 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2496", "outputIndexes": [ 666 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 665, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2498", "outputIndexes": [ 667 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 667 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2499", "outputIndexes": [ 668 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 663, 666, 668, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2502", "outputIndexes": [ 669 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 669 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2503", "outputIndexes": [ 670 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 670 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2504", "outputIndexes": [ 671 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 671, 670 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2505", "outputIndexes": [ 672 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 662, 672, 665 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_96_output_0", "outputIndexes": [ 673 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 673, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_19_output_0", "outputIndexes": [ 674 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 674, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_201_output_0", "outputIndexes": [ 675 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 660, 675, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_27_output_0", "outputIndexes": [ 676 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 676 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_13_output_0", "outputIndexes": [ 677 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 674, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_200_output_0", "outputIndexes": [ 678 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 660, 20, 678, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_26_output_0", "outputIndexes": [ 679 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 677, 679 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_82_output_0", "outputIndexes": [ 680 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 680, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_51_output_0", "outputIndexes": [ 681 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 661, 681 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_31_output_0", "outputIndexes": [ 682 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 606 ], "outputIndexes": [ 6179 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6179 ], "outputIndexes": [ 6180 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/v_proj/Linear", "inputIndexes": [ 6180 ], "outputIndexes": [ 6181 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 1672698824, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6181 ], "outputIndexes": [ 6182 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6182 ], "outputIndexes": [ 683 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 618, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_196_output_0", "outputIndexes": [ 684 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 626, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_197_output_0", "outputIndexes": [ 685 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 684, 685, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_80_output_0", "outputIndexes": [ 686 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 683, 686 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_57_output_0", "outputIndexes": [ 687 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 653, 682, 687, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_62_output_0", "outputIndexes": [ 688 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 688 ], "outputIndexes": [ 6183 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6183 ], "outputIndexes": [ 6184 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/o_proj/Linear", "inputIndexes": [ 6184 ], "outputIndexes": [ 6185 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 1675647966, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6185 ], "outputIndexes": [ 6186 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6186 ], "outputIndexes": [ 689 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 605, 689 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_33_output_0", "outputIndexes": [ 690 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_6/Mul_1_output_0", "inputIndexes": [ 690 ], "outputIndexes": [ 691 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 1699240948, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 691 ], "outputIndexes": [ 6187 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6187 ], "outputIndexes": [ 6188 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/gate_proj/Linear", "inputIndexes": [ 6188 ], "outputIndexes": [ 6189 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 1699281908, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6189 ], "outputIndexes": [ 6190 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6190 ], "outputIndexes": [ 692 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 692 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_6/Mul_output_0", "outputIndexes": [ 693 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 691 ], "outputIndexes": [ 6191 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6191 ], "outputIndexes": [ 6192 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/up_proj/Linear", "inputIndexes": [ 6192 ], "outputIndexes": [ 6193 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 1773009930, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6193 ], "outputIndexes": [ 6194 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6194 ], "outputIndexes": [ 694 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 693, 694 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_6/Mul_output_0", "outputIndexes": [ 695 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 695 ], "outputIndexes": [ 6195 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6195 ], "outputIndexes": [ 6196 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/down_proj/Linear", "inputIndexes": [ 6196 ], "outputIndexes": [ 6197 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 1846737952, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6197 ], "outputIndexes": [ 6198 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.6/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6198 ], "outputIndexes": [ 696 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 690, 696 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_34_output_0", "outputIndexes": [ 697 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 697, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_63_output_0", "outputIndexes": [ 698 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_7/Mul_1_output_0", "inputIndexes": [ 698 ], "outputIndexes": [ 699 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 1920465974, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 699 ], "outputIndexes": [ 6199 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6199 ], "outputIndexes": [ 6200 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/q_proj/Linear", "inputIndexes": [ 6200 ], "outputIndexes": [ 6201 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 1920506934, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6201 ], "outputIndexes": [ 6202 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6202 ], "outputIndexes": [ 700 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 699 ], "main_type": "NONE", "name": "/Shape_98_output_0", "outputIndexes": [ 701 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 701 ], "main_type": "NONE", "name": "Shape2834", "outputIndexes": [ 702 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 701 ], "main_type": "NONE", "name": "Rank2836", "outputIndexes": [ 703 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 703, 703 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2838", "outputIndexes": [ 704 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 704 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2839", "outputIndexes": [ 705 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 704, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2841", "outputIndexes": [ 706 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 706 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2842", "outputIndexes": [ 707 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 702, 705, 707, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2845", "outputIndexes": [ 708 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 708 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2846", "outputIndexes": [ 709 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 709, 709 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2848", "outputIndexes": [ 710 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 701, 710, 704 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_108_output_0", "outputIndexes": [ 711 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 711, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_224_output_0", "outputIndexes": [ 712 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 704 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2856", "outputIndexes": [ 713 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 706 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2859", "outputIndexes": [ 714 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 702, 713, 714, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2862", "outputIndexes": [ 715 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 715 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2863", "outputIndexes": [ 716 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 716 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2864", "outputIndexes": [ 717 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 717, 716 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2865", "outputIndexes": [ 718 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 701, 718, 704 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_109_output_0", "outputIndexes": [ 719 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 719, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_225_output_0", "outputIndexes": [ 720 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 712, 720, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_91_output_0", "outputIndexes": [ 721 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 700, 721 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_64_output_0", "outputIndexes": [ 722 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 722 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_7/Cast_output_0", "outputIndexes": [ 723 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_7/Mul_1_output_0", "inputIndexes": [ 723 ], "outputIndexes": [ 724 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 1944099916, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 724, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_56_output_0", "outputIndexes": [ 725 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 724 ], "main_type": "NONE", "name": "/Shape_100_output_0", "outputIndexes": [ 726 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 726 ], "main_type": "NONE", "name": "Shape2983", "outputIndexes": [ 727 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 726 ], "main_type": "NONE", "name": "Rank2985", "outputIndexes": [ 728 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 728, 728 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2987", "outputIndexes": [ 729 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 729 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2988", "outputIndexes": [ 730 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 729, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2990", "outputIndexes": [ 731 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 731 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2991", "outputIndexes": [ 732 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 727, 730, 732, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2994", "outputIndexes": [ 733 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 733 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2995", "outputIndexes": [ 734 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 734 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2996", "outputIndexes": [ 735 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 735, 734 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2997", "outputIndexes": [ 736 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 726, 736, 729 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_110_output_0", "outputIndexes": [ 737 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 737, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_21_output_0", "outputIndexes": [ 738 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 738, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_231_output_0", "outputIndexes": [ 739 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 724, 739, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_29_output_0", "outputIndexes": [ 740 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 740 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_14_output_0", "outputIndexes": [ 741 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 738, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_230_output_0", "outputIndexes": [ 742 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 724, 20, 742, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_28_output_0", "outputIndexes": [ 743 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 741, 743 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_94_output_0", "outputIndexes": [ 744 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 744, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_57_output_0", "outputIndexes": [ 745 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 725, 745 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_35_output_0", "outputIndexes": [ 746 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 699 ], "outputIndexes": [ 6203 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6203 ], "outputIndexes": [ 6204 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/k_proj/Linear", "inputIndexes": [ 6204 ], "outputIndexes": [ 6205 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 1944100940, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6205 ], "outputIndexes": [ 6206 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6206 ], "outputIndexes": [ 747 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 711, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_226_output_0", "outputIndexes": [ 748 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 719, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_227_output_0", "outputIndexes": [ 749 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 748, 749, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_92_output_0", "outputIndexes": [ 750 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 747, 750 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_65_output_0", "outputIndexes": [ 751 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 751 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_7/Cast_output_0", "outputIndexes": [ 752 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_7/Mul_1_output_0", "inputIndexes": [ 752 ], "outputIndexes": [ 753 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 1947050082, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 753, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_58_output_0", "outputIndexes": [ 754 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 753 ], "main_type": "NONE", "name": "/Shape_101_output_0", "outputIndexes": [ 755 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 755 ], "main_type": "NONE", "name": "Shape2880", "outputIndexes": [ 756 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 755 ], "main_type": "NONE", "name": "Rank2882", "outputIndexes": [ 757 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 757, 757 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2884", "outputIndexes": [ 758 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 758 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2885", "outputIndexes": [ 759 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 758, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2887", "outputIndexes": [ 760 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 760 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze2888", "outputIndexes": [ 761 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 756, 759, 761, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice2891", "outputIndexes": [ 762 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 762 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze2892", "outputIndexes": [ 763 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 763 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2893", "outputIndexes": [ 764 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 764, 763 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp2894", "outputIndexes": [ 765 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 755, 765, 758 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_111_output_0", "outputIndexes": [ 766 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 766, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_22_output_0", "outputIndexes": [ 767 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 767, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_233_output_0", "outputIndexes": [ 768 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 753, 768, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_31_output_0", "outputIndexes": [ 769 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 769 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_15_output_0", "outputIndexes": [ 770 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 767, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_232_output_0", "outputIndexes": [ 771 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 753, 20, 771, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_30_output_0", "outputIndexes": [ 772 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 770, 772 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_95_output_0", "outputIndexes": [ 773 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 773, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_59_output_0", "outputIndexes": [ 774 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 754, 774 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_36_output_0", "outputIndexes": [ 775 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 699 ], "outputIndexes": [ 6207 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6207 ], "outputIndexes": [ 6208 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/v_proj/Linear", "inputIndexes": [ 6208 ], "outputIndexes": [ 6209 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 1947051106, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6209 ], "outputIndexes": [ 6210 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6210 ], "outputIndexes": [ 776 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 711, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_228_output_0", "outputIndexes": [ 777 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 719, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_229_output_0", "outputIndexes": [ 778 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 777, 778, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_93_output_0", "outputIndexes": [ 779 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 776, 779 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_66_output_0", "outputIndexes": [ 780 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 746, 775, 780, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_71_output_0", "outputIndexes": [ 781 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 781 ], "outputIndexes": [ 6211 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6211 ], "outputIndexes": [ 6212 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/o_proj/Linear", "inputIndexes": [ 6212 ], "outputIndexes": [ 6213 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 1950000248, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6213 ], "outputIndexes": [ 6214 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6214 ], "outputIndexes": [ 782 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 698, 782 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_38_output_0", "outputIndexes": [ 783 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_7/Mul_1_output_0", "inputIndexes": [ 783 ], "outputIndexes": [ 784 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 1973593230, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 784 ], "outputIndexes": [ 6215 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6215 ], "outputIndexes": [ 6216 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/gate_proj/Linear", "inputIndexes": [ 6216 ], "outputIndexes": [ 6217 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 1973634190, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6217 ], "outputIndexes": [ 6218 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6218 ], "outputIndexes": [ 785 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 785 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_7/Mul_output_0", "outputIndexes": [ 786 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 784 ], "outputIndexes": [ 6219 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6219 ], "outputIndexes": [ 6220 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/up_proj/Linear", "inputIndexes": [ 6220 ], "outputIndexes": [ 6221 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 2047362212, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6221 ], "outputIndexes": [ 6222 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6222 ], "outputIndexes": [ 787 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 786, 787 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_7/Mul_output_0", "outputIndexes": [ 788 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 788 ], "outputIndexes": [ 6223 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6223 ], "outputIndexes": [ 6224 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/down_proj/Linear", "inputIndexes": [ 6224 ], "outputIndexes": [ 6225 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 2121090234, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6225 ], "outputIndexes": [ 6226 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.7/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6226 ], "outputIndexes": [ 789 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 783, 789 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_39_output_0", "outputIndexes": [ 790 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 790, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_72_output_0", "outputIndexes": [ 791 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_8/Mul_1_output_0", "inputIndexes": [ 791 ], "outputIndexes": [ 792 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 2194818256, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 792 ], "outputIndexes": [ 6227 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6227 ], "outputIndexes": [ 6228 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/q_proj/Linear", "inputIndexes": [ 6228 ], "outputIndexes": [ 6229 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 2194859216, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6229 ], "outputIndexes": [ 6230 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6230 ], "outputIndexes": [ 793 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 792 ], "main_type": "NONE", "name": "/Shape_112_output_0", "outputIndexes": [ 794 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 794 ], "main_type": "NONE", "name": "Shape3222", "outputIndexes": [ 795 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 794 ], "main_type": "NONE", "name": "Rank3224", "outputIndexes": [ 796 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 796, 796 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3226", "outputIndexes": [ 797 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 797 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3227", "outputIndexes": [ 798 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 797, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3229", "outputIndexes": [ 799 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 799 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3230", "outputIndexes": [ 800 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 795, 798, 800, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3233", "outputIndexes": [ 801 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 801 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3234", "outputIndexes": [ 802 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 802, 802 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3236", "outputIndexes": [ 803 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 794, 803, 797 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_123_output_0", "outputIndexes": [ 804 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 804, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_256_output_0", "outputIndexes": [ 805 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 797 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3244", "outputIndexes": [ 806 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 799 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3247", "outputIndexes": [ 807 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 795, 806, 807, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3250", "outputIndexes": [ 808 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 808 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3251", "outputIndexes": [ 809 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 809 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3252", "outputIndexes": [ 810 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 810, 809 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3253", "outputIndexes": [ 811 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 794, 811, 797 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_124_output_0", "outputIndexes": [ 812 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 812, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_257_output_0", "outputIndexes": [ 813 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 805, 813, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_104_output_0", "outputIndexes": [ 814 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 793, 814 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_73_output_0", "outputIndexes": [ 815 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 815 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_8/Cast_output_0", "outputIndexes": [ 816 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_8/Mul_1_output_0", "inputIndexes": [ 816 ], "outputIndexes": [ 817 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 2218452198, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 817, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_64_output_0", "outputIndexes": [ 818 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 817 ], "main_type": "NONE", "name": "/Shape_114_output_0", "outputIndexes": [ 819 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 819 ], "main_type": "NONE", "name": "Shape3372", "outputIndexes": [ 820 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 819 ], "main_type": "NONE", "name": "Rank3374", "outputIndexes": [ 821 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 821, 821 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3376", "outputIndexes": [ 822 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 822 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3377", "outputIndexes": [ 823 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 822, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3379", "outputIndexes": [ 824 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 824 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3380", "outputIndexes": [ 825 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 820, 823, 825, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3383", "outputIndexes": [ 826 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 826 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3384", "outputIndexes": [ 827 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 827 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3385", "outputIndexes": [ 828 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 828, 827 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3386", "outputIndexes": [ 829 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 819, 829, 822 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_125_output_0", "outputIndexes": [ 830 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 830, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_24_output_0", "outputIndexes": [ 831 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 831, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_263_output_0", "outputIndexes": [ 832 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 817, 832, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_33_output_0", "outputIndexes": [ 833 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 833 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_16_output_0", "outputIndexes": [ 834 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 831, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_262_output_0", "outputIndexes": [ 835 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 817, 20, 835, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_32_output_0", "outputIndexes": [ 836 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 834, 836 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_107_output_0", "outputIndexes": [ 837 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 837, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_65_output_0", "outputIndexes": [ 838 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 818, 838 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_40_output_0", "outputIndexes": [ 839 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 792 ], "outputIndexes": [ 6231 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6231 ], "outputIndexes": [ 6232 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/k_proj/Linear", "inputIndexes": [ 6232 ], "outputIndexes": [ 6233 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 2218453222, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6233 ], "outputIndexes": [ 6234 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6234 ], "outputIndexes": [ 840 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 804, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_258_output_0", "outputIndexes": [ 841 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 812, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_259_output_0", "outputIndexes": [ 842 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 841, 842, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_105_output_0", "outputIndexes": [ 843 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 840, 843 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_74_output_0", "outputIndexes": [ 844 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 844 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_8/Cast_output_0", "outputIndexes": [ 845 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_8/Mul_1_output_0", "inputIndexes": [ 845 ], "outputIndexes": [ 846 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 2221402364, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 846, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_66_output_0", "outputIndexes": [ 847 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 846 ], "main_type": "NONE", "name": "/Shape_115_output_0", "outputIndexes": [ 848 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 848 ], "main_type": "NONE", "name": "Shape3268", "outputIndexes": [ 849 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 848 ], "main_type": "NONE", "name": "Rank3270", "outputIndexes": [ 850 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 850, 850 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3272", "outputIndexes": [ 851 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 851 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3273", "outputIndexes": [ 852 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 851, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3275", "outputIndexes": [ 853 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 853 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3276", "outputIndexes": [ 854 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 849, 852, 854, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3279", "outputIndexes": [ 855 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 855 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3280", "outputIndexes": [ 856 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 856 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3281", "outputIndexes": [ 857 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 857, 856 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3282", "outputIndexes": [ 858 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 848, 858, 851 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_126_output_0", "outputIndexes": [ 859 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 859, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_25_output_0", "outputIndexes": [ 860 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 860, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_265_output_0", "outputIndexes": [ 861 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 846, 861, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_35_output_0", "outputIndexes": [ 862 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 862 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_17_output_0", "outputIndexes": [ 863 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 860, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_264_output_0", "outputIndexes": [ 864 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 846, 20, 864, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_34_output_0", "outputIndexes": [ 865 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 863, 865 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_108_output_0", "outputIndexes": [ 866 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 866, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_67_output_0", "outputIndexes": [ 867 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 847, 867 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_41_output_0", "outputIndexes": [ 868 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 792 ], "outputIndexes": [ 6235 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6235 ], "outputIndexes": [ 6236 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/v_proj/Linear", "inputIndexes": [ 6236 ], "outputIndexes": [ 6237 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 2221403388, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6237 ], "outputIndexes": [ 6238 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6238 ], "outputIndexes": [ 869 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 804, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_260_output_0", "outputIndexes": [ 870 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 812, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_261_output_0", "outputIndexes": [ 871 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 870, 871, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_106_output_0", "outputIndexes": [ 872 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 869, 872 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_75_output_0", "outputIndexes": [ 873 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 839, 868, 873, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_80_output_0", "outputIndexes": [ 874 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 874 ], "outputIndexes": [ 6239 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6239 ], "outputIndexes": [ 6240 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/o_proj/Linear", "inputIndexes": [ 6240 ], "outputIndexes": [ 6241 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 2224352530, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6241 ], "outputIndexes": [ 6242 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6242 ], "outputIndexes": [ 875 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 791, 875 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_43_output_0", "outputIndexes": [ 876 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_8/Mul_1_output_0", "inputIndexes": [ 876 ], "outputIndexes": [ 877 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 2247945512, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 877 ], "outputIndexes": [ 6243 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6243 ], "outputIndexes": [ 6244 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/gate_proj/Linear", "inputIndexes": [ 6244 ], "outputIndexes": [ 6245 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 2247986472, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6245 ], "outputIndexes": [ 6246 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6246 ], "outputIndexes": [ 878 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 878 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_8/Mul_output_0", "outputIndexes": [ 879 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 877 ], "outputIndexes": [ 6247 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6247 ], "outputIndexes": [ 6248 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/up_proj/Linear", "inputIndexes": [ 6248 ], "outputIndexes": [ 6249 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 2321714494, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6249 ], "outputIndexes": [ 6250 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6250 ], "outputIndexes": [ 880 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 879, 880 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_8/Mul_output_0", "outputIndexes": [ 881 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 881 ], "outputIndexes": [ 6251 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6251 ], "outputIndexes": [ 6252 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/down_proj/Linear", "inputIndexes": [ 6252 ], "outputIndexes": [ 6253 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 2395442516, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6253 ], "outputIndexes": [ 6254 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.8/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6254 ], "outputIndexes": [ 882 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 876, 882 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_44_output_0", "outputIndexes": [ 883 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 883, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_81_output_0", "outputIndexes": [ 884 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_9/Mul_1_output_0", "inputIndexes": [ 884 ], "outputIndexes": [ 885 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 2469170538, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 885 ], "outputIndexes": [ 6255 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6255 ], "outputIndexes": [ 6256 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/q_proj/Linear", "inputIndexes": [ 6256 ], "outputIndexes": [ 6257 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 2469211498, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6257 ], "outputIndexes": [ 6258 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6258 ], "outputIndexes": [ 886 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 885 ], "main_type": "NONE", "name": "/Shape_126_output_0", "outputIndexes": [ 887 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 887 ], "main_type": "NONE", "name": "Shape3611", "outputIndexes": [ 888 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 887 ], "main_type": "NONE", "name": "Rank3613", "outputIndexes": [ 889 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 889, 889 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3615", "outputIndexes": [ 890 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 890 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3616", "outputIndexes": [ 891 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 890, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3618", "outputIndexes": [ 892 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 892 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3619", "outputIndexes": [ 893 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 888, 891, 893, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3622", "outputIndexes": [ 894 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 894 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3623", "outputIndexes": [ 895 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 895, 895 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3625", "outputIndexes": [ 896 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 887, 896, 890 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_138_output_0", "outputIndexes": [ 897 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 897, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_288_output_0", "outputIndexes": [ 898 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 890 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3633", "outputIndexes": [ 899 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 892 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3636", "outputIndexes": [ 900 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 888, 899, 900, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3639", "outputIndexes": [ 901 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 901 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3640", "outputIndexes": [ 902 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 902 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3641", "outputIndexes": [ 903 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 903, 902 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3642", "outputIndexes": [ 904 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 887, 904, 890 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_139_output_0", "outputIndexes": [ 905 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 905, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_289_output_0", "outputIndexes": [ 906 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 898, 906, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_117_output_0", "outputIndexes": [ 907 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 886, 907 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_82_output_0", "outputIndexes": [ 908 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 908 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_9/Cast_output_0", "outputIndexes": [ 909 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_9/Mul_1_output_0", "inputIndexes": [ 909 ], "outputIndexes": [ 910 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 2492804480, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 910, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_72_output_0", "outputIndexes": [ 911 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 910 ], "main_type": "NONE", "name": "/Shape_128_output_0", "outputIndexes": [ 912 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 912 ], "main_type": "NONE", "name": "Shape3761", "outputIndexes": [ 913 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 912 ], "main_type": "NONE", "name": "Rank3763", "outputIndexes": [ 914 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 914, 914 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3765", "outputIndexes": [ 915 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 915 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3766", "outputIndexes": [ 916 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 915, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3768", "outputIndexes": [ 917 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 917 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3769", "outputIndexes": [ 918 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 913, 916, 918, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3772", "outputIndexes": [ 919 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 919 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3773", "outputIndexes": [ 920 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 920 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3774", "outputIndexes": [ 921 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 921, 920 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3775", "outputIndexes": [ 922 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 912, 922, 915 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_140_output_0", "outputIndexes": [ 923 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 923, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_27_output_0", "outputIndexes": [ 924 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 924, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_295_output_0", "outputIndexes": [ 925 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 910, 925, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_37_output_0", "outputIndexes": [ 926 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 926 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_18_output_0", "outputIndexes": [ 927 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 924, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_294_output_0", "outputIndexes": [ 928 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 910, 20, 928, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_36_output_0", "outputIndexes": [ 929 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 927, 929 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_120_output_0", "outputIndexes": [ 930 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 930, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_73_output_0", "outputIndexes": [ 931 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 911, 931 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_45_output_0", "outputIndexes": [ 932 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 885 ], "outputIndexes": [ 6259 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6259 ], "outputIndexes": [ 6260 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/k_proj/Linear", "inputIndexes": [ 6260 ], "outputIndexes": [ 6261 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 2492805504, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6261 ], "outputIndexes": [ 6262 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6262 ], "outputIndexes": [ 933 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 897, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_290_output_0", "outputIndexes": [ 934 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 905, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_291_output_0", "outputIndexes": [ 935 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 934, 935, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_118_output_0", "outputIndexes": [ 936 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 933, 936 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_83_output_0", "outputIndexes": [ 937 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 937 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_9/Cast_output_0", "outputIndexes": [ 938 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_9/Mul_1_output_0", "inputIndexes": [ 938 ], "outputIndexes": [ 939 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 2495754646, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 939, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_74_output_0", "outputIndexes": [ 940 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 939 ], "main_type": "NONE", "name": "/Shape_129_output_0", "outputIndexes": [ 941 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 941 ], "main_type": "NONE", "name": "Shape3657", "outputIndexes": [ 942 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 941 ], "main_type": "NONE", "name": "Rank3659", "outputIndexes": [ 943 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 943, 943 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3661", "outputIndexes": [ 944 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 944 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3662", "outputIndexes": [ 945 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 944, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3664", "outputIndexes": [ 946 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 946 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze3665", "outputIndexes": [ 947 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 942, 945, 947, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice3668", "outputIndexes": [ 948 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 948 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze3669", "outputIndexes": [ 949 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 949 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3670", "outputIndexes": [ 950 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 950, 949 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp3671", "outputIndexes": [ 951 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 941, 951, 944 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_141_output_0", "outputIndexes": [ 952 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 952, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_28_output_0", "outputIndexes": [ 953 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 953, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_297_output_0", "outputIndexes": [ 954 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 939, 954, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_39_output_0", "outputIndexes": [ 955 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 955 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_19_output_0", "outputIndexes": [ 956 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 953, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_296_output_0", "outputIndexes": [ 957 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 939, 20, 957, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_38_output_0", "outputIndexes": [ 958 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 956, 958 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_121_output_0", "outputIndexes": [ 959 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 959, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_75_output_0", "outputIndexes": [ 960 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 940, 960 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_46_output_0", "outputIndexes": [ 961 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 885 ], "outputIndexes": [ 6263 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6263 ], "outputIndexes": [ 6264 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/v_proj/Linear", "inputIndexes": [ 6264 ], "outputIndexes": [ 6265 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 2495755670, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6265 ], "outputIndexes": [ 6266 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6266 ], "outputIndexes": [ 962 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 897, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_292_output_0", "outputIndexes": [ 963 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 905, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_293_output_0", "outputIndexes": [ 964 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 963, 964, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_119_output_0", "outputIndexes": [ 965 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 962, 965 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_84_output_0", "outputIndexes": [ 966 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 932, 961, 966, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_89_output_0", "outputIndexes": [ 967 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 967 ], "outputIndexes": [ 6267 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6267 ], "outputIndexes": [ 6268 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/o_proj/Linear", "inputIndexes": [ 6268 ], "outputIndexes": [ 6269 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 2498704812, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6269 ], "outputIndexes": [ 6270 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6270 ], "outputIndexes": [ 968 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 884, 968 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_48_output_0", "outputIndexes": [ 969 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_9/Mul_1_output_0", "inputIndexes": [ 969 ], "outputIndexes": [ 970 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 2522297794, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 970 ], "outputIndexes": [ 6271 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6271 ], "outputIndexes": [ 6272 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/gate_proj/Linear", "inputIndexes": [ 6272 ], "outputIndexes": [ 6273 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 2522338754, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6273 ], "outputIndexes": [ 6274 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6274 ], "outputIndexes": [ 971 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 971 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_9/Mul_output_0", "outputIndexes": [ 972 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 970 ], "outputIndexes": [ 6275 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6275 ], "outputIndexes": [ 6276 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/up_proj/Linear", "inputIndexes": [ 6276 ], "outputIndexes": [ 6277 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 2596066776, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6277 ], "outputIndexes": [ 6278 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6278 ], "outputIndexes": [ 973 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 972, 973 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_9/Mul_output_0", "outputIndexes": [ 974 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 974 ], "outputIndexes": [ 6279 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6279 ], "outputIndexes": [ 6280 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/down_proj/Linear", "inputIndexes": [ 6280 ], "outputIndexes": [ 6281 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 2669794798, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6281 ], "outputIndexes": [ 6282 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.9/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6282 ], "outputIndexes": [ 975 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 969, 975 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_49_output_0", "outputIndexes": [ 976 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 976, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_90_output_0", "outputIndexes": [ 977 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_10/Mul_1_output_0", "inputIndexes": [ 977 ], "outputIndexes": [ 978 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 2743522820, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 978 ], "outputIndexes": [ 6283 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6283 ], "outputIndexes": [ 6284 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/q_proj/Linear", "inputIndexes": [ 6284 ], "outputIndexes": [ 6285 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 2743563780, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6285 ], "outputIndexes": [ 6286 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6286 ], "outputIndexes": [ 979 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 978 ], "main_type": "NONE", "name": "/Shape_140_output_0", "outputIndexes": [ 980 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 980 ], "main_type": "NONE", "name": "Shape4000", "outputIndexes": [ 981 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 980 ], "main_type": "NONE", "name": "Rank4002", "outputIndexes": [ 982 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 982, 982 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4004", "outputIndexes": [ 983 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 983 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4005", "outputIndexes": [ 984 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 983, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4007", "outputIndexes": [ 985 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 985 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4008", "outputIndexes": [ 986 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 981, 984, 986, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4011", "outputIndexes": [ 987 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 987 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4012", "outputIndexes": [ 988 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 988, 988 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4014", "outputIndexes": [ 989 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 980, 989, 983 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_153_output_0", "outputIndexes": [ 990 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 990, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_320_output_0", "outputIndexes": [ 991 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 983 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4022", "outputIndexes": [ 992 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 985 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4025", "outputIndexes": [ 993 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 981, 992, 993, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4028", "outputIndexes": [ 994 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 994 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4029", "outputIndexes": [ 995 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 995 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4030", "outputIndexes": [ 996 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 996, 995 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4031", "outputIndexes": [ 997 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 980, 997, 983 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_154_output_0", "outputIndexes": [ 998 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 998, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_321_output_0", "outputIndexes": [ 999 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 991, 999, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_130_output_0", "outputIndexes": [ 1000 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 979, 1000 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_91_output_0", "outputIndexes": [ 1001 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1001 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_10/Cast_output_0", "outputIndexes": [ 1002 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_10/Mul_1_output_0", "inputIndexes": [ 1002 ], "outputIndexes": [ 1003 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 2767156762, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1003, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_80_output_0", "outputIndexes": [ 1004 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1003 ], "main_type": "NONE", "name": "/Shape_142_output_0", "outputIndexes": [ 1005 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1005 ], "main_type": "NONE", "name": "Shape4150", "outputIndexes": [ 1006 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1005 ], "main_type": "NONE", "name": "Rank4152", "outputIndexes": [ 1007 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1007, 1007 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4154", "outputIndexes": [ 1008 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1008 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4155", "outputIndexes": [ 1009 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1008, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4157", "outputIndexes": [ 1010 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1010 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4158", "outputIndexes": [ 1011 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1006, 1009, 1011, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4161", "outputIndexes": [ 1012 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1012 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4162", "outputIndexes": [ 1013 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1013 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4163", "outputIndexes": [ 1014 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1014, 1013 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4164", "outputIndexes": [ 1015 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1005, 1015, 1008 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_155_output_0", "outputIndexes": [ 1016 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1016, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_30_output_0", "outputIndexes": [ 1017 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1017, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_327_output_0", "outputIndexes": [ 1018 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1003, 1018, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_41_output_0", "outputIndexes": [ 1019 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1019 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_20_output_0", "outputIndexes": [ 1020 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1017, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_326_output_0", "outputIndexes": [ 1021 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1003, 20, 1021, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_40_output_0", "outputIndexes": [ 1022 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1020, 1022 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_133_output_0", "outputIndexes": [ 1023 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1023, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_81_output_0", "outputIndexes": [ 1024 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1004, 1024 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_50_output_0", "outputIndexes": [ 1025 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 978 ], "outputIndexes": [ 6287 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6287 ], "outputIndexes": [ 6288 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/k_proj/Linear", "inputIndexes": [ 6288 ], "outputIndexes": [ 6289 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 2767157786, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6289 ], "outputIndexes": [ 6290 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6290 ], "outputIndexes": [ 1026 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 990, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_322_output_0", "outputIndexes": [ 1027 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 998, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_323_output_0", "outputIndexes": [ 1028 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1027, 1028, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_131_output_0", "outputIndexes": [ 1029 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1026, 1029 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_92_output_0", "outputIndexes": [ 1030 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1030 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_10/Cast_output_0", "outputIndexes": [ 1031 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_10/Mul_1_output_0", "inputIndexes": [ 1031 ], "outputIndexes": [ 1032 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 2770106928, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1032, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_82_output_0", "outputIndexes": [ 1033 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1032 ], "main_type": "NONE", "name": "/Shape_143_output_0", "outputIndexes": [ 1034 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1034 ], "main_type": "NONE", "name": "Shape4046", "outputIndexes": [ 1035 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1034 ], "main_type": "NONE", "name": "Rank4048", "outputIndexes": [ 1036 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1036, 1036 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4050", "outputIndexes": [ 1037 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1037 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4051", "outputIndexes": [ 1038 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1037, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4053", "outputIndexes": [ 1039 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1039 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4054", "outputIndexes": [ 1040 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1035, 1038, 1040, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4057", "outputIndexes": [ 1041 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1041 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4058", "outputIndexes": [ 1042 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1042 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4059", "outputIndexes": [ 1043 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1043, 1042 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4060", "outputIndexes": [ 1044 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1034, 1044, 1037 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_156_output_0", "outputIndexes": [ 1045 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1045, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_31_output_0", "outputIndexes": [ 1046 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1046, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_329_output_0", "outputIndexes": [ 1047 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1032, 1047, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_43_output_0", "outputIndexes": [ 1048 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1048 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_21_output_0", "outputIndexes": [ 1049 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1046, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_328_output_0", "outputIndexes": [ 1050 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1032, 20, 1050, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_42_output_0", "outputIndexes": [ 1051 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1049, 1051 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_134_output_0", "outputIndexes": [ 1052 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1052, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_83_output_0", "outputIndexes": [ 1053 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1033, 1053 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_51_output_0", "outputIndexes": [ 1054 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 978 ], "outputIndexes": [ 6291 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6291 ], "outputIndexes": [ 6292 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/v_proj/Linear", "inputIndexes": [ 6292 ], "outputIndexes": [ 6293 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 2770107952, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6293 ], "outputIndexes": [ 6294 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6294 ], "outputIndexes": [ 1055 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 990, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_324_output_0", "outputIndexes": [ 1056 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 998, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_325_output_0", "outputIndexes": [ 1057 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1056, 1057, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_132_output_0", "outputIndexes": [ 1058 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1055, 1058 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_93_output_0", "outputIndexes": [ 1059 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1025, 1054, 1059, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_98_output_0", "outputIndexes": [ 1060 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1060 ], "outputIndexes": [ 6295 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6295 ], "outputIndexes": [ 6296 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/o_proj/Linear", "inputIndexes": [ 6296 ], "outputIndexes": [ 6297 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 2773057094, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6297 ], "outputIndexes": [ 6298 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6298 ], "outputIndexes": [ 1061 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 977, 1061 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_53_output_0", "outputIndexes": [ 1062 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_10/Mul_1_output_0", "inputIndexes": [ 1062 ], "outputIndexes": [ 1063 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 2796650076, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1063 ], "outputIndexes": [ 6299 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6299 ], "outputIndexes": [ 6300 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/gate_proj/Linear", "inputIndexes": [ 6300 ], "outputIndexes": [ 6301 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 2796691036, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6301 ], "outputIndexes": [ 6302 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6302 ], "outputIndexes": [ 1064 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1064 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_10/Mul_output_0", "outputIndexes": [ 1065 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1063 ], "outputIndexes": [ 6303 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6303 ], "outputIndexes": [ 6304 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/up_proj/Linear", "inputIndexes": [ 6304 ], "outputIndexes": [ 6305 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 2870419058, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6305 ], "outputIndexes": [ 6306 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6306 ], "outputIndexes": [ 1066 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1065, 1066 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_10/Mul_output_0", "outputIndexes": [ 1067 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1067 ], "outputIndexes": [ 6307 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6307 ], "outputIndexes": [ 6308 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/down_proj/Linear", "inputIndexes": [ 6308 ], "outputIndexes": [ 6309 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 2944147080, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6309 ], "outputIndexes": [ 6310 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.10/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6310 ], "outputIndexes": [ 1068 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1062, 1068 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_54_output_0", "outputIndexes": [ 1069 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1069, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_99_output_0", "outputIndexes": [ 1070 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_11/Mul_1_output_0", "inputIndexes": [ 1070 ], "outputIndexes": [ 1071 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 3017875102, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1071 ], "outputIndexes": [ 6311 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6311 ], "outputIndexes": [ 6312 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/q_proj/Linear", "inputIndexes": [ 6312 ], "outputIndexes": [ 6313 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 3017916062, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6313 ], "outputIndexes": [ 6314 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6314 ], "outputIndexes": [ 1072 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1071 ], "main_type": "NONE", "name": "/Shape_154_output_0", "outputIndexes": [ 1073 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1073 ], "main_type": "NONE", "name": "Shape4389", "outputIndexes": [ 1074 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1073 ], "main_type": "NONE", "name": "Rank4391", "outputIndexes": [ 1075 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1075, 1075 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4393", "outputIndexes": [ 1076 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1076 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4394", "outputIndexes": [ 1077 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1076, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4396", "outputIndexes": [ 1078 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1078 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4397", "outputIndexes": [ 1079 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1074, 1077, 1079, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4400", "outputIndexes": [ 1080 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1080 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4401", "outputIndexes": [ 1081 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1081, 1081 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4403", "outputIndexes": [ 1082 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1073, 1082, 1076 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_168_output_0", "outputIndexes": [ 1083 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1083, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_352_output_0", "outputIndexes": [ 1084 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1076 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4411", "outputIndexes": [ 1085 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1078 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4414", "outputIndexes": [ 1086 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1074, 1085, 1086, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4417", "outputIndexes": [ 1087 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1087 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4418", "outputIndexes": [ 1088 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1088 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4419", "outputIndexes": [ 1089 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1089, 1088 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4420", "outputIndexes": [ 1090 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1073, 1090, 1076 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_169_output_0", "outputIndexes": [ 1091 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1091, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_353_output_0", "outputIndexes": [ 1092 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1084, 1092, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_143_output_0", "outputIndexes": [ 1093 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1072, 1093 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_100_output_0", "outputIndexes": [ 1094 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1094 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_11/Cast_output_0", "outputIndexes": [ 1095 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_11/Mul_1_output_0", "inputIndexes": [ 1095 ], "outputIndexes": [ 1096 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 3041509044, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1096, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_88_output_0", "outputIndexes": [ 1097 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1096 ], "main_type": "NONE", "name": "/Shape_156_output_0", "outputIndexes": [ 1098 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1098 ], "main_type": "NONE", "name": "Shape4539", "outputIndexes": [ 1099 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1098 ], "main_type": "NONE", "name": "Rank4541", "outputIndexes": [ 1100 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1100, 1100 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4543", "outputIndexes": [ 1101 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1101 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4544", "outputIndexes": [ 1102 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1101, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4546", "outputIndexes": [ 1103 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1103 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4547", "outputIndexes": [ 1104 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1099, 1102, 1104, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4550", "outputIndexes": [ 1105 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1105 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4551", "outputIndexes": [ 1106 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1106 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4552", "outputIndexes": [ 1107 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1107, 1106 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4553", "outputIndexes": [ 1108 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1098, 1108, 1101 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_170_output_0", "outputIndexes": [ 1109 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1109, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_33_output_0", "outputIndexes": [ 1110 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1110, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_359_output_0", "outputIndexes": [ 1111 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1096, 1111, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_45_output_0", "outputIndexes": [ 1112 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1112 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_22_output_0", "outputIndexes": [ 1113 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1110, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_358_output_0", "outputIndexes": [ 1114 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1096, 20, 1114, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_44_output_0", "outputIndexes": [ 1115 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1113, 1115 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_146_output_0", "outputIndexes": [ 1116 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1116, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_89_output_0", "outputIndexes": [ 1117 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1097, 1117 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_55_output_0", "outputIndexes": [ 1118 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1071 ], "outputIndexes": [ 6315 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6315 ], "outputIndexes": [ 6316 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/k_proj/Linear", "inputIndexes": [ 6316 ], "outputIndexes": [ 6317 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 3041510068, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6317 ], "outputIndexes": [ 6318 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6318 ], "outputIndexes": [ 1119 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1083, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_354_output_0", "outputIndexes": [ 1120 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1091, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_355_output_0", "outputIndexes": [ 1121 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1120, 1121, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_144_output_0", "outputIndexes": [ 1122 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1119, 1122 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_101_output_0", "outputIndexes": [ 1123 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1123 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_11/Cast_output_0", "outputIndexes": [ 1124 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_11/Mul_1_output_0", "inputIndexes": [ 1124 ], "outputIndexes": [ 1125 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 3044459210, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1125, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_90_output_0", "outputIndexes": [ 1126 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1125 ], "main_type": "NONE", "name": "/Shape_157_output_0", "outputIndexes": [ 1127 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1127 ], "main_type": "NONE", "name": "Shape4435", "outputIndexes": [ 1128 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1127 ], "main_type": "NONE", "name": "Rank4437", "outputIndexes": [ 1129 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1129, 1129 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4439", "outputIndexes": [ 1130 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1130 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4440", "outputIndexes": [ 1131 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1130, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4442", "outputIndexes": [ 1132 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1132 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4443", "outputIndexes": [ 1133 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1128, 1131, 1133, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4446", "outputIndexes": [ 1134 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1134 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4447", "outputIndexes": [ 1135 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1135 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4448", "outputIndexes": [ 1136 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1136, 1135 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4449", "outputIndexes": [ 1137 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1127, 1137, 1130 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_171_output_0", "outputIndexes": [ 1138 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1138, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_34_output_0", "outputIndexes": [ 1139 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1139, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_361_output_0", "outputIndexes": [ 1140 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1125, 1140, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_47_output_0", "outputIndexes": [ 1141 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1141 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_23_output_0", "outputIndexes": [ 1142 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1139, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_360_output_0", "outputIndexes": [ 1143 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1125, 20, 1143, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_46_output_0", "outputIndexes": [ 1144 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1142, 1144 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_147_output_0", "outputIndexes": [ 1145 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1145, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_91_output_0", "outputIndexes": [ 1146 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1126, 1146 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_56_output_0", "outputIndexes": [ 1147 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1071 ], "outputIndexes": [ 6319 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6319 ], "outputIndexes": [ 6320 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/v_proj/Linear", "inputIndexes": [ 6320 ], "outputIndexes": [ 6321 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 3044460234, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6321 ], "outputIndexes": [ 6322 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6322 ], "outputIndexes": [ 1148 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1083, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_356_output_0", "outputIndexes": [ 1149 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1091, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_357_output_0", "outputIndexes": [ 1150 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1149, 1150, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_145_output_0", "outputIndexes": [ 1151 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1148, 1151 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_102_output_0", "outputIndexes": [ 1152 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1118, 1147, 1152, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_107_output_0", "outputIndexes": [ 1153 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1153 ], "outputIndexes": [ 6323 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6323 ], "outputIndexes": [ 6324 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/o_proj/Linear", "inputIndexes": [ 6324 ], "outputIndexes": [ 6325 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 3047409376, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6325 ], "outputIndexes": [ 6326 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6326 ], "outputIndexes": [ 1154 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1070, 1154 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_58_output_0", "outputIndexes": [ 1155 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_11/Mul_1_output_0", "inputIndexes": [ 1155 ], "outputIndexes": [ 1156 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 3071002358, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1156 ], "outputIndexes": [ 6327 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6327 ], "outputIndexes": [ 6328 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/gate_proj/Linear", "inputIndexes": [ 6328 ], "outputIndexes": [ 6329 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 3071043318, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6329 ], "outputIndexes": [ 6330 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6330 ], "outputIndexes": [ 1157 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1157 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_11/Mul_output_0", "outputIndexes": [ 1158 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1156 ], "outputIndexes": [ 6331 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6331 ], "outputIndexes": [ 6332 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/up_proj/Linear", "inputIndexes": [ 6332 ], "outputIndexes": [ 6333 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 3144771340, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6333 ], "outputIndexes": [ 6334 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6334 ], "outputIndexes": [ 1159 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1158, 1159 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_11/Mul_output_0", "outputIndexes": [ 1160 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1160 ], "outputIndexes": [ 6335 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6335 ], "outputIndexes": [ 6336 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/down_proj/Linear", "inputIndexes": [ 6336 ], "outputIndexes": [ 6337 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 3218499362, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6337 ], "outputIndexes": [ 6338 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.11/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6338 ], "outputIndexes": [ 1161 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1155, 1161 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_59_output_0", "outputIndexes": [ 1162 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1162, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_108_output_0", "outputIndexes": [ 1163 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_12/Mul_1_output_0", "inputIndexes": [ 1163 ], "outputIndexes": [ 1164 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 3292227384, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1164 ], "outputIndexes": [ 6339 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6339 ], "outputIndexes": [ 6340 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/q_proj/Linear", "inputIndexes": [ 6340 ], "outputIndexes": [ 6341 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 3292268344, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6341 ], "outputIndexes": [ 6342 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6342 ], "outputIndexes": [ 1165 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1164 ], "main_type": "NONE", "name": "/Shape_168_output_0", "outputIndexes": [ 1166 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1166 ], "main_type": "NONE", "name": "Shape4778", "outputIndexes": [ 1167 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1166 ], "main_type": "NONE", "name": "Rank4780", "outputIndexes": [ 1168 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1168, 1168 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4782", "outputIndexes": [ 1169 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1169 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4783", "outputIndexes": [ 1170 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1169, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4785", "outputIndexes": [ 1171 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1171 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4786", "outputIndexes": [ 1172 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1167, 1170, 1172, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4789", "outputIndexes": [ 1173 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1173 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4790", "outputIndexes": [ 1174 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1174, 1174 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4792", "outputIndexes": [ 1175 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1166, 1175, 1169 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_183_output_0", "outputIndexes": [ 1176 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1176, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_384_output_0", "outputIndexes": [ 1177 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1169 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4800", "outputIndexes": [ 1178 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1171 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4803", "outputIndexes": [ 1179 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1167, 1178, 1179, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4806", "outputIndexes": [ 1180 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1180 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4807", "outputIndexes": [ 1181 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1181 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4808", "outputIndexes": [ 1182 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1182, 1181 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4809", "outputIndexes": [ 1183 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1166, 1183, 1169 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_184_output_0", "outputIndexes": [ 1184 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1184, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_385_output_0", "outputIndexes": [ 1185 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1177, 1185, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_156_output_0", "outputIndexes": [ 1186 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1165, 1186 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_109_output_0", "outputIndexes": [ 1187 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1187 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_12/Cast_output_0", "outputIndexes": [ 1188 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_12/Mul_1_output_0", "inputIndexes": [ 1188 ], "outputIndexes": [ 1189 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 3315861326, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1189, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_96_output_0", "outputIndexes": [ 1190 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1189 ], "main_type": "NONE", "name": "/Shape_170_output_0", "outputIndexes": [ 1191 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1191 ], "main_type": "NONE", "name": "Shape4928", "outputIndexes": [ 1192 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1191 ], "main_type": "NONE", "name": "Rank4930", "outputIndexes": [ 1193 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1193, 1193 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4932", "outputIndexes": [ 1194 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1194 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4933", "outputIndexes": [ 1195 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1194, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4935", "outputIndexes": [ 1196 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1196 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4936", "outputIndexes": [ 1197 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1192, 1195, 1197, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4939", "outputIndexes": [ 1198 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1198 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4940", "outputIndexes": [ 1199 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1199 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4941", "outputIndexes": [ 1200 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1200, 1199 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4942", "outputIndexes": [ 1201 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1191, 1201, 1194 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_185_output_0", "outputIndexes": [ 1202 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1202, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_36_output_0", "outputIndexes": [ 1203 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1203, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_391_output_0", "outputIndexes": [ 1204 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1189, 1204, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_49_output_0", "outputIndexes": [ 1205 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1205 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_24_output_0", "outputIndexes": [ 1206 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1203, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_390_output_0", "outputIndexes": [ 1207 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1189, 20, 1207, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_48_output_0", "outputIndexes": [ 1208 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1206, 1208 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_159_output_0", "outputIndexes": [ 1209 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1209, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_97_output_0", "outputIndexes": [ 1210 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1190, 1210 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_60_output_0", "outputIndexes": [ 1211 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1164 ], "outputIndexes": [ 6343 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6343 ], "outputIndexes": [ 6344 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/k_proj/Linear", "inputIndexes": [ 6344 ], "outputIndexes": [ 6345 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 3315862350, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6345 ], "outputIndexes": [ 6346 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6346 ], "outputIndexes": [ 1212 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1176, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_386_output_0", "outputIndexes": [ 1213 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1184, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_387_output_0", "outputIndexes": [ 1214 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1213, 1214, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_157_output_0", "outputIndexes": [ 1215 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1212, 1215 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_110_output_0", "outputIndexes": [ 1216 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1216 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_12/Cast_output_0", "outputIndexes": [ 1217 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_12/Mul_1_output_0", "inputIndexes": [ 1217 ], "outputIndexes": [ 1218 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 3318811492, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1218, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_98_output_0", "outputIndexes": [ 1219 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1218 ], "main_type": "NONE", "name": "/Shape_171_output_0", "outputIndexes": [ 1220 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1220 ], "main_type": "NONE", "name": "Shape4824", "outputIndexes": [ 1221 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1220 ], "main_type": "NONE", "name": "Rank4826", "outputIndexes": [ 1222 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1222, 1222 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4828", "outputIndexes": [ 1223 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1223 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4829", "outputIndexes": [ 1224 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1223, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4831", "outputIndexes": [ 1225 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1225 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze4832", "outputIndexes": [ 1226 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1221, 1224, 1226, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice4835", "outputIndexes": [ 1227 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1227 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze4836", "outputIndexes": [ 1228 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1228 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4837", "outputIndexes": [ 1229 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1229, 1228 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp4838", "outputIndexes": [ 1230 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1220, 1230, 1223 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_186_output_0", "outputIndexes": [ 1231 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1231, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_37_output_0", "outputIndexes": [ 1232 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1232, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_393_output_0", "outputIndexes": [ 1233 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1218, 1233, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_51_output_0", "outputIndexes": [ 1234 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1234 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_25_output_0", "outputIndexes": [ 1235 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1232, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_392_output_0", "outputIndexes": [ 1236 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1218, 20, 1236, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_50_output_0", "outputIndexes": [ 1237 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1235, 1237 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_160_output_0", "outputIndexes": [ 1238 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1238, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_99_output_0", "outputIndexes": [ 1239 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1219, 1239 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_61_output_0", "outputIndexes": [ 1240 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1164 ], "outputIndexes": [ 6347 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6347 ], "outputIndexes": [ 6348 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/v_proj/Linear", "inputIndexes": [ 6348 ], "outputIndexes": [ 6349 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 3318812516, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6349 ], "outputIndexes": [ 6350 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6350 ], "outputIndexes": [ 1241 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1176, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_388_output_0", "outputIndexes": [ 1242 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1184, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_389_output_0", "outputIndexes": [ 1243 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1242, 1243, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_158_output_0", "outputIndexes": [ 1244 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1241, 1244 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_111_output_0", "outputIndexes": [ 1245 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1211, 1240, 1245, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_116_output_0", "outputIndexes": [ 1246 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1246 ], "outputIndexes": [ 6351 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6351 ], "outputIndexes": [ 6352 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/o_proj/Linear", "inputIndexes": [ 6352 ], "outputIndexes": [ 6353 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 3321761658, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6353 ], "outputIndexes": [ 6354 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6354 ], "outputIndexes": [ 1247 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1163, 1247 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_63_output_0", "outputIndexes": [ 1248 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_12/Mul_1_output_0", "inputIndexes": [ 1248 ], "outputIndexes": [ 1249 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 3345354640, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1249 ], "outputIndexes": [ 6355 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6355 ], "outputIndexes": [ 6356 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/gate_proj/Linear", "inputIndexes": [ 6356 ], "outputIndexes": [ 6357 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 3345395600, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6357 ], "outputIndexes": [ 6358 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6358 ], "outputIndexes": [ 1250 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1250 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_12/Mul_output_0", "outputIndexes": [ 1251 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1249 ], "outputIndexes": [ 6359 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6359 ], "outputIndexes": [ 6360 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/up_proj/Linear", "inputIndexes": [ 6360 ], "outputIndexes": [ 6361 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 3419123622, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6361 ], "outputIndexes": [ 6362 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6362 ], "outputIndexes": [ 1252 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1251, 1252 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_12/Mul_output_0", "outputIndexes": [ 1253 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1253 ], "outputIndexes": [ 6363 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6363 ], "outputIndexes": [ 6364 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/down_proj/Linear", "inputIndexes": [ 6364 ], "outputIndexes": [ 6365 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 3492851644, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6365 ], "outputIndexes": [ 6366 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.12/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6366 ], "outputIndexes": [ 1254 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1248, 1254 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_64_output_0", "outputIndexes": [ 1255 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1255, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_117_output_0", "outputIndexes": [ 1256 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_13/Mul_1_output_0", "inputIndexes": [ 1256 ], "outputIndexes": [ 1257 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 3566579666, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1257 ], "outputIndexes": [ 6367 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6367 ], "outputIndexes": [ 6368 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/q_proj/Linear", "inputIndexes": [ 6368 ], "outputIndexes": [ 6369 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 3566620626, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6369 ], "outputIndexes": [ 6370 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6370 ], "outputIndexes": [ 1258 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1257 ], "main_type": "NONE", "name": "/Shape_182_output_0", "outputIndexes": [ 1259 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1259 ], "main_type": "NONE", "name": "Shape5167", "outputIndexes": [ 1260 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1259 ], "main_type": "NONE", "name": "Rank5169", "outputIndexes": [ 1261 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1261, 1261 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5171", "outputIndexes": [ 1262 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1262 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5172", "outputIndexes": [ 1263 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1262, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5174", "outputIndexes": [ 1264 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1264 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5175", "outputIndexes": [ 1265 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1260, 1263, 1265, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5178", "outputIndexes": [ 1266 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1266 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5179", "outputIndexes": [ 1267 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1267, 1267 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5181", "outputIndexes": [ 1268 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1259, 1268, 1262 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_198_output_0", "outputIndexes": [ 1269 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1269, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_416_output_0", "outputIndexes": [ 1270 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1262 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5189", "outputIndexes": [ 1271 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1264 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5192", "outputIndexes": [ 1272 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1260, 1271, 1272, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5195", "outputIndexes": [ 1273 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1273 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5196", "outputIndexes": [ 1274 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1274 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5197", "outputIndexes": [ 1275 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1275, 1274 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5198", "outputIndexes": [ 1276 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1259, 1276, 1262 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_199_output_0", "outputIndexes": [ 1277 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1277, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_417_output_0", "outputIndexes": [ 1278 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1270, 1278, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_169_output_0", "outputIndexes": [ 1279 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1258, 1279 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_118_output_0", "outputIndexes": [ 1280 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1280 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_13/Cast_output_0", "outputIndexes": [ 1281 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_13/Mul_1_output_0", "inputIndexes": [ 1281 ], "outputIndexes": [ 1282 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 3590213608, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1282, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_104_output_0", "outputIndexes": [ 1283 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1282 ], "main_type": "NONE", "name": "/Shape_184_output_0", "outputIndexes": [ 1284 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1284 ], "main_type": "NONE", "name": "Shape5317", "outputIndexes": [ 1285 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1284 ], "main_type": "NONE", "name": "Rank5319", "outputIndexes": [ 1286 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1286, 1286 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5321", "outputIndexes": [ 1287 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1287 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5322", "outputIndexes": [ 1288 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1287, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5324", "outputIndexes": [ 1289 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1289 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5325", "outputIndexes": [ 1290 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1285, 1288, 1290, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5328", "outputIndexes": [ 1291 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1291 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5329", "outputIndexes": [ 1292 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1292 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5330", "outputIndexes": [ 1293 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1293, 1292 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5331", "outputIndexes": [ 1294 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1284, 1294, 1287 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_200_output_0", "outputIndexes": [ 1295 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1295, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_39_output_0", "outputIndexes": [ 1296 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1296, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_423_output_0", "outputIndexes": [ 1297 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1282, 1297, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_53_output_0", "outputIndexes": [ 1298 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1298 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_26_output_0", "outputIndexes": [ 1299 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1296, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_422_output_0", "outputIndexes": [ 1300 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1282, 20, 1300, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_52_output_0", "outputIndexes": [ 1301 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1299, 1301 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_172_output_0", "outputIndexes": [ 1302 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1302, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_105_output_0", "outputIndexes": [ 1303 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1283, 1303 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_65_output_0", "outputIndexes": [ 1304 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1257 ], "outputIndexes": [ 6371 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6371 ], "outputIndexes": [ 6372 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/k_proj/Linear", "inputIndexes": [ 6372 ], "outputIndexes": [ 6373 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 3590214632, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6373 ], "outputIndexes": [ 6374 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6374 ], "outputIndexes": [ 1305 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1269, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_418_output_0", "outputIndexes": [ 1306 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1277, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_419_output_0", "outputIndexes": [ 1307 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1306, 1307, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_170_output_0", "outputIndexes": [ 1308 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1305, 1308 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_119_output_0", "outputIndexes": [ 1309 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1309 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_13/Cast_output_0", "outputIndexes": [ 1310 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_13/Mul_1_output_0", "inputIndexes": [ 1310 ], "outputIndexes": [ 1311 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 3593163774, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1311, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_106_output_0", "outputIndexes": [ 1312 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1311 ], "main_type": "NONE", "name": "/Shape_185_output_0", "outputIndexes": [ 1313 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1313 ], "main_type": "NONE", "name": "Shape5213", "outputIndexes": [ 1314 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1313 ], "main_type": "NONE", "name": "Rank5215", "outputIndexes": [ 1315 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1315, 1315 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5217", "outputIndexes": [ 1316 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1316 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5218", "outputIndexes": [ 1317 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1316, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5220", "outputIndexes": [ 1318 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1318 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5221", "outputIndexes": [ 1319 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1314, 1317, 1319, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5224", "outputIndexes": [ 1320 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1320 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5225", "outputIndexes": [ 1321 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1321 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5226", "outputIndexes": [ 1322 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1322, 1321 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5227", "outputIndexes": [ 1323 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1313, 1323, 1316 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_201_output_0", "outputIndexes": [ 1324 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1324, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_40_output_0", "outputIndexes": [ 1325 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1325, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_425_output_0", "outputIndexes": [ 1326 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1311, 1326, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_55_output_0", "outputIndexes": [ 1327 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1327 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_27_output_0", "outputIndexes": [ 1328 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1325, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_424_output_0", "outputIndexes": [ 1329 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1311, 20, 1329, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_54_output_0", "outputIndexes": [ 1330 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1328, 1330 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_173_output_0", "outputIndexes": [ 1331 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1331, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_107_output_0", "outputIndexes": [ 1332 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1312, 1332 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_66_output_0", "outputIndexes": [ 1333 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1257 ], "outputIndexes": [ 6375 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6375 ], "outputIndexes": [ 6376 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/v_proj/Linear", "inputIndexes": [ 6376 ], "outputIndexes": [ 6377 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 3593164798, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6377 ], "outputIndexes": [ 6378 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6378 ], "outputIndexes": [ 1334 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1269, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_420_output_0", "outputIndexes": [ 1335 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1277, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_421_output_0", "outputIndexes": [ 1336 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1335, 1336, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_171_output_0", "outputIndexes": [ 1337 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1334, 1337 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_120_output_0", "outputIndexes": [ 1338 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1304, 1333, 1338, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_125_output_0", "outputIndexes": [ 1339 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1339 ], "outputIndexes": [ 6379 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6379 ], "outputIndexes": [ 6380 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/o_proj/Linear", "inputIndexes": [ 6380 ], "outputIndexes": [ 6381 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 3596113940, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6381 ], "outputIndexes": [ 6382 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6382 ], "outputIndexes": [ 1340 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1256, 1340 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_68_output_0", "outputIndexes": [ 1341 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_13/Mul_1_output_0", "inputIndexes": [ 1341 ], "outputIndexes": [ 1342 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 3619706922, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1342 ], "outputIndexes": [ 6383 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6383 ], "outputIndexes": [ 6384 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/gate_proj/Linear", "inputIndexes": [ 6384 ], "outputIndexes": [ 6385 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 3619747882, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6385 ], "outputIndexes": [ 6386 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6386 ], "outputIndexes": [ 1343 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1343 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_13/Mul_output_0", "outputIndexes": [ 1344 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1342 ], "outputIndexes": [ 6387 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6387 ], "outputIndexes": [ 6388 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/up_proj/Linear", "inputIndexes": [ 6388 ], "outputIndexes": [ 6389 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 3693475904, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6389 ], "outputIndexes": [ 6390 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6390 ], "outputIndexes": [ 1345 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1344, 1345 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_13/Mul_output_0", "outputIndexes": [ 1346 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1346 ], "outputIndexes": [ 6391 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6391 ], "outputIndexes": [ 6392 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/down_proj/Linear", "inputIndexes": [ 6392 ], "outputIndexes": [ 6393 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 3767203926, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6393 ], "outputIndexes": [ 6394 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.13/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6394 ], "outputIndexes": [ 1347 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1341, 1347 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_69_output_0", "outputIndexes": [ 1348 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1348, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_126_output_0", "outputIndexes": [ 1349 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_14/Mul_1_output_0", "inputIndexes": [ 1349 ], "outputIndexes": [ 1350 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 3840931948, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1350 ], "outputIndexes": [ 6395 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6395 ], "outputIndexes": [ 6396 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/q_proj/Linear", "inputIndexes": [ 6396 ], "outputIndexes": [ 6397 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 3840972908, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6397 ], "outputIndexes": [ 6398 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6398 ], "outputIndexes": [ 1351 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1350 ], "main_type": "NONE", "name": "/Shape_196_output_0", "outputIndexes": [ 1352 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1352 ], "main_type": "NONE", "name": "Shape5556", "outputIndexes": [ 1353 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1352 ], "main_type": "NONE", "name": "Rank5558", "outputIndexes": [ 1354 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1354, 1354 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5560", "outputIndexes": [ 1355 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1355 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5561", "outputIndexes": [ 1356 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1355, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5563", "outputIndexes": [ 1357 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1357 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5564", "outputIndexes": [ 1358 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1353, 1356, 1358, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5567", "outputIndexes": [ 1359 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1359 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5568", "outputIndexes": [ 1360 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1360, 1360 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5570", "outputIndexes": [ 1361 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1352, 1361, 1355 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_213_output_0", "outputIndexes": [ 1362 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1362, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_448_output_0", "outputIndexes": [ 1363 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1355 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5578", "outputIndexes": [ 1364 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1357 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5581", "outputIndexes": [ 1365 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1353, 1364, 1365, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5584", "outputIndexes": [ 1366 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1366 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5585", "outputIndexes": [ 1367 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1367 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5586", "outputIndexes": [ 1368 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1368, 1367 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5587", "outputIndexes": [ 1369 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1352, 1369, 1355 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_214_output_0", "outputIndexes": [ 1370 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1370, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_449_output_0", "outputIndexes": [ 1371 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1363, 1371, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_182_output_0", "outputIndexes": [ 1372 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1351, 1372 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_127_output_0", "outputIndexes": [ 1373 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1373 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_14/Cast_output_0", "outputIndexes": [ 1374 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_14/Mul_1_output_0", "inputIndexes": [ 1374 ], "outputIndexes": [ 1375 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 3864565890, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1375, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_112_output_0", "outputIndexes": [ 1376 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1375 ], "main_type": "NONE", "name": "/Shape_198_output_0", "outputIndexes": [ 1377 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1377 ], "main_type": "NONE", "name": "Shape5706", "outputIndexes": [ 1378 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1377 ], "main_type": "NONE", "name": "Rank5708", "outputIndexes": [ 1379 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1379, 1379 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5710", "outputIndexes": [ 1380 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1380 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5711", "outputIndexes": [ 1381 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1380, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5713", "outputIndexes": [ 1382 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1382 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5714", "outputIndexes": [ 1383 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1378, 1381, 1383, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5717", "outputIndexes": [ 1384 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1384 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5718", "outputIndexes": [ 1385 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1385 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5719", "outputIndexes": [ 1386 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1386, 1385 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5720", "outputIndexes": [ 1387 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1377, 1387, 1380 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_215_output_0", "outputIndexes": [ 1388 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1388, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_42_output_0", "outputIndexes": [ 1389 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1389, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_455_output_0", "outputIndexes": [ 1390 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1375, 1390, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_57_output_0", "outputIndexes": [ 1391 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1391 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_28_output_0", "outputIndexes": [ 1392 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1389, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_454_output_0", "outputIndexes": [ 1393 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1375, 20, 1393, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_56_output_0", "outputIndexes": [ 1394 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1392, 1394 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_185_output_0", "outputIndexes": [ 1395 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1395, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_113_output_0", "outputIndexes": [ 1396 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1376, 1396 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_70_output_0", "outputIndexes": [ 1397 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1350 ], "outputIndexes": [ 6399 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6399 ], "outputIndexes": [ 6400 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/k_proj/Linear", "inputIndexes": [ 6400 ], "outputIndexes": [ 6401 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 3864566914, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6401 ], "outputIndexes": [ 6402 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6402 ], "outputIndexes": [ 1398 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1362, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_450_output_0", "outputIndexes": [ 1399 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1370, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_451_output_0", "outputIndexes": [ 1400 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1399, 1400, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_183_output_0", "outputIndexes": [ 1401 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1398, 1401 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_128_output_0", "outputIndexes": [ 1402 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1402 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_14/Cast_output_0", "outputIndexes": [ 1403 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_14/Mul_1_output_0", "inputIndexes": [ 1403 ], "outputIndexes": [ 1404 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 3867516056, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1404, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_114_output_0", "outputIndexes": [ 1405 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1404 ], "main_type": "NONE", "name": "/Shape_199_output_0", "outputIndexes": [ 1406 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1406 ], "main_type": "NONE", "name": "Shape5602", "outputIndexes": [ 1407 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1406 ], "main_type": "NONE", "name": "Rank5604", "outputIndexes": [ 1408 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1408, 1408 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5606", "outputIndexes": [ 1409 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1409 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5607", "outputIndexes": [ 1410 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1409, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5609", "outputIndexes": [ 1411 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1411 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5610", "outputIndexes": [ 1412 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1407, 1410, 1412, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5613", "outputIndexes": [ 1413 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1413 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5614", "outputIndexes": [ 1414 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1414 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5615", "outputIndexes": [ 1415 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1415, 1414 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5616", "outputIndexes": [ 1416 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1406, 1416, 1409 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_216_output_0", "outputIndexes": [ 1417 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1417, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_43_output_0", "outputIndexes": [ 1418 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1418, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_457_output_0", "outputIndexes": [ 1419 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1404, 1419, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_59_output_0", "outputIndexes": [ 1420 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1420 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_29_output_0", "outputIndexes": [ 1421 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1418, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_456_output_0", "outputIndexes": [ 1422 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1404, 20, 1422, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_58_output_0", "outputIndexes": [ 1423 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1421, 1423 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_186_output_0", "outputIndexes": [ 1424 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1424, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_115_output_0", "outputIndexes": [ 1425 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1405, 1425 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_71_output_0", "outputIndexes": [ 1426 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1350 ], "outputIndexes": [ 6403 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6403 ], "outputIndexes": [ 6404 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/v_proj/Linear", "inputIndexes": [ 6404 ], "outputIndexes": [ 6405 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 3867517080, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6405 ], "outputIndexes": [ 6406 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6406 ], "outputIndexes": [ 1427 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1362, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_452_output_0", "outputIndexes": [ 1428 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1370, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_453_output_0", "outputIndexes": [ 1429 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1428, 1429, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_184_output_0", "outputIndexes": [ 1430 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1427, 1430 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_129_output_0", "outputIndexes": [ 1431 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1397, 1426, 1431, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_134_output_0", "outputIndexes": [ 1432 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1432 ], "outputIndexes": [ 6407 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6407 ], "outputIndexes": [ 6408 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/o_proj/Linear", "inputIndexes": [ 6408 ], "outputIndexes": [ 6409 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 3870466222, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6409 ], "outputIndexes": [ 6410 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6410 ], "outputIndexes": [ 1433 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1349, 1433 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_73_output_0", "outputIndexes": [ 1434 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_14/Mul_1_output_0", "inputIndexes": [ 1434 ], "outputIndexes": [ 1435 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 3894059204, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1435 ], "outputIndexes": [ 6411 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6411 ], "outputIndexes": [ 6412 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/gate_proj/Linear", "inputIndexes": [ 6412 ], "outputIndexes": [ 6413 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 3894100164, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6413 ], "outputIndexes": [ 6414 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6414 ], "outputIndexes": [ 1436 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1436 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_14/Mul_output_0", "outputIndexes": [ 1437 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1435 ], "outputIndexes": [ 6415 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6415 ], "outputIndexes": [ 6416 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/up_proj/Linear", "inputIndexes": [ 6416 ], "outputIndexes": [ 6417 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 3967828186, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6417 ], "outputIndexes": [ 6418 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6418 ], "outputIndexes": [ 1438 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1437, 1438 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_14/Mul_output_0", "outputIndexes": [ 1439 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1439 ], "outputIndexes": [ 6419 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6419 ], "outputIndexes": [ 6420 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/down_proj/Linear", "inputIndexes": [ 6420 ], "outputIndexes": [ 6421 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 4041556208, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6421 ], "outputIndexes": [ 6422 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.14/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6422 ], "outputIndexes": [ 1440 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1434, 1440 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_74_output_0", "outputIndexes": [ 1441 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1441, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_135_output_0", "outputIndexes": [ 1442 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_15/Mul_1_output_0", "inputIndexes": [ 1442 ], "outputIndexes": [ 1443 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 4115284230, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1443 ], "outputIndexes": [ 6423 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6423 ], "outputIndexes": [ 6424 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/q_proj/Linear", "inputIndexes": [ 6424 ], "outputIndexes": [ 6425 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 4115325190, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6425 ], "outputIndexes": [ 6426 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6426 ], "outputIndexes": [ 1444 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1443 ], "main_type": "NONE", "name": "/Shape_210_output_0", "outputIndexes": [ 1445 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1445 ], "main_type": "NONE", "name": "Shape5945", "outputIndexes": [ 1446 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1445 ], "main_type": "NONE", "name": "Rank5947", "outputIndexes": [ 1447 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1447, 1447 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5949", "outputIndexes": [ 1448 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1448 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5950", "outputIndexes": [ 1449 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1448, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5952", "outputIndexes": [ 1450 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1450 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5953", "outputIndexes": [ 1451 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1446, 1449, 1451, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5956", "outputIndexes": [ 1452 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1452 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5957", "outputIndexes": [ 1453 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1453, 1453 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5959", "outputIndexes": [ 1454 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1445, 1454, 1448 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_228_output_0", "outputIndexes": [ 1455 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1455, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_480_output_0", "outputIndexes": [ 1456 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1448 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5967", "outputIndexes": [ 1457 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1450 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5970", "outputIndexes": [ 1458 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1446, 1457, 1458, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice5973", "outputIndexes": [ 1459 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1459 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze5974", "outputIndexes": [ 1460 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1460 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5975", "outputIndexes": [ 1461 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1461, 1460 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5976", "outputIndexes": [ 1462 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1445, 1462, 1448 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_229_output_0", "outputIndexes": [ 1463 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1463, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_481_output_0", "outputIndexes": [ 1464 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1456, 1464, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_195_output_0", "outputIndexes": [ 1465 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1444, 1465 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_136_output_0", "outputIndexes": [ 1466 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1466 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_15/Cast_output_0", "outputIndexes": [ 1467 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_15/Mul_1_output_0", "inputIndexes": [ 1467 ], "outputIndexes": [ 1468 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 4138918172, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1468, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_120_output_0", "outputIndexes": [ 1469 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1468 ], "main_type": "NONE", "name": "/Shape_212_output_0", "outputIndexes": [ 1470 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1470 ], "main_type": "NONE", "name": "Shape6095", "outputIndexes": [ 1471 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1470 ], "main_type": "NONE", "name": "Rank6097", "outputIndexes": [ 1472 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1472, 1472 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6099", "outputIndexes": [ 1473 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1473 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6100", "outputIndexes": [ 1474 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1473, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6102", "outputIndexes": [ 1475 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1475 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6103", "outputIndexes": [ 1476 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1471, 1474, 1476, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6106", "outputIndexes": [ 1477 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1477 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6107", "outputIndexes": [ 1478 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1478 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6108", "outputIndexes": [ 1479 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1479, 1478 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6109", "outputIndexes": [ 1480 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1470, 1480, 1473 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_230_output_0", "outputIndexes": [ 1481 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1481, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_45_output_0", "outputIndexes": [ 1482 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1482, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_487_output_0", "outputIndexes": [ 1483 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1468, 1483, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_61_output_0", "outputIndexes": [ 1484 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1484 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_30_output_0", "outputIndexes": [ 1485 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1482, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_486_output_0", "outputIndexes": [ 1486 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1468, 20, 1486, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_60_output_0", "outputIndexes": [ 1487 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1485, 1487 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_198_output_0", "outputIndexes": [ 1488 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1488, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_121_output_0", "outputIndexes": [ 1489 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1469, 1489 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_75_output_0", "outputIndexes": [ 1490 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1443 ], "outputIndexes": [ 6427 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6427 ], "outputIndexes": [ 6428 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/k_proj/Linear", "inputIndexes": [ 6428 ], "outputIndexes": [ 6429 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 4138919196, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6429 ], "outputIndexes": [ 6430 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6430 ], "outputIndexes": [ 1491 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1455, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_482_output_0", "outputIndexes": [ 1492 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1463, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_483_output_0", "outputIndexes": [ 1493 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1492, 1493, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_196_output_0", "outputIndexes": [ 1494 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1491, 1494 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_137_output_0", "outputIndexes": [ 1495 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1495 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_15/Cast_output_0", "outputIndexes": [ 1496 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_15/Mul_1_output_0", "inputIndexes": [ 1496 ], "outputIndexes": [ 1497 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 4141868338, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1497, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_122_output_0", "outputIndexes": [ 1498 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1497 ], "main_type": "NONE", "name": "/Shape_213_output_0", "outputIndexes": [ 1499 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1499 ], "main_type": "NONE", "name": "Shape5991", "outputIndexes": [ 1500 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1499 ], "main_type": "NONE", "name": "Rank5993", "outputIndexes": [ 1501 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1501, 1501 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5995", "outputIndexes": [ 1502 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1502 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5996", "outputIndexes": [ 1503 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1502, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp5998", "outputIndexes": [ 1504 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1504 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze5999", "outputIndexes": [ 1505 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1500, 1503, 1505, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6002", "outputIndexes": [ 1506 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1506 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6003", "outputIndexes": [ 1507 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1507 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6004", "outputIndexes": [ 1508 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1508, 1507 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6005", "outputIndexes": [ 1509 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1499, 1509, 1502 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_231_output_0", "outputIndexes": [ 1510 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1510, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_46_output_0", "outputIndexes": [ 1511 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1511, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_489_output_0", "outputIndexes": [ 1512 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1497, 1512, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_63_output_0", "outputIndexes": [ 1513 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1513 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_31_output_0", "outputIndexes": [ 1514 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1511, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_488_output_0", "outputIndexes": [ 1515 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1497, 20, 1515, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_62_output_0", "outputIndexes": [ 1516 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1514, 1516 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_199_output_0", "outputIndexes": [ 1517 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1517, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_123_output_0", "outputIndexes": [ 1518 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1498, 1518 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_76_output_0", "outputIndexes": [ 1519 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1443 ], "outputIndexes": [ 6431 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6431 ], "outputIndexes": [ 6432 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/v_proj/Linear", "inputIndexes": [ 6432 ], "outputIndexes": [ 6433 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 4141869362, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6433 ], "outputIndexes": [ 6434 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6434 ], "outputIndexes": [ 1520 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1455, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_484_output_0", "outputIndexes": [ 1521 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1463, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_485_output_0", "outputIndexes": [ 1522 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1521, 1522, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_197_output_0", "outputIndexes": [ 1523 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1520, 1523 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_138_output_0", "outputIndexes": [ 1524 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1490, 1519, 1524, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_143_output_0", "outputIndexes": [ 1525 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1525 ], "outputIndexes": [ 6435 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6435 ], "outputIndexes": [ 6436 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/o_proj/Linear", "inputIndexes": [ 6436 ], "outputIndexes": [ 6437 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 4144818504, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6437 ], "outputIndexes": [ 6438 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6438 ], "outputIndexes": [ 1526 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1442, 1526 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_78_output_0", "outputIndexes": [ 1527 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_15/Mul_1_output_0", "inputIndexes": [ 1527 ], "outputIndexes": [ 1528 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 4168411486, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1528 ], "outputIndexes": [ 6439 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6439 ], "outputIndexes": [ 6440 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/gate_proj/Linear", "inputIndexes": [ 6440 ], "outputIndexes": [ 6441 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 4168452446, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6441 ], "outputIndexes": [ 6442 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6442 ], "outputIndexes": [ 1529 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1529 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_15/Mul_output_0", "outputIndexes": [ 1530 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1528 ], "outputIndexes": [ 6443 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6443 ], "outputIndexes": [ 6444 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/up_proj/Linear", "inputIndexes": [ 6444 ], "outputIndexes": [ 6445 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 4242180468, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6445 ], "outputIndexes": [ 6446 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6446 ], "outputIndexes": [ 1531 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1530, 1531 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_15/Mul_output_0", "outputIndexes": [ 1532 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1532 ], "outputIndexes": [ 6447 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6447 ], "outputIndexes": [ 6448 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/down_proj/Linear", "inputIndexes": [ 6448 ], "outputIndexes": [ 6449 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 4315908490, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6449 ], "outputIndexes": [ 6450 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.15/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6450 ], "outputIndexes": [ 1533 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1527, 1533 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_79_output_0", "outputIndexes": [ 1534 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1534, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_144_output_0", "outputIndexes": [ 1535 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_16/Mul_1_output_0", "inputIndexes": [ 1535 ], "outputIndexes": [ 1536 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 4389636512, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1536 ], "outputIndexes": [ 6451 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6451 ], "outputIndexes": [ 6452 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/q_proj/Linear", "inputIndexes": [ 6452 ], "outputIndexes": [ 6453 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 4389677472, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6453 ], "outputIndexes": [ 6454 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6454 ], "outputIndexes": [ 1537 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1536 ], "main_type": "NONE", "name": "/Shape_224_output_0", "outputIndexes": [ 1538 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1538 ], "main_type": "NONE", "name": "Shape6334", "outputIndexes": [ 1539 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1538 ], "main_type": "NONE", "name": "Rank6336", "outputIndexes": [ 1540 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1540, 1540 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6338", "outputIndexes": [ 1541 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1541 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6339", "outputIndexes": [ 1542 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1541, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6341", "outputIndexes": [ 1543 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1543 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6342", "outputIndexes": [ 1544 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1539, 1542, 1544, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6345", "outputIndexes": [ 1545 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1545 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6346", "outputIndexes": [ 1546 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1546, 1546 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6348", "outputIndexes": [ 1547 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1538, 1547, 1541 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_243_output_0", "outputIndexes": [ 1548 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1548, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_512_output_0", "outputIndexes": [ 1549 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1541 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6356", "outputIndexes": [ 1550 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1543 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6359", "outputIndexes": [ 1551 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1539, 1550, 1551, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6362", "outputIndexes": [ 1552 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1552 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6363", "outputIndexes": [ 1553 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1553 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6364", "outputIndexes": [ 1554 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1554, 1553 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6365", "outputIndexes": [ 1555 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1538, 1555, 1541 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_244_output_0", "outputIndexes": [ 1556 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1556, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_513_output_0", "outputIndexes": [ 1557 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1549, 1557, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_208_output_0", "outputIndexes": [ 1558 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1537, 1558 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_145_output_0", "outputIndexes": [ 1559 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1559 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_16/Cast_output_0", "outputIndexes": [ 1560 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_16/Mul_1_output_0", "inputIndexes": [ 1560 ], "outputIndexes": [ 1561 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 4413270454, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1561, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_128_output_0", "outputIndexes": [ 1562 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1561 ], "main_type": "NONE", "name": "/Shape_226_output_0", "outputIndexes": [ 1563 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1563 ], "main_type": "NONE", "name": "Shape6484", "outputIndexes": [ 1564 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1563 ], "main_type": "NONE", "name": "Rank6486", "outputIndexes": [ 1565 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1565, 1565 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6488", "outputIndexes": [ 1566 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1566 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6489", "outputIndexes": [ 1567 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1566, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6491", "outputIndexes": [ 1568 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1568 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6492", "outputIndexes": [ 1569 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1564, 1567, 1569, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6495", "outputIndexes": [ 1570 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1570 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6496", "outputIndexes": [ 1571 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1571 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6497", "outputIndexes": [ 1572 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1572, 1571 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6498", "outputIndexes": [ 1573 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1563, 1573, 1566 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_245_output_0", "outputIndexes": [ 1574 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1574, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_48_output_0", "outputIndexes": [ 1575 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1575, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_519_output_0", "outputIndexes": [ 1576 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1561, 1576, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_65_output_0", "outputIndexes": [ 1577 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1577 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_32_output_0", "outputIndexes": [ 1578 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1575, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_518_output_0", "outputIndexes": [ 1579 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1561, 20, 1579, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_64_output_0", "outputIndexes": [ 1580 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1578, 1580 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_211_output_0", "outputIndexes": [ 1581 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1581, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_129_output_0", "outputIndexes": [ 1582 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1562, 1582 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_80_output_0", "outputIndexes": [ 1583 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1536 ], "outputIndexes": [ 6455 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6455 ], "outputIndexes": [ 6456 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/k_proj/Linear", "inputIndexes": [ 6456 ], "outputIndexes": [ 6457 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 4413271478, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6457 ], "outputIndexes": [ 6458 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6458 ], "outputIndexes": [ 1584 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1548, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_514_output_0", "outputIndexes": [ 1585 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1556, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_515_output_0", "outputIndexes": [ 1586 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1585, 1586, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_209_output_0", "outputIndexes": [ 1587 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1584, 1587 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_146_output_0", "outputIndexes": [ 1588 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1588 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_16/Cast_output_0", "outputIndexes": [ 1589 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_16/Mul_1_output_0", "inputIndexes": [ 1589 ], "outputIndexes": [ 1590 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 4416220620, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1590, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_130_output_0", "outputIndexes": [ 1591 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1590 ], "main_type": "NONE", "name": "/Shape_227_output_0", "outputIndexes": [ 1592 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1592 ], "main_type": "NONE", "name": "Shape6380", "outputIndexes": [ 1593 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1592 ], "main_type": "NONE", "name": "Rank6382", "outputIndexes": [ 1594 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1594, 1594 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6384", "outputIndexes": [ 1595 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1595 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6385", "outputIndexes": [ 1596 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1595, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6387", "outputIndexes": [ 1597 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1597 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6388", "outputIndexes": [ 1598 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1593, 1596, 1598, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6391", "outputIndexes": [ 1599 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1599 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6392", "outputIndexes": [ 1600 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1600 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6393", "outputIndexes": [ 1601 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1601, 1600 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6394", "outputIndexes": [ 1602 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1592, 1602, 1595 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_246_output_0", "outputIndexes": [ 1603 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1603, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_49_output_0", "outputIndexes": [ 1604 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1604, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_521_output_0", "outputIndexes": [ 1605 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1590, 1605, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_67_output_0", "outputIndexes": [ 1606 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1606 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_33_output_0", "outputIndexes": [ 1607 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1604, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_520_output_0", "outputIndexes": [ 1608 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1590, 20, 1608, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_66_output_0", "outputIndexes": [ 1609 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1607, 1609 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_212_output_0", "outputIndexes": [ 1610 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1610, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_131_output_0", "outputIndexes": [ 1611 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1591, 1611 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_81_output_0", "outputIndexes": [ 1612 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1536 ], "outputIndexes": [ 6459 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6459 ], "outputIndexes": [ 6460 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/v_proj/Linear", "inputIndexes": [ 6460 ], "outputIndexes": [ 6461 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 4416221644, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6461 ], "outputIndexes": [ 6462 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6462 ], "outputIndexes": [ 1613 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1548, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_516_output_0", "outputIndexes": [ 1614 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1556, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_517_output_0", "outputIndexes": [ 1615 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1614, 1615, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_210_output_0", "outputIndexes": [ 1616 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1613, 1616 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_147_output_0", "outputIndexes": [ 1617 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1583, 1612, 1617, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_152_output_0", "outputIndexes": [ 1618 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1618 ], "outputIndexes": [ 6463 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6463 ], "outputIndexes": [ 6464 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/o_proj/Linear", "inputIndexes": [ 6464 ], "outputIndexes": [ 6465 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 4419170786, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6465 ], "outputIndexes": [ 6466 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6466 ], "outputIndexes": [ 1619 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1535, 1619 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_83_output_0", "outputIndexes": [ 1620 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_16/Mul_1_output_0", "inputIndexes": [ 1620 ], "outputIndexes": [ 1621 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 4442763768, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1621 ], "outputIndexes": [ 6467 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6467 ], "outputIndexes": [ 6468 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/gate_proj/Linear", "inputIndexes": [ 6468 ], "outputIndexes": [ 6469 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 4442804728, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6469 ], "outputIndexes": [ 6470 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6470 ], "outputIndexes": [ 1622 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1622 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_16/Mul_output_0", "outputIndexes": [ 1623 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1621 ], "outputIndexes": [ 6471 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6471 ], "outputIndexes": [ 6472 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/up_proj/Linear", "inputIndexes": [ 6472 ], "outputIndexes": [ 6473 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 4516532750, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6473 ], "outputIndexes": [ 6474 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6474 ], "outputIndexes": [ 1624 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1623, 1624 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_16/Mul_output_0", "outputIndexes": [ 1625 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1625 ], "outputIndexes": [ 6475 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6475 ], "outputIndexes": [ 6476 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/down_proj/Linear", "inputIndexes": [ 6476 ], "outputIndexes": [ 6477 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 4590260772, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6477 ], "outputIndexes": [ 6478 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.16/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6478 ], "outputIndexes": [ 1626 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1620, 1626 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_84_output_0", "outputIndexes": [ 1627 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1627, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_153_output_0", "outputIndexes": [ 1628 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_17/Mul_1_output_0", "inputIndexes": [ 1628 ], "outputIndexes": [ 1629 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 4663988794, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1629 ], "outputIndexes": [ 6479 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6479 ], "outputIndexes": [ 6480 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/q_proj/Linear", "inputIndexes": [ 6480 ], "outputIndexes": [ 6481 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 4664029754, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6481 ], "outputIndexes": [ 6482 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6482 ], "outputIndexes": [ 1630 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1629 ], "main_type": "NONE", "name": "/Shape_238_output_0", "outputIndexes": [ 1631 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1631 ], "main_type": "NONE", "name": "Shape6723", "outputIndexes": [ 1632 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1631 ], "main_type": "NONE", "name": "Rank6725", "outputIndexes": [ 1633 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1633, 1633 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6727", "outputIndexes": [ 1634 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1634 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6728", "outputIndexes": [ 1635 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1634, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6730", "outputIndexes": [ 1636 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1636 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6731", "outputIndexes": [ 1637 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1632, 1635, 1637, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6734", "outputIndexes": [ 1638 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1638 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6735", "outputIndexes": [ 1639 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1639, 1639 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6737", "outputIndexes": [ 1640 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1631, 1640, 1634 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_258_output_0", "outputIndexes": [ 1641 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1641, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_544_output_0", "outputIndexes": [ 1642 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1634 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6745", "outputIndexes": [ 1643 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1636 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6748", "outputIndexes": [ 1644 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1632, 1643, 1644, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6751", "outputIndexes": [ 1645 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1645 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6752", "outputIndexes": [ 1646 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1646 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6753", "outputIndexes": [ 1647 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1647, 1646 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6754", "outputIndexes": [ 1648 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1631, 1648, 1634 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_259_output_0", "outputIndexes": [ 1649 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1649, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_545_output_0", "outputIndexes": [ 1650 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1642, 1650, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_221_output_0", "outputIndexes": [ 1651 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1630, 1651 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_154_output_0", "outputIndexes": [ 1652 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1652 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_17/Cast_output_0", "outputIndexes": [ 1653 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_17/Mul_1_output_0", "inputIndexes": [ 1653 ], "outputIndexes": [ 1654 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 4687622736, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1654, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_136_output_0", "outputIndexes": [ 1655 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1654 ], "main_type": "NONE", "name": "/Shape_240_output_0", "outputIndexes": [ 1656 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1656 ], "main_type": "NONE", "name": "Shape6873", "outputIndexes": [ 1657 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1656 ], "main_type": "NONE", "name": "Rank6875", "outputIndexes": [ 1658 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1658, 1658 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6877", "outputIndexes": [ 1659 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1659 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6878", "outputIndexes": [ 1660 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1659, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6880", "outputIndexes": [ 1661 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1661 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6881", "outputIndexes": [ 1662 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1657, 1660, 1662, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6884", "outputIndexes": [ 1663 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1663 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6885", "outputIndexes": [ 1664 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1664 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6886", "outputIndexes": [ 1665 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1665, 1664 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6887", "outputIndexes": [ 1666 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1656, 1666, 1659 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_260_output_0", "outputIndexes": [ 1667 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1667, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_51_output_0", "outputIndexes": [ 1668 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1668, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_551_output_0", "outputIndexes": [ 1669 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1654, 1669, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_69_output_0", "outputIndexes": [ 1670 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1670 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_34_output_0", "outputIndexes": [ 1671 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1668, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_550_output_0", "outputIndexes": [ 1672 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1654, 20, 1672, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_68_output_0", "outputIndexes": [ 1673 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1671, 1673 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_224_output_0", "outputIndexes": [ 1674 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1674, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_137_output_0", "outputIndexes": [ 1675 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1655, 1675 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_85_output_0", "outputIndexes": [ 1676 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1629 ], "outputIndexes": [ 6483 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6483 ], "outputIndexes": [ 6484 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/k_proj/Linear", "inputIndexes": [ 6484 ], "outputIndexes": [ 6485 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 4687623760, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6485 ], "outputIndexes": [ 6486 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6486 ], "outputIndexes": [ 1677 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1641, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_546_output_0", "outputIndexes": [ 1678 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1649, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_547_output_0", "outputIndexes": [ 1679 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1678, 1679, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_222_output_0", "outputIndexes": [ 1680 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1677, 1680 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_155_output_0", "outputIndexes": [ 1681 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1681 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_17/Cast_output_0", "outputIndexes": [ 1682 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_17/Mul_1_output_0", "inputIndexes": [ 1682 ], "outputIndexes": [ 1683 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 4690572902, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1683, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_138_output_0", "outputIndexes": [ 1684 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1683 ], "main_type": "NONE", "name": "/Shape_241_output_0", "outputIndexes": [ 1685 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1685 ], "main_type": "NONE", "name": "Shape6769", "outputIndexes": [ 1686 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1685 ], "main_type": "NONE", "name": "Rank6771", "outputIndexes": [ 1687 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1687, 1687 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6773", "outputIndexes": [ 1688 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1688 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6774", "outputIndexes": [ 1689 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1688, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6776", "outputIndexes": [ 1690 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1690 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze6777", "outputIndexes": [ 1691 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1686, 1689, 1691, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice6780", "outputIndexes": [ 1692 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1692 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze6781", "outputIndexes": [ 1693 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1693 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6782", "outputIndexes": [ 1694 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1694, 1693 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp6783", "outputIndexes": [ 1695 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1685, 1695, 1688 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_261_output_0", "outputIndexes": [ 1696 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1696, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_52_output_0", "outputIndexes": [ 1697 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1697, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_553_output_0", "outputIndexes": [ 1698 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1683, 1698, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_71_output_0", "outputIndexes": [ 1699 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1699 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_35_output_0", "outputIndexes": [ 1700 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1697, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_552_output_0", "outputIndexes": [ 1701 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1683, 20, 1701, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_70_output_0", "outputIndexes": [ 1702 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1700, 1702 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_225_output_0", "outputIndexes": [ 1703 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1703, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_139_output_0", "outputIndexes": [ 1704 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1684, 1704 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_86_output_0", "outputIndexes": [ 1705 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1629 ], "outputIndexes": [ 6487 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6487 ], "outputIndexes": [ 6488 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/v_proj/Linear", "inputIndexes": [ 6488 ], "outputIndexes": [ 6489 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 4690573926, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6489 ], "outputIndexes": [ 6490 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6490 ], "outputIndexes": [ 1706 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1641, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_548_output_0", "outputIndexes": [ 1707 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1649, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_549_output_0", "outputIndexes": [ 1708 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1707, 1708, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_223_output_0", "outputIndexes": [ 1709 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1706, 1709 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_156_output_0", "outputIndexes": [ 1710 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1676, 1705, 1710, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_161_output_0", "outputIndexes": [ 1711 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1711 ], "outputIndexes": [ 6491 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6491 ], "outputIndexes": [ 6492 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/o_proj/Linear", "inputIndexes": [ 6492 ], "outputIndexes": [ 6493 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 4693523068, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6493 ], "outputIndexes": [ 6494 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6494 ], "outputIndexes": [ 1712 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1628, 1712 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_88_output_0", "outputIndexes": [ 1713 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_17/Mul_1_output_0", "inputIndexes": [ 1713 ], "outputIndexes": [ 1714 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 4717116050, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1714 ], "outputIndexes": [ 6495 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6495 ], "outputIndexes": [ 6496 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/gate_proj/Linear", "inputIndexes": [ 6496 ], "outputIndexes": [ 6497 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 4717157010, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6497 ], "outputIndexes": [ 6498 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6498 ], "outputIndexes": [ 1715 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1715 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_17/Mul_output_0", "outputIndexes": [ 1716 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1714 ], "outputIndexes": [ 6499 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6499 ], "outputIndexes": [ 6500 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/up_proj/Linear", "inputIndexes": [ 6500 ], "outputIndexes": [ 6501 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 4790885032, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6501 ], "outputIndexes": [ 6502 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6502 ], "outputIndexes": [ 1717 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1716, 1717 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_17/Mul_output_0", "outputIndexes": [ 1718 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1718 ], "outputIndexes": [ 6503 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6503 ], "outputIndexes": [ 6504 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/down_proj/Linear", "inputIndexes": [ 6504 ], "outputIndexes": [ 6505 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 4864613054, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6505 ], "outputIndexes": [ 6506 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.17/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6506 ], "outputIndexes": [ 1719 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1713, 1719 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_89_output_0", "outputIndexes": [ 1720 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1720, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_162_output_0", "outputIndexes": [ 1721 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_18/Mul_1_output_0", "inputIndexes": [ 1721 ], "outputIndexes": [ 1722 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 4938341076, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1722 ], "outputIndexes": [ 6507 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6507 ], "outputIndexes": [ 6508 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/q_proj/Linear", "inputIndexes": [ 6508 ], "outputIndexes": [ 6509 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 4938382036, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6509 ], "outputIndexes": [ 6510 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6510 ], "outputIndexes": [ 1723 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1722 ], "main_type": "NONE", "name": "/Shape_252_output_0", "outputIndexes": [ 1724 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1724 ], "main_type": "NONE", "name": "Shape7112", "outputIndexes": [ 1725 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1724 ], "main_type": "NONE", "name": "Rank7114", "outputIndexes": [ 1726 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1726, 1726 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7116", "outputIndexes": [ 1727 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1727 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7117", "outputIndexes": [ 1728 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1727, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7119", "outputIndexes": [ 1729 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1729 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7120", "outputIndexes": [ 1730 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1725, 1728, 1730, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7123", "outputIndexes": [ 1731 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1731 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7124", "outputIndexes": [ 1732 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1732, 1732 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7126", "outputIndexes": [ 1733 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1724, 1733, 1727 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_273_output_0", "outputIndexes": [ 1734 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1734, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_576_output_0", "outputIndexes": [ 1735 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1727 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7134", "outputIndexes": [ 1736 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1729 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7137", "outputIndexes": [ 1737 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1725, 1736, 1737, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7140", "outputIndexes": [ 1738 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1738 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7141", "outputIndexes": [ 1739 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1739 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7142", "outputIndexes": [ 1740 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1740, 1739 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7143", "outputIndexes": [ 1741 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1724, 1741, 1727 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_274_output_0", "outputIndexes": [ 1742 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1742, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_577_output_0", "outputIndexes": [ 1743 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1735, 1743, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_234_output_0", "outputIndexes": [ 1744 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1723, 1744 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_163_output_0", "outputIndexes": [ 1745 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1745 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_18/Cast_output_0", "outputIndexes": [ 1746 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_18/Mul_1_output_0", "inputIndexes": [ 1746 ], "outputIndexes": [ 1747 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 4961975018, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1747, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_144_output_0", "outputIndexes": [ 1748 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1747 ], "main_type": "NONE", "name": "/Shape_254_output_0", "outputIndexes": [ 1749 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1749 ], "main_type": "NONE", "name": "Shape7262", "outputIndexes": [ 1750 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1749 ], "main_type": "NONE", "name": "Rank7264", "outputIndexes": [ 1751 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1751, 1751 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7266", "outputIndexes": [ 1752 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1752 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7267", "outputIndexes": [ 1753 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1752, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7269", "outputIndexes": [ 1754 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1754 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7270", "outputIndexes": [ 1755 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1750, 1753, 1755, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7273", "outputIndexes": [ 1756 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1756 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7274", "outputIndexes": [ 1757 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1757 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7275", "outputIndexes": [ 1758 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1758, 1757 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7276", "outputIndexes": [ 1759 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1749, 1759, 1752 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_275_output_0", "outputIndexes": [ 1760 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1760, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_54_output_0", "outputIndexes": [ 1761 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1761, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_583_output_0", "outputIndexes": [ 1762 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1747, 1762, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_73_output_0", "outputIndexes": [ 1763 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1763 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_36_output_0", "outputIndexes": [ 1764 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1761, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_582_output_0", "outputIndexes": [ 1765 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1747, 20, 1765, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_72_output_0", "outputIndexes": [ 1766 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1764, 1766 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_237_output_0", "outputIndexes": [ 1767 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1767, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_145_output_0", "outputIndexes": [ 1768 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1748, 1768 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_90_output_0", "outputIndexes": [ 1769 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1722 ], "outputIndexes": [ 6511 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6511 ], "outputIndexes": [ 6512 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/k_proj/Linear", "inputIndexes": [ 6512 ], "outputIndexes": [ 6513 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 4961976042, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6513 ], "outputIndexes": [ 6514 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6514 ], "outputIndexes": [ 1770 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1734, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_578_output_0", "outputIndexes": [ 1771 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1742, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_579_output_0", "outputIndexes": [ 1772 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1771, 1772, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_235_output_0", "outputIndexes": [ 1773 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1770, 1773 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_164_output_0", "outputIndexes": [ 1774 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1774 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_18/Cast_output_0", "outputIndexes": [ 1775 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_18/Mul_1_output_0", "inputIndexes": [ 1775 ], "outputIndexes": [ 1776 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 4964925184, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1776, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_146_output_0", "outputIndexes": [ 1777 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1776 ], "main_type": "NONE", "name": "/Shape_255_output_0", "outputIndexes": [ 1778 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1778 ], "main_type": "NONE", "name": "Shape7158", "outputIndexes": [ 1779 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1778 ], "main_type": "NONE", "name": "Rank7160", "outputIndexes": [ 1780 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1780, 1780 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7162", "outputIndexes": [ 1781 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1781 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7163", "outputIndexes": [ 1782 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1781, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7165", "outputIndexes": [ 1783 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1783 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7166", "outputIndexes": [ 1784 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1779, 1782, 1784, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7169", "outputIndexes": [ 1785 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1785 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7170", "outputIndexes": [ 1786 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1786 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7171", "outputIndexes": [ 1787 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1787, 1786 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7172", "outputIndexes": [ 1788 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1778, 1788, 1781 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_276_output_0", "outputIndexes": [ 1789 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1789, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_55_output_0", "outputIndexes": [ 1790 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1790, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_585_output_0", "outputIndexes": [ 1791 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1776, 1791, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_75_output_0", "outputIndexes": [ 1792 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1792 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_37_output_0", "outputIndexes": [ 1793 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1790, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_584_output_0", "outputIndexes": [ 1794 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1776, 20, 1794, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_74_output_0", "outputIndexes": [ 1795 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1793, 1795 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_238_output_0", "outputIndexes": [ 1796 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1796, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_147_output_0", "outputIndexes": [ 1797 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1777, 1797 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_91_output_0", "outputIndexes": [ 1798 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1722 ], "outputIndexes": [ 6515 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6515 ], "outputIndexes": [ 6516 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/v_proj/Linear", "inputIndexes": [ 6516 ], "outputIndexes": [ 6517 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 4964926208, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6517 ], "outputIndexes": [ 6518 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6518 ], "outputIndexes": [ 1799 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1734, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_580_output_0", "outputIndexes": [ 1800 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1742, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_581_output_0", "outputIndexes": [ 1801 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1800, 1801, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_236_output_0", "outputIndexes": [ 1802 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1799, 1802 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_165_output_0", "outputIndexes": [ 1803 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1769, 1798, 1803, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_170_output_0", "outputIndexes": [ 1804 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1804 ], "outputIndexes": [ 6519 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6519 ], "outputIndexes": [ 6520 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/o_proj/Linear", "inputIndexes": [ 6520 ], "outputIndexes": [ 6521 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 4967875350, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6521 ], "outputIndexes": [ 6522 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6522 ], "outputIndexes": [ 1805 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1721, 1805 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_93_output_0", "outputIndexes": [ 1806 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_18/Mul_1_output_0", "inputIndexes": [ 1806 ], "outputIndexes": [ 1807 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 4991468332, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1807 ], "outputIndexes": [ 6523 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6523 ], "outputIndexes": [ 6524 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/gate_proj/Linear", "inputIndexes": [ 6524 ], "outputIndexes": [ 6525 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 4991509292, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6525 ], "outputIndexes": [ 6526 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6526 ], "outputIndexes": [ 1808 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1808 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_18/Mul_output_0", "outputIndexes": [ 1809 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1807 ], "outputIndexes": [ 6527 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6527 ], "outputIndexes": [ 6528 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/up_proj/Linear", "inputIndexes": [ 6528 ], "outputIndexes": [ 6529 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 5065237314, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6529 ], "outputIndexes": [ 6530 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6530 ], "outputIndexes": [ 1810 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1809, 1810 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_18/Mul_output_0", "outputIndexes": [ 1811 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1811 ], "outputIndexes": [ 6531 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6531 ], "outputIndexes": [ 6532 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/down_proj/Linear", "inputIndexes": [ 6532 ], "outputIndexes": [ 6533 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 5138965336, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6533 ], "outputIndexes": [ 6534 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.18/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6534 ], "outputIndexes": [ 1812 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1806, 1812 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_94_output_0", "outputIndexes": [ 1813 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1813, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_171_output_0", "outputIndexes": [ 1814 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_19/Mul_1_output_0", "inputIndexes": [ 1814 ], "outputIndexes": [ 1815 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 5212693358, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1815 ], "outputIndexes": [ 6535 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6535 ], "outputIndexes": [ 6536 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/q_proj/Linear", "inputIndexes": [ 6536 ], "outputIndexes": [ 6537 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 5212734318, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6537 ], "outputIndexes": [ 6538 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6538 ], "outputIndexes": [ 1816 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1815 ], "main_type": "NONE", "name": "/Shape_266_output_0", "outputIndexes": [ 1817 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1817 ], "main_type": "NONE", "name": "Shape7501", "outputIndexes": [ 1818 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1817 ], "main_type": "NONE", "name": "Rank7503", "outputIndexes": [ 1819 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1819, 1819 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7505", "outputIndexes": [ 1820 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1820 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7506", "outputIndexes": [ 1821 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1820, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7508", "outputIndexes": [ 1822 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1822 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7509", "outputIndexes": [ 1823 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1818, 1821, 1823, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7512", "outputIndexes": [ 1824 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1824 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7513", "outputIndexes": [ 1825 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1825, 1825 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7515", "outputIndexes": [ 1826 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1817, 1826, 1820 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_288_output_0", "outputIndexes": [ 1827 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1827, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_608_output_0", "outputIndexes": [ 1828 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1820 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7523", "outputIndexes": [ 1829 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1822 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7526", "outputIndexes": [ 1830 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1818, 1829, 1830, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7529", "outputIndexes": [ 1831 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1831 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7530", "outputIndexes": [ 1832 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1832 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7531", "outputIndexes": [ 1833 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1833, 1832 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7532", "outputIndexes": [ 1834 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1817, 1834, 1820 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_289_output_0", "outputIndexes": [ 1835 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1835, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_609_output_0", "outputIndexes": [ 1836 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1828, 1836, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_247_output_0", "outputIndexes": [ 1837 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1816, 1837 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_172_output_0", "outputIndexes": [ 1838 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1838 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_19/Cast_output_0", "outputIndexes": [ 1839 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_19/Mul_1_output_0", "inputIndexes": [ 1839 ], "outputIndexes": [ 1840 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 5236327300, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1840, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_152_output_0", "outputIndexes": [ 1841 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1840 ], "main_type": "NONE", "name": "/Shape_268_output_0", "outputIndexes": [ 1842 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1842 ], "main_type": "NONE", "name": "Shape7651", "outputIndexes": [ 1843 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1842 ], "main_type": "NONE", "name": "Rank7653", "outputIndexes": [ 1844 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1844, 1844 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7655", "outputIndexes": [ 1845 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1845 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7656", "outputIndexes": [ 1846 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1845, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7658", "outputIndexes": [ 1847 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1847 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7659", "outputIndexes": [ 1848 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1843, 1846, 1848, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7662", "outputIndexes": [ 1849 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1849 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7663", "outputIndexes": [ 1850 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1850 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7664", "outputIndexes": [ 1851 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1851, 1850 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7665", "outputIndexes": [ 1852 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1842, 1852, 1845 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_290_output_0", "outputIndexes": [ 1853 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1853, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_57_output_0", "outputIndexes": [ 1854 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1854, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_615_output_0", "outputIndexes": [ 1855 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1840, 1855, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_77_output_0", "outputIndexes": [ 1856 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1856 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_38_output_0", "outputIndexes": [ 1857 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1854, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_614_output_0", "outputIndexes": [ 1858 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1840, 20, 1858, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_76_output_0", "outputIndexes": [ 1859 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1857, 1859 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_250_output_0", "outputIndexes": [ 1860 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1860, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_153_output_0", "outputIndexes": [ 1861 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1841, 1861 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_95_output_0", "outputIndexes": [ 1862 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1815 ], "outputIndexes": [ 6539 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6539 ], "outputIndexes": [ 6540 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/k_proj/Linear", "inputIndexes": [ 6540 ], "outputIndexes": [ 6541 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 5236328324, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6541 ], "outputIndexes": [ 6542 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6542 ], "outputIndexes": [ 1863 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1827, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_610_output_0", "outputIndexes": [ 1864 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1835, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_611_output_0", "outputIndexes": [ 1865 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1864, 1865, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_248_output_0", "outputIndexes": [ 1866 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1863, 1866 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_173_output_0", "outputIndexes": [ 1867 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1867 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_19/Cast_output_0", "outputIndexes": [ 1868 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_19/Mul_1_output_0", "inputIndexes": [ 1868 ], "outputIndexes": [ 1869 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 5239277466, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1869, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_154_output_0", "outputIndexes": [ 1870 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1869 ], "main_type": "NONE", "name": "/Shape_269_output_0", "outputIndexes": [ 1871 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1871 ], "main_type": "NONE", "name": "Shape7547", "outputIndexes": [ 1872 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1871 ], "main_type": "NONE", "name": "Rank7549", "outputIndexes": [ 1873 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1873, 1873 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7551", "outputIndexes": [ 1874 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1874 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7552", "outputIndexes": [ 1875 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1874, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7554", "outputIndexes": [ 1876 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1876 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7555", "outputIndexes": [ 1877 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1872, 1875, 1877, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7558", "outputIndexes": [ 1878 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1878 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7559", "outputIndexes": [ 1879 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1879 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7560", "outputIndexes": [ 1880 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1880, 1879 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7561", "outputIndexes": [ 1881 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1871, 1881, 1874 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_291_output_0", "outputIndexes": [ 1882 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1882, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_58_output_0", "outputIndexes": [ 1883 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1883, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_617_output_0", "outputIndexes": [ 1884 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1869, 1884, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_79_output_0", "outputIndexes": [ 1885 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1885 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_39_output_0", "outputIndexes": [ 1886 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1883, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_616_output_0", "outputIndexes": [ 1887 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1869, 20, 1887, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_78_output_0", "outputIndexes": [ 1888 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1886, 1888 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_251_output_0", "outputIndexes": [ 1889 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1889, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_155_output_0", "outputIndexes": [ 1890 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1870, 1890 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_96_output_0", "outputIndexes": [ 1891 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1815 ], "outputIndexes": [ 6543 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6543 ], "outputIndexes": [ 6544 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/v_proj/Linear", "inputIndexes": [ 6544 ], "outputIndexes": [ 6545 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 5239278490, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6545 ], "outputIndexes": [ 6546 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6546 ], "outputIndexes": [ 1892 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1827, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_612_output_0", "outputIndexes": [ 1893 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1835, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_613_output_0", "outputIndexes": [ 1894 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1893, 1894, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_249_output_0", "outputIndexes": [ 1895 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1892, 1895 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_174_output_0", "outputIndexes": [ 1896 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1862, 1891, 1896, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_179_output_0", "outputIndexes": [ 1897 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1897 ], "outputIndexes": [ 6547 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6547 ], "outputIndexes": [ 6548 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/o_proj/Linear", "inputIndexes": [ 6548 ], "outputIndexes": [ 6549 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 5242227632, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6549 ], "outputIndexes": [ 6550 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6550 ], "outputIndexes": [ 1898 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1814, 1898 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_98_output_0", "outputIndexes": [ 1899 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_19/Mul_1_output_0", "inputIndexes": [ 1899 ], "outputIndexes": [ 1900 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 5265820614, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1900 ], "outputIndexes": [ 6551 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6551 ], "outputIndexes": [ 6552 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/gate_proj/Linear", "inputIndexes": [ 6552 ], "outputIndexes": [ 6553 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 5265861574, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6553 ], "outputIndexes": [ 6554 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6554 ], "outputIndexes": [ 1901 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1901 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_19/Mul_output_0", "outputIndexes": [ 1902 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1900 ], "outputIndexes": [ 6555 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6555 ], "outputIndexes": [ 6556 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/up_proj/Linear", "inputIndexes": [ 6556 ], "outputIndexes": [ 6557 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 5339589596, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6557 ], "outputIndexes": [ 6558 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6558 ], "outputIndexes": [ 1903 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1902, 1903 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_19/Mul_output_0", "outputIndexes": [ 1904 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1904 ], "outputIndexes": [ 6559 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6559 ], "outputIndexes": [ 6560 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/down_proj/Linear", "inputIndexes": [ 6560 ], "outputIndexes": [ 6561 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 5413317618, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6561 ], "outputIndexes": [ 6562 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.19/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6562 ], "outputIndexes": [ 1905 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1899, 1905 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_99_output_0", "outputIndexes": [ 1906 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1906, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_180_output_0", "outputIndexes": [ 1907 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_20/Mul_1_output_0", "inputIndexes": [ 1907 ], "outputIndexes": [ 1908 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 5487045640, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1908 ], "outputIndexes": [ 6563 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6563 ], "outputIndexes": [ 6564 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/q_proj/Linear", "inputIndexes": [ 6564 ], "outputIndexes": [ 6565 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 5487086600, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6565 ], "outputIndexes": [ 6566 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6566 ], "outputIndexes": [ 1909 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1908 ], "main_type": "NONE", "name": "/Shape_280_output_0", "outputIndexes": [ 1910 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1910 ], "main_type": "NONE", "name": "Shape7890", "outputIndexes": [ 1911 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1910 ], "main_type": "NONE", "name": "Rank7892", "outputIndexes": [ 1912 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1912, 1912 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7894", "outputIndexes": [ 1913 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1913 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7895", "outputIndexes": [ 1914 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1913, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7897", "outputIndexes": [ 1915 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1915 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7898", "outputIndexes": [ 1916 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1911, 1914, 1916, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7901", "outputIndexes": [ 1917 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1917 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7902", "outputIndexes": [ 1918 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1918, 1918 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7904", "outputIndexes": [ 1919 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1910, 1919, 1913 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_303_output_0", "outputIndexes": [ 1920 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1920, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_640_output_0", "outputIndexes": [ 1921 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1913 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7912", "outputIndexes": [ 1922 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1915 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7915", "outputIndexes": [ 1923 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1911, 1922, 1923, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7918", "outputIndexes": [ 1924 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1924 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7919", "outputIndexes": [ 1925 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 1925 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7920", "outputIndexes": [ 1926 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1926, 1925 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7921", "outputIndexes": [ 1927 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1910, 1927, 1913 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_304_output_0", "outputIndexes": [ 1928 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1928, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_641_output_0", "outputIndexes": [ 1929 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1921, 1929, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_260_output_0", "outputIndexes": [ 1930 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1909, 1930 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_181_output_0", "outputIndexes": [ 1931 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1931 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_20/Cast_output_0", "outputIndexes": [ 1932 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_20/Mul_1_output_0", "inputIndexes": [ 1932 ], "outputIndexes": [ 1933 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 5510679582, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1933, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_160_output_0", "outputIndexes": [ 1934 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1933 ], "main_type": "NONE", "name": "/Shape_282_output_0", "outputIndexes": [ 1935 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1935 ], "main_type": "NONE", "name": "Shape8040", "outputIndexes": [ 1936 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1935 ], "main_type": "NONE", "name": "Rank8042", "outputIndexes": [ 1937 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1937, 1937 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8044", "outputIndexes": [ 1938 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1938 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8045", "outputIndexes": [ 1939 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1938, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8047", "outputIndexes": [ 1940 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1940 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8048", "outputIndexes": [ 1941 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1936, 1939, 1941, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8051", "outputIndexes": [ 1942 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1942 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8052", "outputIndexes": [ 1943 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1943 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8053", "outputIndexes": [ 1944 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1944, 1943 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8054", "outputIndexes": [ 1945 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1935, 1945, 1938 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_305_output_0", "outputIndexes": [ 1946 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1946, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_60_output_0", "outputIndexes": [ 1947 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1947, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_647_output_0", "outputIndexes": [ 1948 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1933, 1948, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_81_output_0", "outputIndexes": [ 1949 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1949 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_40_output_0", "outputIndexes": [ 1950 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1947, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_646_output_0", "outputIndexes": [ 1951 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1933, 20, 1951, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_80_output_0", "outputIndexes": [ 1952 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1950, 1952 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_263_output_0", "outputIndexes": [ 1953 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1953, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_161_output_0", "outputIndexes": [ 1954 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1934, 1954 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_100_output_0", "outputIndexes": [ 1955 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1908 ], "outputIndexes": [ 6567 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6567 ], "outputIndexes": [ 6568 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/k_proj/Linear", "inputIndexes": [ 6568 ], "outputIndexes": [ 6569 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 5510680606, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6569 ], "outputIndexes": [ 6570 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6570 ], "outputIndexes": [ 1956 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1920, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_642_output_0", "outputIndexes": [ 1957 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1928, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_643_output_0", "outputIndexes": [ 1958 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1957, 1958, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_261_output_0", "outputIndexes": [ 1959 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1956, 1959 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_182_output_0", "outputIndexes": [ 1960 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1960 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_20/Cast_output_0", "outputIndexes": [ 1961 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_20/Mul_1_output_0", "inputIndexes": [ 1961 ], "outputIndexes": [ 1962 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 5513629748, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1962, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_162_output_0", "outputIndexes": [ 1963 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1962 ], "main_type": "NONE", "name": "/Shape_283_output_0", "outputIndexes": [ 1964 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1964 ], "main_type": "NONE", "name": "Shape7936", "outputIndexes": [ 1965 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 1964 ], "main_type": "NONE", "name": "Rank7938", "outputIndexes": [ 1966 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1966, 1966 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7940", "outputIndexes": [ 1967 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1967 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7941", "outputIndexes": [ 1968 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1967, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7943", "outputIndexes": [ 1969 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1969 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze7944", "outputIndexes": [ 1970 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1965, 1968, 1970, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice7947", "outputIndexes": [ 1971 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1971 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze7948", "outputIndexes": [ 1972 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 1972 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7949", "outputIndexes": [ 1973 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1973, 1972 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp7950", "outputIndexes": [ 1974 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1964, 1974, 1967 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_306_output_0", "outputIndexes": [ 1975 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1975, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_61_output_0", "outputIndexes": [ 1976 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1976, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_649_output_0", "outputIndexes": [ 1977 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1962, 1977, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_83_output_0", "outputIndexes": [ 1978 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1978 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_41_output_0", "outputIndexes": [ 1979 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1976, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_648_output_0", "outputIndexes": [ 1980 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1962, 20, 1980, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_82_output_0", "outputIndexes": [ 1981 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1979, 1981 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_264_output_0", "outputIndexes": [ 1982 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1982, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_163_output_0", "outputIndexes": [ 1983 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1963, 1983 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_101_output_0", "outputIndexes": [ 1984 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1908 ], "outputIndexes": [ 6571 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6571 ], "outputIndexes": [ 6572 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/v_proj/Linear", "inputIndexes": [ 6572 ], "outputIndexes": [ 6573 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 5513630772, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6573 ], "outputIndexes": [ 6574 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6574 ], "outputIndexes": [ 1985 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1920, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_644_output_0", "outputIndexes": [ 1986 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1928, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_645_output_0", "outputIndexes": [ 1987 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1986, 1987, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_262_output_0", "outputIndexes": [ 1988 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1985, 1988 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_183_output_0", "outputIndexes": [ 1989 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1955, 1984, 1989, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_188_output_0", "outputIndexes": [ 1990 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1990 ], "outputIndexes": [ 6575 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6575 ], "outputIndexes": [ 6576 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/o_proj/Linear", "inputIndexes": [ 6576 ], "outputIndexes": [ 6577 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 5516579914, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6577 ], "outputIndexes": [ 6578 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6578 ], "outputIndexes": [ 1991 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1907, 1991 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_103_output_0", "outputIndexes": [ 1992 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_20/Mul_1_output_0", "inputIndexes": [ 1992 ], "outputIndexes": [ 1993 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 5540172896, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1993 ], "outputIndexes": [ 6579 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6579 ], "outputIndexes": [ 6580 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/gate_proj/Linear", "inputIndexes": [ 6580 ], "outputIndexes": [ 6581 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 5540213856, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6581 ], "outputIndexes": [ 6582 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6582 ], "outputIndexes": [ 1994 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1994 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_20/Mul_output_0", "outputIndexes": [ 1995 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1993 ], "outputIndexes": [ 6583 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6583 ], "outputIndexes": [ 6584 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/up_proj/Linear", "inputIndexes": [ 6584 ], "outputIndexes": [ 6585 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 5613941878, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6585 ], "outputIndexes": [ 6586 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6586 ], "outputIndexes": [ 1996 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1995, 1996 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_20/Mul_output_0", "outputIndexes": [ 1997 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 1997 ], "outputIndexes": [ 6587 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6587 ], "outputIndexes": [ 6588 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/down_proj/Linear", "inputIndexes": [ 6588 ], "outputIndexes": [ 6589 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 5687669900, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6589 ], "outputIndexes": [ 6590 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.20/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6590 ], "outputIndexes": [ 1998 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1992, 1998 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_104_output_0", "outputIndexes": [ 1999 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 1999, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_189_output_0", "outputIndexes": [ 2000 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_21/Mul_1_output_0", "inputIndexes": [ 2000 ], "outputIndexes": [ 2001 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 5761397922, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2001 ], "outputIndexes": [ 6591 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6591 ], "outputIndexes": [ 6592 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/q_proj/Linear", "inputIndexes": [ 6592 ], "outputIndexes": [ 6593 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 5761438882, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6593 ], "outputIndexes": [ 6594 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6594 ], "outputIndexes": [ 2002 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2001 ], "main_type": "NONE", "name": "/Shape_294_output_0", "outputIndexes": [ 2003 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2003 ], "main_type": "NONE", "name": "Shape8279", "outputIndexes": [ 2004 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2003 ], "main_type": "NONE", "name": "Rank8281", "outputIndexes": [ 2005 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2005, 2005 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8283", "outputIndexes": [ 2006 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2006 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8284", "outputIndexes": [ 2007 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2006, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8286", "outputIndexes": [ 2008 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2008 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8287", "outputIndexes": [ 2009 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2004, 2007, 2009, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8290", "outputIndexes": [ 2010 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2010 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8291", "outputIndexes": [ 2011 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2011, 2011 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8293", "outputIndexes": [ 2012 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2003, 2012, 2006 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_318_output_0", "outputIndexes": [ 2013 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2013, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_672_output_0", "outputIndexes": [ 2014 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2006 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8301", "outputIndexes": [ 2015 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2008 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8304", "outputIndexes": [ 2016 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2004, 2015, 2016, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8307", "outputIndexes": [ 2017 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2017 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8308", "outputIndexes": [ 2018 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2018 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8309", "outputIndexes": [ 2019 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2019, 2018 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8310", "outputIndexes": [ 2020 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2003, 2020, 2006 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_319_output_0", "outputIndexes": [ 2021 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2021, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_673_output_0", "outputIndexes": [ 2022 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2014, 2022, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_273_output_0", "outputIndexes": [ 2023 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2002, 2023 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_190_output_0", "outputIndexes": [ 2024 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2024 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_21/Cast_output_0", "outputIndexes": [ 2025 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_21/Mul_1_output_0", "inputIndexes": [ 2025 ], "outputIndexes": [ 2026 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 5785031864, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2026, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_168_output_0", "outputIndexes": [ 2027 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2026 ], "main_type": "NONE", "name": "/Shape_296_output_0", "outputIndexes": [ 2028 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2028 ], "main_type": "NONE", "name": "Shape8429", "outputIndexes": [ 2029 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2028 ], "main_type": "NONE", "name": "Rank8431", "outputIndexes": [ 2030 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2030, 2030 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8433", "outputIndexes": [ 2031 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2031 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8434", "outputIndexes": [ 2032 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2031, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8436", "outputIndexes": [ 2033 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2033 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8437", "outputIndexes": [ 2034 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2029, 2032, 2034, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8440", "outputIndexes": [ 2035 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2035 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8441", "outputIndexes": [ 2036 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2036 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8442", "outputIndexes": [ 2037 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2037, 2036 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8443", "outputIndexes": [ 2038 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2028, 2038, 2031 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_320_output_0", "outputIndexes": [ 2039 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2039, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_63_output_0", "outputIndexes": [ 2040 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2040, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_679_output_0", "outputIndexes": [ 2041 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2026, 2041, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_85_output_0", "outputIndexes": [ 2042 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2042 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_42_output_0", "outputIndexes": [ 2043 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2040, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_678_output_0", "outputIndexes": [ 2044 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2026, 20, 2044, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_84_output_0", "outputIndexes": [ 2045 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2043, 2045 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_276_output_0", "outputIndexes": [ 2046 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2046, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_169_output_0", "outputIndexes": [ 2047 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2027, 2047 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_105_output_0", "outputIndexes": [ 2048 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2001 ], "outputIndexes": [ 6595 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6595 ], "outputIndexes": [ 6596 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/k_proj/Linear", "inputIndexes": [ 6596 ], "outputIndexes": [ 6597 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 5785032888, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6597 ], "outputIndexes": [ 6598 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6598 ], "outputIndexes": [ 2049 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2013, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_674_output_0", "outputIndexes": [ 2050 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2021, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_675_output_0", "outputIndexes": [ 2051 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2050, 2051, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_274_output_0", "outputIndexes": [ 2052 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2049, 2052 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_191_output_0", "outputIndexes": [ 2053 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2053 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_21/Cast_output_0", "outputIndexes": [ 2054 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_21/Mul_1_output_0", "inputIndexes": [ 2054 ], "outputIndexes": [ 2055 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 5787982030, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2055, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_170_output_0", "outputIndexes": [ 2056 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2055 ], "main_type": "NONE", "name": "/Shape_297_output_0", "outputIndexes": [ 2057 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2057 ], "main_type": "NONE", "name": "Shape8325", "outputIndexes": [ 2058 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2057 ], "main_type": "NONE", "name": "Rank8327", "outputIndexes": [ 2059 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2059, 2059 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8329", "outputIndexes": [ 2060 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2060 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8330", "outputIndexes": [ 2061 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2060, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8332", "outputIndexes": [ 2062 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2062 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8333", "outputIndexes": [ 2063 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2058, 2061, 2063, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8336", "outputIndexes": [ 2064 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2064 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8337", "outputIndexes": [ 2065 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2065 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8338", "outputIndexes": [ 2066 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2066, 2065 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8339", "outputIndexes": [ 2067 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2057, 2067, 2060 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_321_output_0", "outputIndexes": [ 2068 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2068, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_64_output_0", "outputIndexes": [ 2069 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2069, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_681_output_0", "outputIndexes": [ 2070 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2055, 2070, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_87_output_0", "outputIndexes": [ 2071 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2071 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_43_output_0", "outputIndexes": [ 2072 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2069, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_680_output_0", "outputIndexes": [ 2073 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2055, 20, 2073, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_86_output_0", "outputIndexes": [ 2074 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2072, 2074 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_277_output_0", "outputIndexes": [ 2075 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2075, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_171_output_0", "outputIndexes": [ 2076 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2056, 2076 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_106_output_0", "outputIndexes": [ 2077 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2001 ], "outputIndexes": [ 6599 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6599 ], "outputIndexes": [ 6600 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/v_proj/Linear", "inputIndexes": [ 6600 ], "outputIndexes": [ 6601 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 5787983054, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6601 ], "outputIndexes": [ 6602 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6602 ], "outputIndexes": [ 2078 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2013, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_676_output_0", "outputIndexes": [ 2079 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2021, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_677_output_0", "outputIndexes": [ 2080 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2079, 2080, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_275_output_0", "outputIndexes": [ 2081 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2078, 2081 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_192_output_0", "outputIndexes": [ 2082 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2048, 2077, 2082, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_197_output_0", "outputIndexes": [ 2083 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2083 ], "outputIndexes": [ 6603 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6603 ], "outputIndexes": [ 6604 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/o_proj/Linear", "inputIndexes": [ 6604 ], "outputIndexes": [ 6605 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 5790932196, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6605 ], "outputIndexes": [ 6606 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6606 ], "outputIndexes": [ 2084 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2000, 2084 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_108_output_0", "outputIndexes": [ 2085 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_21/Mul_1_output_0", "inputIndexes": [ 2085 ], "outputIndexes": [ 2086 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 5814525178, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2086 ], "outputIndexes": [ 6607 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6607 ], "outputIndexes": [ 6608 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/gate_proj/Linear", "inputIndexes": [ 6608 ], "outputIndexes": [ 6609 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 5814566138, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6609 ], "outputIndexes": [ 6610 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6610 ], "outputIndexes": [ 2087 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2087 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_21/Mul_output_0", "outputIndexes": [ 2088 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2086 ], "outputIndexes": [ 6611 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6611 ], "outputIndexes": [ 6612 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/up_proj/Linear", "inputIndexes": [ 6612 ], "outputIndexes": [ 6613 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 5888294160, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6613 ], "outputIndexes": [ 6614 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6614 ], "outputIndexes": [ 2089 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2088, 2089 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_21/Mul_output_0", "outputIndexes": [ 2090 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2090 ], "outputIndexes": [ 6615 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6615 ], "outputIndexes": [ 6616 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/down_proj/Linear", "inputIndexes": [ 6616 ], "outputIndexes": [ 6617 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 5962022182, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6617 ], "outputIndexes": [ 6618 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.21/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6618 ], "outputIndexes": [ 2091 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2085, 2091 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_109_output_0", "outputIndexes": [ 2092 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2092, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_198_output_0", "outputIndexes": [ 2093 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_22/Mul_1_output_0", "inputIndexes": [ 2093 ], "outputIndexes": [ 2094 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 6035750204, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2094 ], "outputIndexes": [ 6619 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6619 ], "outputIndexes": [ 6620 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/q_proj/Linear", "inputIndexes": [ 6620 ], "outputIndexes": [ 6621 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 6035791164, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6621 ], "outputIndexes": [ 6622 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6622 ], "outputIndexes": [ 2095 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2094 ], "main_type": "NONE", "name": "/Shape_308_output_0", "outputIndexes": [ 2096 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2096 ], "main_type": "NONE", "name": "Shape8668", "outputIndexes": [ 2097 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2096 ], "main_type": "NONE", "name": "Rank8670", "outputIndexes": [ 2098 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2098, 2098 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8672", "outputIndexes": [ 2099 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2099 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8673", "outputIndexes": [ 2100 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2099, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8675", "outputIndexes": [ 2101 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2101 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8676", "outputIndexes": [ 2102 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2097, 2100, 2102, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8679", "outputIndexes": [ 2103 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2103 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8680", "outputIndexes": [ 2104 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2104, 2104 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8682", "outputIndexes": [ 2105 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2096, 2105, 2099 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_333_output_0", "outputIndexes": [ 2106 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2106, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_704_output_0", "outputIndexes": [ 2107 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2099 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8690", "outputIndexes": [ 2108 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2101 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8693", "outputIndexes": [ 2109 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2097, 2108, 2109, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8696", "outputIndexes": [ 2110 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2110 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8697", "outputIndexes": [ 2111 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2111 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8698", "outputIndexes": [ 2112 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2112, 2111 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8699", "outputIndexes": [ 2113 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2096, 2113, 2099 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_334_output_0", "outputIndexes": [ 2114 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2114, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_705_output_0", "outputIndexes": [ 2115 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2107, 2115, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_286_output_0", "outputIndexes": [ 2116 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2095, 2116 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_199_output_0", "outputIndexes": [ 2117 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2117 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_22/Cast_output_0", "outputIndexes": [ 2118 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_22/Mul_1_output_0", "inputIndexes": [ 2118 ], "outputIndexes": [ 2119 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 6059384146, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2119, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_176_output_0", "outputIndexes": [ 2120 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2119 ], "main_type": "NONE", "name": "/Shape_310_output_0", "outputIndexes": [ 2121 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2121 ], "main_type": "NONE", "name": "Shape8818", "outputIndexes": [ 2122 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2121 ], "main_type": "NONE", "name": "Rank8820", "outputIndexes": [ 2123 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2123, 2123 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8822", "outputIndexes": [ 2124 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2124 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8823", "outputIndexes": [ 2125 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2124, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8825", "outputIndexes": [ 2126 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2126 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8826", "outputIndexes": [ 2127 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2122, 2125, 2127, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8829", "outputIndexes": [ 2128 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2128 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8830", "outputIndexes": [ 2129 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2129 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8831", "outputIndexes": [ 2130 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2130, 2129 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8832", "outputIndexes": [ 2131 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2121, 2131, 2124 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_335_output_0", "outputIndexes": [ 2132 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2132, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_66_output_0", "outputIndexes": [ 2133 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2133, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_711_output_0", "outputIndexes": [ 2134 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2119, 2134, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_89_output_0", "outputIndexes": [ 2135 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2135 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_44_output_0", "outputIndexes": [ 2136 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2133, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_710_output_0", "outputIndexes": [ 2137 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2119, 20, 2137, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_88_output_0", "outputIndexes": [ 2138 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2136, 2138 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_289_output_0", "outputIndexes": [ 2139 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2139, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_177_output_0", "outputIndexes": [ 2140 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2120, 2140 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_110_output_0", "outputIndexes": [ 2141 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2094 ], "outputIndexes": [ 6623 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6623 ], "outputIndexes": [ 6624 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/k_proj/Linear", "inputIndexes": [ 6624 ], "outputIndexes": [ 6625 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 6059385170, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6625 ], "outputIndexes": [ 6626 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6626 ], "outputIndexes": [ 2142 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2106, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_706_output_0", "outputIndexes": [ 2143 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2114, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_707_output_0", "outputIndexes": [ 2144 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2143, 2144, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_287_output_0", "outputIndexes": [ 2145 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2142, 2145 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_200_output_0", "outputIndexes": [ 2146 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2146 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_22/Cast_output_0", "outputIndexes": [ 2147 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_22/Mul_1_output_0", "inputIndexes": [ 2147 ], "outputIndexes": [ 2148 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 6062334312, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2148, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_178_output_0", "outputIndexes": [ 2149 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2148 ], "main_type": "NONE", "name": "/Shape_311_output_0", "outputIndexes": [ 2150 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2150 ], "main_type": "NONE", "name": "Shape8714", "outputIndexes": [ 2151 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2150 ], "main_type": "NONE", "name": "Rank8716", "outputIndexes": [ 2152 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2152, 2152 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8718", "outputIndexes": [ 2153 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2153 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8719", "outputIndexes": [ 2154 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2153, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8721", "outputIndexes": [ 2155 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2155 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze8722", "outputIndexes": [ 2156 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2151, 2154, 2156, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice8725", "outputIndexes": [ 2157 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2157 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze8726", "outputIndexes": [ 2158 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2158 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8727", "outputIndexes": [ 2159 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2159, 2158 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp8728", "outputIndexes": [ 2160 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2150, 2160, 2153 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_336_output_0", "outputIndexes": [ 2161 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2161, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_67_output_0", "outputIndexes": [ 2162 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2162, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_713_output_0", "outputIndexes": [ 2163 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2148, 2163, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_91_output_0", "outputIndexes": [ 2164 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2164 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_45_output_0", "outputIndexes": [ 2165 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2162, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_712_output_0", "outputIndexes": [ 2166 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2148, 20, 2166, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_90_output_0", "outputIndexes": [ 2167 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2165, 2167 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_290_output_0", "outputIndexes": [ 2168 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2168, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_179_output_0", "outputIndexes": [ 2169 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2149, 2169 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_111_output_0", "outputIndexes": [ 2170 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2094 ], "outputIndexes": [ 6627 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6627 ], "outputIndexes": [ 6628 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/v_proj/Linear", "inputIndexes": [ 6628 ], "outputIndexes": [ 6629 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 6062335336, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6629 ], "outputIndexes": [ 6630 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6630 ], "outputIndexes": [ 2171 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2106, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_708_output_0", "outputIndexes": [ 2172 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2114, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_709_output_0", "outputIndexes": [ 2173 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2172, 2173, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_288_output_0", "outputIndexes": [ 2174 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2171, 2174 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_201_output_0", "outputIndexes": [ 2175 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2141, 2170, 2175, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_206_output_0", "outputIndexes": [ 2176 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2176 ], "outputIndexes": [ 6631 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6631 ], "outputIndexes": [ 6632 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/o_proj/Linear", "inputIndexes": [ 6632 ], "outputIndexes": [ 6633 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 6065284478, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6633 ], "outputIndexes": [ 6634 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6634 ], "outputIndexes": [ 2177 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2093, 2177 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_113_output_0", "outputIndexes": [ 2178 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_22/Mul_1_output_0", "inputIndexes": [ 2178 ], "outputIndexes": [ 2179 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 6088877460, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2179 ], "outputIndexes": [ 6635 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6635 ], "outputIndexes": [ 6636 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/gate_proj/Linear", "inputIndexes": [ 6636 ], "outputIndexes": [ 6637 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 6088918420, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6637 ], "outputIndexes": [ 6638 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6638 ], "outputIndexes": [ 2180 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2180 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_22/Mul_output_0", "outputIndexes": [ 2181 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2179 ], "outputIndexes": [ 6639 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6639 ], "outputIndexes": [ 6640 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/up_proj/Linear", "inputIndexes": [ 6640 ], "outputIndexes": [ 6641 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 6162646442, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6641 ], "outputIndexes": [ 6642 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6642 ], "outputIndexes": [ 2182 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2181, 2182 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_22/Mul_output_0", "outputIndexes": [ 2183 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2183 ], "outputIndexes": [ 6643 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6643 ], "outputIndexes": [ 6644 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/down_proj/Linear", "inputIndexes": [ 6644 ], "outputIndexes": [ 6645 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 6236374464, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6645 ], "outputIndexes": [ 6646 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.22/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6646 ], "outputIndexes": [ 2184 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2178, 2184 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_114_output_0", "outputIndexes": [ 2185 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2185, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_207_output_0", "outputIndexes": [ 2186 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_23/Mul_1_output_0", "inputIndexes": [ 2186 ], "outputIndexes": [ 2187 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 6310102486, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2187 ], "outputIndexes": [ 6647 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6647 ], "outputIndexes": [ 6648 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/q_proj/Linear", "inputIndexes": [ 6648 ], "outputIndexes": [ 6649 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 6310143446, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6649 ], "outputIndexes": [ 6650 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6650 ], "outputIndexes": [ 2188 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2187 ], "main_type": "NONE", "name": "/Shape_322_output_0", "outputIndexes": [ 2189 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2189 ], "main_type": "NONE", "name": "Shape9057", "outputIndexes": [ 2190 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2189 ], "main_type": "NONE", "name": "Rank9059", "outputIndexes": [ 2191 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2191, 2191 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9061", "outputIndexes": [ 2192 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2192 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9062", "outputIndexes": [ 2193 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2192, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9064", "outputIndexes": [ 2194 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2194 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9065", "outputIndexes": [ 2195 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2190, 2193, 2195, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9068", "outputIndexes": [ 2196 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2196 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9069", "outputIndexes": [ 2197 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2197, 2197 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9071", "outputIndexes": [ 2198 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2189, 2198, 2192 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_348_output_0", "outputIndexes": [ 2199 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2199, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_736_output_0", "outputIndexes": [ 2200 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2192 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9079", "outputIndexes": [ 2201 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2194 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9082", "outputIndexes": [ 2202 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2190, 2201, 2202, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9085", "outputIndexes": [ 2203 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2203 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9086", "outputIndexes": [ 2204 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2204 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9087", "outputIndexes": [ 2205 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2205, 2204 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9088", "outputIndexes": [ 2206 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2189, 2206, 2192 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_349_output_0", "outputIndexes": [ 2207 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2207, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_737_output_0", "outputIndexes": [ 2208 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2200, 2208, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_299_output_0", "outputIndexes": [ 2209 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2188, 2209 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_208_output_0", "outputIndexes": [ 2210 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2210 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_23/Cast_output_0", "outputIndexes": [ 2211 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_23/Mul_1_output_0", "inputIndexes": [ 2211 ], "outputIndexes": [ 2212 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 6333736428, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2212, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_184_output_0", "outputIndexes": [ 2213 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2212 ], "main_type": "NONE", "name": "/Shape_324_output_0", "outputIndexes": [ 2214 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2214 ], "main_type": "NONE", "name": "Shape9207", "outputIndexes": [ 2215 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2214 ], "main_type": "NONE", "name": "Rank9209", "outputIndexes": [ 2216 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2216, 2216 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9211", "outputIndexes": [ 2217 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2217 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9212", "outputIndexes": [ 2218 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2217, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9214", "outputIndexes": [ 2219 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2219 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9215", "outputIndexes": [ 2220 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2215, 2218, 2220, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9218", "outputIndexes": [ 2221 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2221 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9219", "outputIndexes": [ 2222 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2222 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9220", "outputIndexes": [ 2223 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2223, 2222 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9221", "outputIndexes": [ 2224 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2214, 2224, 2217 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_350_output_0", "outputIndexes": [ 2225 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2225, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_69_output_0", "outputIndexes": [ 2226 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2226, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_743_output_0", "outputIndexes": [ 2227 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2212, 2227, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_93_output_0", "outputIndexes": [ 2228 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2228 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_46_output_0", "outputIndexes": [ 2229 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2226, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_742_output_0", "outputIndexes": [ 2230 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2212, 20, 2230, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_92_output_0", "outputIndexes": [ 2231 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2229, 2231 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_302_output_0", "outputIndexes": [ 2232 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2232, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_185_output_0", "outputIndexes": [ 2233 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2213, 2233 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_115_output_0", "outputIndexes": [ 2234 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2187 ], "outputIndexes": [ 6651 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6651 ], "outputIndexes": [ 6652 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/k_proj/Linear", "inputIndexes": [ 6652 ], "outputIndexes": [ 6653 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 6333737452, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6653 ], "outputIndexes": [ 6654 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6654 ], "outputIndexes": [ 2235 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2199, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_738_output_0", "outputIndexes": [ 2236 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2207, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_739_output_0", "outputIndexes": [ 2237 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2236, 2237, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_300_output_0", "outputIndexes": [ 2238 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2235, 2238 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_209_output_0", "outputIndexes": [ 2239 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2239 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_23/Cast_output_0", "outputIndexes": [ 2240 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_23/Mul_1_output_0", "inputIndexes": [ 2240 ], "outputIndexes": [ 2241 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 6336686594, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2241, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_186_output_0", "outputIndexes": [ 2242 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2241 ], "main_type": "NONE", "name": "/Shape_325_output_0", "outputIndexes": [ 2243 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2243 ], "main_type": "NONE", "name": "Shape9103", "outputIndexes": [ 2244 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2243 ], "main_type": "NONE", "name": "Rank9105", "outputIndexes": [ 2245 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2245, 2245 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9107", "outputIndexes": [ 2246 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2246 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9108", "outputIndexes": [ 2247 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2246, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9110", "outputIndexes": [ 2248 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2248 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9111", "outputIndexes": [ 2249 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2244, 2247, 2249, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9114", "outputIndexes": [ 2250 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2250 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9115", "outputIndexes": [ 2251 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2251 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9116", "outputIndexes": [ 2252 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2252, 2251 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9117", "outputIndexes": [ 2253 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2243, 2253, 2246 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_351_output_0", "outputIndexes": [ 2254 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2254, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_70_output_0", "outputIndexes": [ 2255 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2255, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_745_output_0", "outputIndexes": [ 2256 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2241, 2256, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_95_output_0", "outputIndexes": [ 2257 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2257 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_47_output_0", "outputIndexes": [ 2258 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2255, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_744_output_0", "outputIndexes": [ 2259 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2241, 20, 2259, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_94_output_0", "outputIndexes": [ 2260 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2258, 2260 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_303_output_0", "outputIndexes": [ 2261 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2261, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_187_output_0", "outputIndexes": [ 2262 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2242, 2262 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_116_output_0", "outputIndexes": [ 2263 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2187 ], "outputIndexes": [ 6655 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6655 ], "outputIndexes": [ 6656 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/v_proj/Linear", "inputIndexes": [ 6656 ], "outputIndexes": [ 6657 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 6336687618, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6657 ], "outputIndexes": [ 6658 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6658 ], "outputIndexes": [ 2264 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2199, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_740_output_0", "outputIndexes": [ 2265 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2207, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_741_output_0", "outputIndexes": [ 2266 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2265, 2266, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_301_output_0", "outputIndexes": [ 2267 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2264, 2267 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_210_output_0", "outputIndexes": [ 2268 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2234, 2263, 2268, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_215_output_0", "outputIndexes": [ 2269 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2269 ], "outputIndexes": [ 6659 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6659 ], "outputIndexes": [ 6660 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/o_proj/Linear", "inputIndexes": [ 6660 ], "outputIndexes": [ 6661 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 6339636760, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6661 ], "outputIndexes": [ 6662 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6662 ], "outputIndexes": [ 2270 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2186, 2270 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_118_output_0", "outputIndexes": [ 2271 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_23/Mul_1_output_0", "inputIndexes": [ 2271 ], "outputIndexes": [ 2272 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 6363229742, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2272 ], "outputIndexes": [ 6663 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6663 ], "outputIndexes": [ 6664 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/gate_proj/Linear", "inputIndexes": [ 6664 ], "outputIndexes": [ 6665 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 6363270702, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6665 ], "outputIndexes": [ 6666 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6666 ], "outputIndexes": [ 2273 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2273 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_23/Mul_output_0", "outputIndexes": [ 2274 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2272 ], "outputIndexes": [ 6667 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6667 ], "outputIndexes": [ 6668 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/up_proj/Linear", "inputIndexes": [ 6668 ], "outputIndexes": [ 6669 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 6436998724, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6669 ], "outputIndexes": [ 6670 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6670 ], "outputIndexes": [ 2275 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2274, 2275 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_23/Mul_output_0", "outputIndexes": [ 2276 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2276 ], "outputIndexes": [ 6671 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6671 ], "outputIndexes": [ 6672 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/down_proj/Linear", "inputIndexes": [ 6672 ], "outputIndexes": [ 6673 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 6510726746, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6673 ], "outputIndexes": [ 6674 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.23/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6674 ], "outputIndexes": [ 2277 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2271, 2277 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_119_output_0", "outputIndexes": [ 2278 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2278, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_216_output_0", "outputIndexes": [ 2279 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_24/Mul_1_output_0", "inputIndexes": [ 2279 ], "outputIndexes": [ 2280 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 6584454768, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2280 ], "outputIndexes": [ 6675 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6675 ], "outputIndexes": [ 6676 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/q_proj/Linear", "inputIndexes": [ 6676 ], "outputIndexes": [ 6677 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 6584495728, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6677 ], "outputIndexes": [ 6678 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6678 ], "outputIndexes": [ 2281 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2280 ], "main_type": "NONE", "name": "/Shape_336_output_0", "outputIndexes": [ 2282 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2282 ], "main_type": "NONE", "name": "Shape9446", "outputIndexes": [ 2283 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2282 ], "main_type": "NONE", "name": "Rank9448", "outputIndexes": [ 2284 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2284, 2284 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9450", "outputIndexes": [ 2285 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2285 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9451", "outputIndexes": [ 2286 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2285, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9453", "outputIndexes": [ 2287 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2287 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9454", "outputIndexes": [ 2288 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2283, 2286, 2288, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9457", "outputIndexes": [ 2289 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2289 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9458", "outputIndexes": [ 2290 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2290, 2290 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9460", "outputIndexes": [ 2291 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2282, 2291, 2285 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_363_output_0", "outputIndexes": [ 2292 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2292, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_768_output_0", "outputIndexes": [ 2293 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2285 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9468", "outputIndexes": [ 2294 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2287 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9471", "outputIndexes": [ 2295 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2283, 2294, 2295, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9474", "outputIndexes": [ 2296 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2296 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9475", "outputIndexes": [ 2297 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2297 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9476", "outputIndexes": [ 2298 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2298, 2297 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9477", "outputIndexes": [ 2299 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2282, 2299, 2285 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_364_output_0", "outputIndexes": [ 2300 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2300, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_769_output_0", "outputIndexes": [ 2301 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2293, 2301, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_312_output_0", "outputIndexes": [ 2302 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2281, 2302 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_217_output_0", "outputIndexes": [ 2303 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2303 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_24/Cast_output_0", "outputIndexes": [ 2304 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_24/Mul_1_output_0", "inputIndexes": [ 2304 ], "outputIndexes": [ 2305 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 6608088710, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2305, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_192_output_0", "outputIndexes": [ 2306 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2305 ], "main_type": "NONE", "name": "/Shape_338_output_0", "outputIndexes": [ 2307 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2307 ], "main_type": "NONE", "name": "Shape9596", "outputIndexes": [ 2308 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2307 ], "main_type": "NONE", "name": "Rank9598", "outputIndexes": [ 2309 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2309, 2309 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9600", "outputIndexes": [ 2310 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2310 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9601", "outputIndexes": [ 2311 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2310, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9603", "outputIndexes": [ 2312 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2312 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9604", "outputIndexes": [ 2313 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2308, 2311, 2313, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9607", "outputIndexes": [ 2314 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2314 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9608", "outputIndexes": [ 2315 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2315 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9609", "outputIndexes": [ 2316 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2316, 2315 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9610", "outputIndexes": [ 2317 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2307, 2317, 2310 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_365_output_0", "outputIndexes": [ 2318 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2318, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_72_output_0", "outputIndexes": [ 2319 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2319, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_775_output_0", "outputIndexes": [ 2320 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2305, 2320, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_97_output_0", "outputIndexes": [ 2321 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2321 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_48_output_0", "outputIndexes": [ 2322 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2319, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_774_output_0", "outputIndexes": [ 2323 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2305, 20, 2323, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_96_output_0", "outputIndexes": [ 2324 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2322, 2324 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_315_output_0", "outputIndexes": [ 2325 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2325, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_193_output_0", "outputIndexes": [ 2326 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2306, 2326 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_120_output_0", "outputIndexes": [ 2327 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2280 ], "outputIndexes": [ 6679 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6679 ], "outputIndexes": [ 6680 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/k_proj/Linear", "inputIndexes": [ 6680 ], "outputIndexes": [ 6681 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 6608089734, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6681 ], "outputIndexes": [ 6682 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6682 ], "outputIndexes": [ 2328 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2292, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_770_output_0", "outputIndexes": [ 2329 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2300, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_771_output_0", "outputIndexes": [ 2330 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2329, 2330, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_313_output_0", "outputIndexes": [ 2331 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2328, 2331 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_218_output_0", "outputIndexes": [ 2332 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2332 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_24/Cast_output_0", "outputIndexes": [ 2333 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_24/Mul_1_output_0", "inputIndexes": [ 2333 ], "outputIndexes": [ 2334 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 6611038876, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2334, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_194_output_0", "outputIndexes": [ 2335 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2334 ], "main_type": "NONE", "name": "/Shape_339_output_0", "outputIndexes": [ 2336 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2336 ], "main_type": "NONE", "name": "Shape9492", "outputIndexes": [ 2337 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2336 ], "main_type": "NONE", "name": "Rank9494", "outputIndexes": [ 2338 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2338, 2338 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9496", "outputIndexes": [ 2339 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2339 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9497", "outputIndexes": [ 2340 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2339, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9499", "outputIndexes": [ 2341 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2341 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9500", "outputIndexes": [ 2342 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2337, 2340, 2342, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9503", "outputIndexes": [ 2343 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2343 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9504", "outputIndexes": [ 2344 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2344 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9505", "outputIndexes": [ 2345 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2345, 2344 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9506", "outputIndexes": [ 2346 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2336, 2346, 2339 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_366_output_0", "outputIndexes": [ 2347 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2347, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_73_output_0", "outputIndexes": [ 2348 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2348, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_777_output_0", "outputIndexes": [ 2349 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2334, 2349, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_99_output_0", "outputIndexes": [ 2350 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2350 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_49_output_0", "outputIndexes": [ 2351 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2348, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_776_output_0", "outputIndexes": [ 2352 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2334, 20, 2352, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_98_output_0", "outputIndexes": [ 2353 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2351, 2353 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_316_output_0", "outputIndexes": [ 2354 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2354, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_195_output_0", "outputIndexes": [ 2355 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2335, 2355 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_121_output_0", "outputIndexes": [ 2356 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2280 ], "outputIndexes": [ 6683 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6683 ], "outputIndexes": [ 6684 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/v_proj/Linear", "inputIndexes": [ 6684 ], "outputIndexes": [ 6685 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 6611039900, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6685 ], "outputIndexes": [ 6686 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6686 ], "outputIndexes": [ 2357 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2292, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_772_output_0", "outputIndexes": [ 2358 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2300, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_773_output_0", "outputIndexes": [ 2359 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2358, 2359, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_314_output_0", "outputIndexes": [ 2360 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2357, 2360 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_219_output_0", "outputIndexes": [ 2361 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2327, 2356, 2361, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_224_output_0", "outputIndexes": [ 2362 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2362 ], "outputIndexes": [ 6687 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6687 ], "outputIndexes": [ 6688 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/o_proj/Linear", "inputIndexes": [ 6688 ], "outputIndexes": [ 6689 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 6613989042, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6689 ], "outputIndexes": [ 6690 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6690 ], "outputIndexes": [ 2363 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2279, 2363 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_123_output_0", "outputIndexes": [ 2364 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_24/Mul_1_output_0", "inputIndexes": [ 2364 ], "outputIndexes": [ 2365 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 6637582024, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2365 ], "outputIndexes": [ 6691 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6691 ], "outputIndexes": [ 6692 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/gate_proj/Linear", "inputIndexes": [ 6692 ], "outputIndexes": [ 6693 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 6637622984, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6693 ], "outputIndexes": [ 6694 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6694 ], "outputIndexes": [ 2366 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2366 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_24/Mul_output_0", "outputIndexes": [ 2367 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2365 ], "outputIndexes": [ 6695 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6695 ], "outputIndexes": [ 6696 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/up_proj/Linear", "inputIndexes": [ 6696 ], "outputIndexes": [ 6697 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 6711351006, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6697 ], "outputIndexes": [ 6698 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6698 ], "outputIndexes": [ 2368 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2367, 2368 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_24/Mul_output_0", "outputIndexes": [ 2369 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2369 ], "outputIndexes": [ 6699 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6699 ], "outputIndexes": [ 6700 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/down_proj/Linear", "inputIndexes": [ 6700 ], "outputIndexes": [ 6701 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 6785079028, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6701 ], "outputIndexes": [ 6702 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.24/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6702 ], "outputIndexes": [ 2370 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2364, 2370 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_124_output_0", "outputIndexes": [ 2371 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2371, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_225_output_0", "outputIndexes": [ 2372 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_25/Mul_1_output_0", "inputIndexes": [ 2372 ], "outputIndexes": [ 2373 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 6858807050, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2373 ], "outputIndexes": [ 6703 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6703 ], "outputIndexes": [ 6704 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/q_proj/Linear", "inputIndexes": [ 6704 ], "outputIndexes": [ 6705 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 6858848010, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6705 ], "outputIndexes": [ 6706 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6706 ], "outputIndexes": [ 2374 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2373 ], "main_type": "NONE", "name": "/Shape_350_output_0", "outputIndexes": [ 2375 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2375 ], "main_type": "NONE", "name": "Shape9835", "outputIndexes": [ 2376 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2375 ], "main_type": "NONE", "name": "Rank9837", "outputIndexes": [ 2377 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2377, 2377 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9839", "outputIndexes": [ 2378 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2378 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9840", "outputIndexes": [ 2379 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2378, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9842", "outputIndexes": [ 2380 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2380 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9843", "outputIndexes": [ 2381 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2376, 2379, 2381, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9846", "outputIndexes": [ 2382 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2382 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9847", "outputIndexes": [ 2383 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2383, 2383 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9849", "outputIndexes": [ 2384 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2375, 2384, 2378 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_378_output_0", "outputIndexes": [ 2385 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2385, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_800_output_0", "outputIndexes": [ 2386 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2378 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9857", "outputIndexes": [ 2387 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2380 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9860", "outputIndexes": [ 2388 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2376, 2387, 2388, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9863", "outputIndexes": [ 2389 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2389 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9864", "outputIndexes": [ 2390 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2390 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9865", "outputIndexes": [ 2391 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2391, 2390 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9866", "outputIndexes": [ 2392 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2375, 2392, 2378 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_379_output_0", "outputIndexes": [ 2393 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2393, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_801_output_0", "outputIndexes": [ 2394 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2386, 2394, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_325_output_0", "outputIndexes": [ 2395 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2374, 2395 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_226_output_0", "outputIndexes": [ 2396 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2396 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_25/Cast_output_0", "outputIndexes": [ 2397 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_25/Mul_1_output_0", "inputIndexes": [ 2397 ], "outputIndexes": [ 2398 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 6882440992, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2398, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_200_output_0", "outputIndexes": [ 2399 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2398 ], "main_type": "NONE", "name": "/Shape_352_output_0", "outputIndexes": [ 2400 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2400 ], "main_type": "NONE", "name": "Shape9985", "outputIndexes": [ 2401 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2400 ], "main_type": "NONE", "name": "Rank9987", "outputIndexes": [ 2402 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2402, 2402 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9989", "outputIndexes": [ 2403 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2403 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9990", "outputIndexes": [ 2404 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2403, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9992", "outputIndexes": [ 2405 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2405 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9993", "outputIndexes": [ 2406 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2401, 2404, 2406, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9996", "outputIndexes": [ 2407 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2407 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9997", "outputIndexes": [ 2408 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2408 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9998", "outputIndexes": [ 2409 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2409, 2408 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9999", "outputIndexes": [ 2410 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2400, 2410, 2403 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_380_output_0", "outputIndexes": [ 2411 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2411, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_75_output_0", "outputIndexes": [ 2412 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2412, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_807_output_0", "outputIndexes": [ 2413 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2398, 2413, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_101_output_0", "outputIndexes": [ 2414 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2414 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_50_output_0", "outputIndexes": [ 2415 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2412, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_806_output_0", "outputIndexes": [ 2416 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2398, 20, 2416, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_100_output_0", "outputIndexes": [ 2417 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2415, 2417 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_328_output_0", "outputIndexes": [ 2418 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2418, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_201_output_0", "outputIndexes": [ 2419 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2399, 2419 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_125_output_0", "outputIndexes": [ 2420 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2373 ], "outputIndexes": [ 6707 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6707 ], "outputIndexes": [ 6708 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/k_proj/Linear", "inputIndexes": [ 6708 ], "outputIndexes": [ 6709 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 6882442016, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6709 ], "outputIndexes": [ 6710 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6710 ], "outputIndexes": [ 2421 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2385, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_802_output_0", "outputIndexes": [ 2422 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2393, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_803_output_0", "outputIndexes": [ 2423 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2422, 2423, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_326_output_0", "outputIndexes": [ 2424 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2421, 2424 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_227_output_0", "outputIndexes": [ 2425 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2425 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_25/Cast_output_0", "outputIndexes": [ 2426 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_25/Mul_1_output_0", "inputIndexes": [ 2426 ], "outputIndexes": [ 2427 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 6885391158, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2427, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_202_output_0", "outputIndexes": [ 2428 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2427 ], "main_type": "NONE", "name": "/Shape_353_output_0", "outputIndexes": [ 2429 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2429 ], "main_type": "NONE", "name": "Shape9881", "outputIndexes": [ 2430 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2429 ], "main_type": "NONE", "name": "Rank9883", "outputIndexes": [ 2431 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2431, 2431 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9885", "outputIndexes": [ 2432 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2432 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9886", "outputIndexes": [ 2433 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2432, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9888", "outputIndexes": [ 2434 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2434 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze9889", "outputIndexes": [ 2435 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2430, 2433, 2435, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice9892", "outputIndexes": [ 2436 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2436 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze9893", "outputIndexes": [ 2437 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2437 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9894", "outputIndexes": [ 2438 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2438, 2437 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp9895", "outputIndexes": [ 2439 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2429, 2439, 2432 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_381_output_0", "outputIndexes": [ 2440 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2440, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_76_output_0", "outputIndexes": [ 2441 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2441, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_809_output_0", "outputIndexes": [ 2442 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2427, 2442, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_103_output_0", "outputIndexes": [ 2443 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2443 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_51_output_0", "outputIndexes": [ 2444 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2441, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_808_output_0", "outputIndexes": [ 2445 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2427, 20, 2445, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_102_output_0", "outputIndexes": [ 2446 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2444, 2446 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_329_output_0", "outputIndexes": [ 2447 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2447, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_203_output_0", "outputIndexes": [ 2448 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2428, 2448 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_126_output_0", "outputIndexes": [ 2449 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2373 ], "outputIndexes": [ 6711 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6711 ], "outputIndexes": [ 6712 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/v_proj/Linear", "inputIndexes": [ 6712 ], "outputIndexes": [ 6713 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 6885392182, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6713 ], "outputIndexes": [ 6714 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6714 ], "outputIndexes": [ 2450 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2385, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_804_output_0", "outputIndexes": [ 2451 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2393, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_805_output_0", "outputIndexes": [ 2452 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2451, 2452, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_327_output_0", "outputIndexes": [ 2453 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2450, 2453 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_228_output_0", "outputIndexes": [ 2454 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2420, 2449, 2454, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_233_output_0", "outputIndexes": [ 2455 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2455 ], "outputIndexes": [ 6715 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6715 ], "outputIndexes": [ 6716 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/o_proj/Linear", "inputIndexes": [ 6716 ], "outputIndexes": [ 6717 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 6888341324, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6717 ], "outputIndexes": [ 6718 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6718 ], "outputIndexes": [ 2456 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2372, 2456 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_128_output_0", "outputIndexes": [ 2457 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_25/Mul_1_output_0", "inputIndexes": [ 2457 ], "outputIndexes": [ 2458 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 6911934306, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2458 ], "outputIndexes": [ 6719 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6719 ], "outputIndexes": [ 6720 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/gate_proj/Linear", "inputIndexes": [ 6720 ], "outputIndexes": [ 6721 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 6911975266, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6721 ], "outputIndexes": [ 6722 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6722 ], "outputIndexes": [ 2459 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2459 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_25/Mul_output_0", "outputIndexes": [ 2460 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2458 ], "outputIndexes": [ 6723 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6723 ], "outputIndexes": [ 6724 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/up_proj/Linear", "inputIndexes": [ 6724 ], "outputIndexes": [ 6725 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 6985703288, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6725 ], "outputIndexes": [ 6726 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6726 ], "outputIndexes": [ 2461 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2460, 2461 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_25/Mul_output_0", "outputIndexes": [ 2462 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2462 ], "outputIndexes": [ 6727 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6727 ], "outputIndexes": [ 6728 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/down_proj/Linear", "inputIndexes": [ 6728 ], "outputIndexes": [ 6729 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 7059431310, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6729 ], "outputIndexes": [ 6730 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.25/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6730 ], "outputIndexes": [ 2463 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2457, 2463 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_129_output_0", "outputIndexes": [ 2464 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2464, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_234_output_0", "outputIndexes": [ 2465 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_26/Mul_1_output_0", "inputIndexes": [ 2465 ], "outputIndexes": [ 2466 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 7133159332, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2466 ], "outputIndexes": [ 6731 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6731 ], "outputIndexes": [ 6732 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/q_proj/Linear", "inputIndexes": [ 6732 ], "outputIndexes": [ 6733 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 7133200292, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6733 ], "outputIndexes": [ 6734 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6734 ], "outputIndexes": [ 2467 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2466 ], "main_type": "NONE", "name": "/Shape_364_output_0", "outputIndexes": [ 2468 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2468 ], "main_type": "NONE", "name": "Shape10224", "outputIndexes": [ 2469 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2468 ], "main_type": "NONE", "name": "Rank10226", "outputIndexes": [ 2470 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2470, 2470 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10228", "outputIndexes": [ 2471 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2471 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10229", "outputIndexes": [ 2472 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2471, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10231", "outputIndexes": [ 2473 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2473 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10232", "outputIndexes": [ 2474 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2469, 2472, 2474, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10235", "outputIndexes": [ 2475 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2475 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10236", "outputIndexes": [ 2476 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2476, 2476 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10238", "outputIndexes": [ 2477 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2468, 2477, 2471 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_393_output_0", "outputIndexes": [ 2478 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2478, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_832_output_0", "outputIndexes": [ 2479 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2471 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10246", "outputIndexes": [ 2480 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2473 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10249", "outputIndexes": [ 2481 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2469, 2480, 2481, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10252", "outputIndexes": [ 2482 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2482 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10253", "outputIndexes": [ 2483 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2483 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10254", "outputIndexes": [ 2484 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2484, 2483 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10255", "outputIndexes": [ 2485 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2468, 2485, 2471 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_394_output_0", "outputIndexes": [ 2486 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2486, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_833_output_0", "outputIndexes": [ 2487 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2479, 2487, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_338_output_0", "outputIndexes": [ 2488 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2467, 2488 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_235_output_0", "outputIndexes": [ 2489 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2489 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_26/Cast_output_0", "outputIndexes": [ 2490 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_26/Mul_1_output_0", "inputIndexes": [ 2490 ], "outputIndexes": [ 2491 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 7156793274, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2491, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_208_output_0", "outputIndexes": [ 2492 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2491 ], "main_type": "NONE", "name": "/Shape_366_output_0", "outputIndexes": [ 2493 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2493 ], "main_type": "NONE", "name": "Shape10374", "outputIndexes": [ 2494 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2493 ], "main_type": "NONE", "name": "Rank10376", "outputIndexes": [ 2495 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2495, 2495 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10378", "outputIndexes": [ 2496 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2496 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10379", "outputIndexes": [ 2497 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2496, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10381", "outputIndexes": [ 2498 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2498 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10382", "outputIndexes": [ 2499 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2494, 2497, 2499, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10385", "outputIndexes": [ 2500 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2500 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10386", "outputIndexes": [ 2501 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2501 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10387", "outputIndexes": [ 2502 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2502, 2501 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10388", "outputIndexes": [ 2503 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2493, 2503, 2496 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_395_output_0", "outputIndexes": [ 2504 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2504, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_78_output_0", "outputIndexes": [ 2505 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2505, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_839_output_0", "outputIndexes": [ 2506 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2491, 2506, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_105_output_0", "outputIndexes": [ 2507 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2507 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_52_output_0", "outputIndexes": [ 2508 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2505, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_838_output_0", "outputIndexes": [ 2509 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2491, 20, 2509, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_104_output_0", "outputIndexes": [ 2510 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2508, 2510 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_341_output_0", "outputIndexes": [ 2511 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2511, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_209_output_0", "outputIndexes": [ 2512 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2492, 2512 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_130_output_0", "outputIndexes": [ 2513 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2466 ], "outputIndexes": [ 6735 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6735 ], "outputIndexes": [ 6736 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/k_proj/Linear", "inputIndexes": [ 6736 ], "outputIndexes": [ 6737 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 7156794298, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6737 ], "outputIndexes": [ 6738 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6738 ], "outputIndexes": [ 2514 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2478, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_834_output_0", "outputIndexes": [ 2515 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2486, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_835_output_0", "outputIndexes": [ 2516 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2515, 2516, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_339_output_0", "outputIndexes": [ 2517 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2514, 2517 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_236_output_0", "outputIndexes": [ 2518 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2518 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_26/Cast_output_0", "outputIndexes": [ 2519 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_26/Mul_1_output_0", "inputIndexes": [ 2519 ], "outputIndexes": [ 2520 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 7159743440, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2520, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_210_output_0", "outputIndexes": [ 2521 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2520 ], "main_type": "NONE", "name": "/Shape_367_output_0", "outputIndexes": [ 2522 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2522 ], "main_type": "NONE", "name": "Shape10270", "outputIndexes": [ 2523 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2522 ], "main_type": "NONE", "name": "Rank10272", "outputIndexes": [ 2524 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2524, 2524 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10274", "outputIndexes": [ 2525 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2525 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10275", "outputIndexes": [ 2526 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2525, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10277", "outputIndexes": [ 2527 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2527 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10278", "outputIndexes": [ 2528 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2523, 2526, 2528, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10281", "outputIndexes": [ 2529 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2529 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10282", "outputIndexes": [ 2530 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2530 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10283", "outputIndexes": [ 2531 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2531, 2530 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10284", "outputIndexes": [ 2532 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2522, 2532, 2525 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_396_output_0", "outputIndexes": [ 2533 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2533, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_79_output_0", "outputIndexes": [ 2534 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2534, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_841_output_0", "outputIndexes": [ 2535 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2520, 2535, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_107_output_0", "outputIndexes": [ 2536 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2536 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_53_output_0", "outputIndexes": [ 2537 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2534, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_840_output_0", "outputIndexes": [ 2538 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2520, 20, 2538, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_106_output_0", "outputIndexes": [ 2539 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2537, 2539 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_342_output_0", "outputIndexes": [ 2540 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2540, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_211_output_0", "outputIndexes": [ 2541 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2521, 2541 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_131_output_0", "outputIndexes": [ 2542 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2466 ], "outputIndexes": [ 6739 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6739 ], "outputIndexes": [ 6740 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/v_proj/Linear", "inputIndexes": [ 6740 ], "outputIndexes": [ 6741 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 7159744464, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6741 ], "outputIndexes": [ 6742 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6742 ], "outputIndexes": [ 2543 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2478, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_836_output_0", "outputIndexes": [ 2544 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2486, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_837_output_0", "outputIndexes": [ 2545 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2544, 2545, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_340_output_0", "outputIndexes": [ 2546 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2543, 2546 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_237_output_0", "outputIndexes": [ 2547 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2513, 2542, 2547, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_242_output_0", "outputIndexes": [ 2548 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2548 ], "outputIndexes": [ 6743 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6743 ], "outputIndexes": [ 6744 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/o_proj/Linear", "inputIndexes": [ 6744 ], "outputIndexes": [ 6745 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 7162693606, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6745 ], "outputIndexes": [ 6746 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6746 ], "outputIndexes": [ 2549 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2465, 2549 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_133_output_0", "outputIndexes": [ 2550 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_26/Mul_1_output_0", "inputIndexes": [ 2550 ], "outputIndexes": [ 2551 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 7186286588, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2551 ], "outputIndexes": [ 6747 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6747 ], "outputIndexes": [ 6748 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/gate_proj/Linear", "inputIndexes": [ 6748 ], "outputIndexes": [ 6749 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 7186327548, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6749 ], "outputIndexes": [ 6750 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6750 ], "outputIndexes": [ 2552 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2552 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_26/Mul_output_0", "outputIndexes": [ 2553 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2551 ], "outputIndexes": [ 6751 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6751 ], "outputIndexes": [ 6752 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/up_proj/Linear", "inputIndexes": [ 6752 ], "outputIndexes": [ 6753 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 7260055570, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6753 ], "outputIndexes": [ 6754 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6754 ], "outputIndexes": [ 2554 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2553, 2554 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_26/Mul_output_0", "outputIndexes": [ 2555 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2555 ], "outputIndexes": [ 6755 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6755 ], "outputIndexes": [ 6756 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/down_proj/Linear", "inputIndexes": [ 6756 ], "outputIndexes": [ 6757 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 7333783592, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6757 ], "outputIndexes": [ 6758 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.26/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6758 ], "outputIndexes": [ 2556 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2550, 2556 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_134_output_0", "outputIndexes": [ 2557 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2557, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_243_output_0", "outputIndexes": [ 2558 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_27/Mul_1_output_0", "inputIndexes": [ 2558 ], "outputIndexes": [ 2559 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 7407511614, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2559 ], "outputIndexes": [ 6759 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6759 ], "outputIndexes": [ 6760 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/q_proj/Linear", "inputIndexes": [ 6760 ], "outputIndexes": [ 6761 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 7407552574, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6761 ], "outputIndexes": [ 6762 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6762 ], "outputIndexes": [ 2560 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2559 ], "main_type": "NONE", "name": "/Shape_378_output_0", "outputIndexes": [ 2561 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2561 ], "main_type": "NONE", "name": "Shape10613", "outputIndexes": [ 2562 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2561 ], "main_type": "NONE", "name": "Rank10615", "outputIndexes": [ 2563 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2563, 2563 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10617", "outputIndexes": [ 2564 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2564 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10618", "outputIndexes": [ 2565 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2564, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10620", "outputIndexes": [ 2566 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2566 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10621", "outputIndexes": [ 2567 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2562, 2565, 2567, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10624", "outputIndexes": [ 2568 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2568 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10625", "outputIndexes": [ 2569 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2569, 2569 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10627", "outputIndexes": [ 2570 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2561, 2570, 2564 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_408_output_0", "outputIndexes": [ 2571 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2571, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_864_output_0", "outputIndexes": [ 2572 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2564 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10635", "outputIndexes": [ 2573 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2566 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10638", "outputIndexes": [ 2574 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2562, 2573, 2574, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10641", "outputIndexes": [ 2575 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2575 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10642", "outputIndexes": [ 2576 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2576 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10643", "outputIndexes": [ 2577 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2577, 2576 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10644", "outputIndexes": [ 2578 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2561, 2578, 2564 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_409_output_0", "outputIndexes": [ 2579 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2579, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_865_output_0", "outputIndexes": [ 2580 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2572, 2580, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_351_output_0", "outputIndexes": [ 2581 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2560, 2581 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_244_output_0", "outputIndexes": [ 2582 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2582 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_27/Cast_output_0", "outputIndexes": [ 2583 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_27/Mul_1_output_0", "inputIndexes": [ 2583 ], "outputIndexes": [ 2584 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 7431145556, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2584, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_216_output_0", "outputIndexes": [ 2585 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2584 ], "main_type": "NONE", "name": "/Shape_380_output_0", "outputIndexes": [ 2586 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2586 ], "main_type": "NONE", "name": "Shape10763", "outputIndexes": [ 2587 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2586 ], "main_type": "NONE", "name": "Rank10765", "outputIndexes": [ 2588 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2588, 2588 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10767", "outputIndexes": [ 2589 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2589 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10768", "outputIndexes": [ 2590 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2589, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10770", "outputIndexes": [ 2591 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2591 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10771", "outputIndexes": [ 2592 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2587, 2590, 2592, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10774", "outputIndexes": [ 2593 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2593 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10775", "outputIndexes": [ 2594 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2594 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10776", "outputIndexes": [ 2595 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2595, 2594 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10777", "outputIndexes": [ 2596 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2586, 2596, 2589 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_410_output_0", "outputIndexes": [ 2597 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2597, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_81_output_0", "outputIndexes": [ 2598 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2598, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_871_output_0", "outputIndexes": [ 2599 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2584, 2599, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_109_output_0", "outputIndexes": [ 2600 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2600 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_54_output_0", "outputIndexes": [ 2601 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2598, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_870_output_0", "outputIndexes": [ 2602 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2584, 20, 2602, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_108_output_0", "outputIndexes": [ 2603 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2601, 2603 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_354_output_0", "outputIndexes": [ 2604 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2604, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_217_output_0", "outputIndexes": [ 2605 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2585, 2605 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_135_output_0", "outputIndexes": [ 2606 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2559 ], "outputIndexes": [ 6763 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6763 ], "outputIndexes": [ 6764 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/k_proj/Linear", "inputIndexes": [ 6764 ], "outputIndexes": [ 6765 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 7431146580, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6765 ], "outputIndexes": [ 6766 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6766 ], "outputIndexes": [ 2607 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2571, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_866_output_0", "outputIndexes": [ 2608 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2579, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_867_output_0", "outputIndexes": [ 2609 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2608, 2609, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_352_output_0", "outputIndexes": [ 2610 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2607, 2610 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_245_output_0", "outputIndexes": [ 2611 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2611 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_27/Cast_output_0", "outputIndexes": [ 2612 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_27/Mul_1_output_0", "inputIndexes": [ 2612 ], "outputIndexes": [ 2613 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 7434095722, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2613, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_218_output_0", "outputIndexes": [ 2614 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2613 ], "main_type": "NONE", "name": "/Shape_381_output_0", "outputIndexes": [ 2615 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2615 ], "main_type": "NONE", "name": "Shape10659", "outputIndexes": [ 2616 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2615 ], "main_type": "NONE", "name": "Rank10661", "outputIndexes": [ 2617 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2617, 2617 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10663", "outputIndexes": [ 2618 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2618 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10664", "outputIndexes": [ 2619 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2618, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10666", "outputIndexes": [ 2620 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2620 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze10667", "outputIndexes": [ 2621 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2616, 2619, 2621, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice10670", "outputIndexes": [ 2622 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2622 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze10671", "outputIndexes": [ 2623 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2623 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10672", "outputIndexes": [ 2624 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2624, 2623 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp10673", "outputIndexes": [ 2625 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2615, 2625, 2618 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_411_output_0", "outputIndexes": [ 2626 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2626, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_82_output_0", "outputIndexes": [ 2627 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2627, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_873_output_0", "outputIndexes": [ 2628 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2613, 2628, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_111_output_0", "outputIndexes": [ 2629 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2629 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_55_output_0", "outputIndexes": [ 2630 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2627, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_872_output_0", "outputIndexes": [ 2631 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2613, 20, 2631, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_110_output_0", "outputIndexes": [ 2632 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2630, 2632 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_355_output_0", "outputIndexes": [ 2633 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2633, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_219_output_0", "outputIndexes": [ 2634 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2614, 2634 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_136_output_0", "outputIndexes": [ 2635 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2559 ], "outputIndexes": [ 6767 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6767 ], "outputIndexes": [ 6768 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/v_proj/Linear", "inputIndexes": [ 6768 ], "outputIndexes": [ 6769 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 7434096746, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6769 ], "outputIndexes": [ 6770 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6770 ], "outputIndexes": [ 2636 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2571, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_868_output_0", "outputIndexes": [ 2637 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2579, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_869_output_0", "outputIndexes": [ 2638 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2637, 2638, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_353_output_0", "outputIndexes": [ 2639 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2636, 2639 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_246_output_0", "outputIndexes": [ 2640 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2606, 2635, 2640, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_251_output_0", "outputIndexes": [ 2641 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2641 ], "outputIndexes": [ 6771 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6771 ], "outputIndexes": [ 6772 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/o_proj/Linear", "inputIndexes": [ 6772 ], "outputIndexes": [ 6773 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 7437045888, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6773 ], "outputIndexes": [ 6774 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6774 ], "outputIndexes": [ 2642 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2558, 2642 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_138_output_0", "outputIndexes": [ 2643 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_27/Mul_1_output_0", "inputIndexes": [ 2643 ], "outputIndexes": [ 2644 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 7460638870, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2644 ], "outputIndexes": [ 6775 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6775 ], "outputIndexes": [ 6776 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/gate_proj/Linear", "inputIndexes": [ 6776 ], "outputIndexes": [ 6777 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 7460679830, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6777 ], "outputIndexes": [ 6778 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6778 ], "outputIndexes": [ 2645 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2645 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_27/Mul_output_0", "outputIndexes": [ 2646 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2644 ], "outputIndexes": [ 6779 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6779 ], "outputIndexes": [ 6780 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/up_proj/Linear", "inputIndexes": [ 6780 ], "outputIndexes": [ 6781 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 7534407852, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6781 ], "outputIndexes": [ 6782 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6782 ], "outputIndexes": [ 2647 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2646, 2647 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_27/Mul_output_0", "outputIndexes": [ 2648 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2648 ], "outputIndexes": [ 6783 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6783 ], "outputIndexes": [ 6784 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/down_proj/Linear", "inputIndexes": [ 6784 ], "outputIndexes": [ 6785 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 7608135874, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6785 ], "outputIndexes": [ 6786 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.27/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6786 ], "outputIndexes": [ 2649 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2643, 2649 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_139_output_0", "outputIndexes": [ 2650 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2650, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_252_output_0", "outputIndexes": [ 2651 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_28/Mul_1_output_0", "inputIndexes": [ 2651 ], "outputIndexes": [ 2652 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 7681863896, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2652 ], "outputIndexes": [ 6787 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6787 ], "outputIndexes": [ 6788 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/q_proj/Linear", "inputIndexes": [ 6788 ], "outputIndexes": [ 6789 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 7681904856, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6789 ], "outputIndexes": [ 6790 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6790 ], "outputIndexes": [ 2653 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2652 ], "main_type": "NONE", "name": "/Shape_392_output_0", "outputIndexes": [ 2654 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2654 ], "main_type": "NONE", "name": "Shape11002", "outputIndexes": [ 2655 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2654 ], "main_type": "NONE", "name": "Rank11004", "outputIndexes": [ 2656 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2656, 2656 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11006", "outputIndexes": [ 2657 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2657 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11007", "outputIndexes": [ 2658 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2657, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11009", "outputIndexes": [ 2659 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2659 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11010", "outputIndexes": [ 2660 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2655, 2658, 2660, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11013", "outputIndexes": [ 2661 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2661 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11014", "outputIndexes": [ 2662 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2662, 2662 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11016", "outputIndexes": [ 2663 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2654, 2663, 2657 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_423_output_0", "outputIndexes": [ 2664 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2664, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_896_output_0", "outputIndexes": [ 2665 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2657 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11024", "outputIndexes": [ 2666 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2659 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11027", "outputIndexes": [ 2667 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2655, 2666, 2667, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11030", "outputIndexes": [ 2668 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2668 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11031", "outputIndexes": [ 2669 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2669 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11032", "outputIndexes": [ 2670 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2670, 2669 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11033", "outputIndexes": [ 2671 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2654, 2671, 2657 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_424_output_0", "outputIndexes": [ 2672 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2672, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_897_output_0", "outputIndexes": [ 2673 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2665, 2673, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_364_output_0", "outputIndexes": [ 2674 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2653, 2674 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_253_output_0", "outputIndexes": [ 2675 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2675 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_28/Cast_output_0", "outputIndexes": [ 2676 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_28/Mul_1_output_0", "inputIndexes": [ 2676 ], "outputIndexes": [ 2677 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 7705497838, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2677, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_224_output_0", "outputIndexes": [ 2678 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2677 ], "main_type": "NONE", "name": "/Shape_394_output_0", "outputIndexes": [ 2679 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2679 ], "main_type": "NONE", "name": "Shape11152", "outputIndexes": [ 2680 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2679 ], "main_type": "NONE", "name": "Rank11154", "outputIndexes": [ 2681 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2681, 2681 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11156", "outputIndexes": [ 2682 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2682 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11157", "outputIndexes": [ 2683 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2682, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11159", "outputIndexes": [ 2684 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2684 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11160", "outputIndexes": [ 2685 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2680, 2683, 2685, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11163", "outputIndexes": [ 2686 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2686 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11164", "outputIndexes": [ 2687 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2687 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11165", "outputIndexes": [ 2688 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2688, 2687 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11166", "outputIndexes": [ 2689 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2679, 2689, 2682 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_425_output_0", "outputIndexes": [ 2690 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2690, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_84_output_0", "outputIndexes": [ 2691 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2691, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_903_output_0", "outputIndexes": [ 2692 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2677, 2692, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_113_output_0", "outputIndexes": [ 2693 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2693 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_56_output_0", "outputIndexes": [ 2694 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2691, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_902_output_0", "outputIndexes": [ 2695 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2677, 20, 2695, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_112_output_0", "outputIndexes": [ 2696 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2694, 2696 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_367_output_0", "outputIndexes": [ 2697 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2697, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_225_output_0", "outputIndexes": [ 2698 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2678, 2698 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_140_output_0", "outputIndexes": [ 2699 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2652 ], "outputIndexes": [ 6791 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6791 ], "outputIndexes": [ 6792 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/k_proj/Linear", "inputIndexes": [ 6792 ], "outputIndexes": [ 6793 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 7705498862, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6793 ], "outputIndexes": [ 6794 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6794 ], "outputIndexes": [ 2700 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2664, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_898_output_0", "outputIndexes": [ 2701 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2672, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_899_output_0", "outputIndexes": [ 2702 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2701, 2702, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_365_output_0", "outputIndexes": [ 2703 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2700, 2703 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_254_output_0", "outputIndexes": [ 2704 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2704 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_28/Cast_output_0", "outputIndexes": [ 2705 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_28/Mul_1_output_0", "inputIndexes": [ 2705 ], "outputIndexes": [ 2706 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 7708448004, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2706, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_226_output_0", "outputIndexes": [ 2707 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2706 ], "main_type": "NONE", "name": "/Shape_395_output_0", "outputIndexes": [ 2708 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2708 ], "main_type": "NONE", "name": "Shape11048", "outputIndexes": [ 2709 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2708 ], "main_type": "NONE", "name": "Rank11050", "outputIndexes": [ 2710 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2710, 2710 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11052", "outputIndexes": [ 2711 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2711 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11053", "outputIndexes": [ 2712 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2711, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11055", "outputIndexes": [ 2713 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2713 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11056", "outputIndexes": [ 2714 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2709, 2712, 2714, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11059", "outputIndexes": [ 2715 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2715 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11060", "outputIndexes": [ 2716 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2716 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11061", "outputIndexes": [ 2717 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2717, 2716 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11062", "outputIndexes": [ 2718 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2708, 2718, 2711 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_426_output_0", "outputIndexes": [ 2719 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2719, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_85_output_0", "outputIndexes": [ 2720 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2720, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_905_output_0", "outputIndexes": [ 2721 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2706, 2721, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_115_output_0", "outputIndexes": [ 2722 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2722 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_57_output_0", "outputIndexes": [ 2723 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2720, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_904_output_0", "outputIndexes": [ 2724 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2706, 20, 2724, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_114_output_0", "outputIndexes": [ 2725 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2723, 2725 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_368_output_0", "outputIndexes": [ 2726 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2726, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_227_output_0", "outputIndexes": [ 2727 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2707, 2727 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_141_output_0", "outputIndexes": [ 2728 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2652 ], "outputIndexes": [ 6795 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6795 ], "outputIndexes": [ 6796 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/v_proj/Linear", "inputIndexes": [ 6796 ], "outputIndexes": [ 6797 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 7708449028, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6797 ], "outputIndexes": [ 6798 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6798 ], "outputIndexes": [ 2729 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2664, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_900_output_0", "outputIndexes": [ 2730 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2672, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_901_output_0", "outputIndexes": [ 2731 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2730, 2731, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_366_output_0", "outputIndexes": [ 2732 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2729, 2732 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_255_output_0", "outputIndexes": [ 2733 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2699, 2728, 2733, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_260_output_0", "outputIndexes": [ 2734 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2734 ], "outputIndexes": [ 6799 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6799 ], "outputIndexes": [ 6800 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/o_proj/Linear", "inputIndexes": [ 6800 ], "outputIndexes": [ 6801 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 7711398170, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6801 ], "outputIndexes": [ 6802 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6802 ], "outputIndexes": [ 2735 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2651, 2735 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_143_output_0", "outputIndexes": [ 2736 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_28/Mul_1_output_0", "inputIndexes": [ 2736 ], "outputIndexes": [ 2737 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 7734991152, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2737 ], "outputIndexes": [ 6803 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6803 ], "outputIndexes": [ 6804 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/gate_proj/Linear", "inputIndexes": [ 6804 ], "outputIndexes": [ 6805 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 7735032112, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6805 ], "outputIndexes": [ 6806 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6806 ], "outputIndexes": [ 2738 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2738 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_28/Mul_output_0", "outputIndexes": [ 2739 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2737 ], "outputIndexes": [ 6807 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6807 ], "outputIndexes": [ 6808 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/up_proj/Linear", "inputIndexes": [ 6808 ], "outputIndexes": [ 6809 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 7808760134, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6809 ], "outputIndexes": [ 6810 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6810 ], "outputIndexes": [ 2740 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2739, 2740 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_28/Mul_output_0", "outputIndexes": [ 2741 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2741 ], "outputIndexes": [ 6811 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6811 ], "outputIndexes": [ 6812 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/down_proj/Linear", "inputIndexes": [ 6812 ], "outputIndexes": [ 6813 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 7882488156, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6813 ], "outputIndexes": [ 6814 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.28/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6814 ], "outputIndexes": [ 2742 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2736, 2742 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_144_output_0", "outputIndexes": [ 2743 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2743, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_261_output_0", "outputIndexes": [ 2744 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_29/Mul_1_output_0", "inputIndexes": [ 2744 ], "outputIndexes": [ 2745 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 7956216178, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2745 ], "outputIndexes": [ 6815 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6815 ], "outputIndexes": [ 6816 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/q_proj/Linear", "inputIndexes": [ 6816 ], "outputIndexes": [ 6817 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 7956257138, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6817 ], "outputIndexes": [ 6818 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6818 ], "outputIndexes": [ 2746 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2745 ], "main_type": "NONE", "name": "/Shape_406_output_0", "outputIndexes": [ 2747 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2747 ], "main_type": "NONE", "name": "Shape11391", "outputIndexes": [ 2748 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2747 ], "main_type": "NONE", "name": "Rank11393", "outputIndexes": [ 2749 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2749, 2749 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11395", "outputIndexes": [ 2750 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2750 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11396", "outputIndexes": [ 2751 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2750, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11398", "outputIndexes": [ 2752 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2752 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11399", "outputIndexes": [ 2753 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2748, 2751, 2753, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11402", "outputIndexes": [ 2754 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2754 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11403", "outputIndexes": [ 2755 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2755, 2755 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11405", "outputIndexes": [ 2756 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2747, 2756, 2750 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_438_output_0", "outputIndexes": [ 2757 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2757, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_928_output_0", "outputIndexes": [ 2758 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2750 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11413", "outputIndexes": [ 2759 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2752 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11416", "outputIndexes": [ 2760 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2748, 2759, 2760, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11419", "outputIndexes": [ 2761 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2761 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11420", "outputIndexes": [ 2762 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2762 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11421", "outputIndexes": [ 2763 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2763, 2762 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11422", "outputIndexes": [ 2764 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2747, 2764, 2750 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_439_output_0", "outputIndexes": [ 2765 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2765, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_929_output_0", "outputIndexes": [ 2766 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2758, 2766, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_377_output_0", "outputIndexes": [ 2767 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2746, 2767 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_262_output_0", "outputIndexes": [ 2768 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2768 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_29/Cast_output_0", "outputIndexes": [ 2769 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_29/Mul_1_output_0", "inputIndexes": [ 2769 ], "outputIndexes": [ 2770 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 7979850120, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2770, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_232_output_0", "outputIndexes": [ 2771 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2770 ], "main_type": "NONE", "name": "/Shape_408_output_0", "outputIndexes": [ 2772 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2772 ], "main_type": "NONE", "name": "Shape11541", "outputIndexes": [ 2773 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2772 ], "main_type": "NONE", "name": "Rank11543", "outputIndexes": [ 2774 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2774, 2774 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11545", "outputIndexes": [ 2775 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2775 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11546", "outputIndexes": [ 2776 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2775, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11548", "outputIndexes": [ 2777 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2777 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11549", "outputIndexes": [ 2778 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2773, 2776, 2778, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11552", "outputIndexes": [ 2779 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2779 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11553", "outputIndexes": [ 2780 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2780 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11554", "outputIndexes": [ 2781 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2781, 2780 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11555", "outputIndexes": [ 2782 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2772, 2782, 2775 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_440_output_0", "outputIndexes": [ 2783 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2783, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_87_output_0", "outputIndexes": [ 2784 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2784, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_935_output_0", "outputIndexes": [ 2785 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2770, 2785, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_117_output_0", "outputIndexes": [ 2786 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2786 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_58_output_0", "outputIndexes": [ 2787 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2784, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_934_output_0", "outputIndexes": [ 2788 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2770, 20, 2788, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_116_output_0", "outputIndexes": [ 2789 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2787, 2789 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_380_output_0", "outputIndexes": [ 2790 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2790, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_233_output_0", "outputIndexes": [ 2791 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2771, 2791 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_145_output_0", "outputIndexes": [ 2792 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2745 ], "outputIndexes": [ 6819 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6819 ], "outputIndexes": [ 6820 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/k_proj/Linear", "inputIndexes": [ 6820 ], "outputIndexes": [ 6821 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 7979851144, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6821 ], "outputIndexes": [ 6822 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6822 ], "outputIndexes": [ 2793 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2757, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_930_output_0", "outputIndexes": [ 2794 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2765, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_931_output_0", "outputIndexes": [ 2795 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2794, 2795, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_378_output_0", "outputIndexes": [ 2796 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2793, 2796 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_263_output_0", "outputIndexes": [ 2797 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2797 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_29/Cast_output_0", "outputIndexes": [ 2798 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_29/Mul_1_output_0", "inputIndexes": [ 2798 ], "outputIndexes": [ 2799 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 7982800286, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2799, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_234_output_0", "outputIndexes": [ 2800 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2799 ], "main_type": "NONE", "name": "/Shape_409_output_0", "outputIndexes": [ 2801 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2801 ], "main_type": "NONE", "name": "Shape11437", "outputIndexes": [ 2802 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2801 ], "main_type": "NONE", "name": "Rank11439", "outputIndexes": [ 2803 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2803, 2803 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11441", "outputIndexes": [ 2804 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2804 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11442", "outputIndexes": [ 2805 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2804, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11444", "outputIndexes": [ 2806 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2806 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11445", "outputIndexes": [ 2807 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2802, 2805, 2807, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11448", "outputIndexes": [ 2808 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2808 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11449", "outputIndexes": [ 2809 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2809 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11450", "outputIndexes": [ 2810 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2810, 2809 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11451", "outputIndexes": [ 2811 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2801, 2811, 2804 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_441_output_0", "outputIndexes": [ 2812 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2812, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_88_output_0", "outputIndexes": [ 2813 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2813, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_937_output_0", "outputIndexes": [ 2814 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2799, 2814, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_119_output_0", "outputIndexes": [ 2815 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2815 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_59_output_0", "outputIndexes": [ 2816 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2813, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_936_output_0", "outputIndexes": [ 2817 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2799, 20, 2817, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_118_output_0", "outputIndexes": [ 2818 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2816, 2818 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_381_output_0", "outputIndexes": [ 2819 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2819, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_235_output_0", "outputIndexes": [ 2820 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2800, 2820 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_146_output_0", "outputIndexes": [ 2821 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2745 ], "outputIndexes": [ 6823 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6823 ], "outputIndexes": [ 6824 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/v_proj/Linear", "inputIndexes": [ 6824 ], "outputIndexes": [ 6825 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 7982801310, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6825 ], "outputIndexes": [ 6826 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6826 ], "outputIndexes": [ 2822 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2757, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_932_output_0", "outputIndexes": [ 2823 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2765, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_933_output_0", "outputIndexes": [ 2824 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2823, 2824, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_379_output_0", "outputIndexes": [ 2825 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2822, 2825 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_264_output_0", "outputIndexes": [ 2826 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2792, 2821, 2826, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_269_output_0", "outputIndexes": [ 2827 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2827 ], "outputIndexes": [ 6827 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6827 ], "outputIndexes": [ 6828 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/o_proj/Linear", "inputIndexes": [ 6828 ], "outputIndexes": [ 6829 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 7985750452, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6829 ], "outputIndexes": [ 6830 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6830 ], "outputIndexes": [ 2828 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2744, 2828 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_148_output_0", "outputIndexes": [ 2829 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_29/Mul_1_output_0", "inputIndexes": [ 2829 ], "outputIndexes": [ 2830 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 8009343434, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2830 ], "outputIndexes": [ 6831 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6831 ], "outputIndexes": [ 6832 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/gate_proj/Linear", "inputIndexes": [ 6832 ], "outputIndexes": [ 6833 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 8009384394, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6833 ], "outputIndexes": [ 6834 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6834 ], "outputIndexes": [ 2831 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2831 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_29/Mul_output_0", "outputIndexes": [ 2832 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2830 ], "outputIndexes": [ 6835 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6835 ], "outputIndexes": [ 6836 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/up_proj/Linear", "inputIndexes": [ 6836 ], "outputIndexes": [ 6837 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 8083112416, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6837 ], "outputIndexes": [ 6838 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6838 ], "outputIndexes": [ 2833 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2832, 2833 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_29/Mul_output_0", "outputIndexes": [ 2834 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2834 ], "outputIndexes": [ 6839 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6839 ], "outputIndexes": [ 6840 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/down_proj/Linear", "inputIndexes": [ 6840 ], "outputIndexes": [ 6841 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 8156840438, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6841 ], "outputIndexes": [ 6842 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.29/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6842 ], "outputIndexes": [ 2835 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2829, 2835 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_149_output_0", "outputIndexes": [ 2836 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2836, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_270_output_0", "outputIndexes": [ 2837 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_30/Mul_1_output_0", "inputIndexes": [ 2837 ], "outputIndexes": [ 2838 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 8230568460, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2838 ], "outputIndexes": [ 6843 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6843 ], "outputIndexes": [ 6844 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/q_proj/Linear", "inputIndexes": [ 6844 ], "outputIndexes": [ 6845 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 8230609420, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6845 ], "outputIndexes": [ 6846 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6846 ], "outputIndexes": [ 2839 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2838 ], "main_type": "NONE", "name": "/Shape_420_output_0", "outputIndexes": [ 2840 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2840 ], "main_type": "NONE", "name": "Shape11780", "outputIndexes": [ 2841 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2840 ], "main_type": "NONE", "name": "Rank11782", "outputIndexes": [ 2842 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2842, 2842 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11784", "outputIndexes": [ 2843 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2843 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11785", "outputIndexes": [ 2844 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2843, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11787", "outputIndexes": [ 2845 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2845 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11788", "outputIndexes": [ 2846 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2841, 2844, 2846, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11791", "outputIndexes": [ 2847 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2847 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11792", "outputIndexes": [ 2848 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2848, 2848 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11794", "outputIndexes": [ 2849 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2840, 2849, 2843 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_453_output_0", "outputIndexes": [ 2850 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2850, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_960_output_0", "outputIndexes": [ 2851 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2843 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11802", "outputIndexes": [ 2852 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2845 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11805", "outputIndexes": [ 2853 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2841, 2852, 2853, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11808", "outputIndexes": [ 2854 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2854 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11809", "outputIndexes": [ 2855 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2855 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11810", "outputIndexes": [ 2856 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2856, 2855 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11811", "outputIndexes": [ 2857 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2840, 2857, 2843 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_454_output_0", "outputIndexes": [ 2858 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2858, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_961_output_0", "outputIndexes": [ 2859 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2851, 2859, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_390_output_0", "outputIndexes": [ 2860 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2839, 2860 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_271_output_0", "outputIndexes": [ 2861 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2861 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_30/Cast_output_0", "outputIndexes": [ 2862 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_30/Mul_1_output_0", "inputIndexes": [ 2862 ], "outputIndexes": [ 2863 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 8254202402, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2863, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_240_output_0", "outputIndexes": [ 2864 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2863 ], "main_type": "NONE", "name": "/Shape_422_output_0", "outputIndexes": [ 2865 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2865 ], "main_type": "NONE", "name": "Shape11930", "outputIndexes": [ 2866 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2865 ], "main_type": "NONE", "name": "Rank11932", "outputIndexes": [ 2867 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2867, 2867 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11934", "outputIndexes": [ 2868 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2868 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11935", "outputIndexes": [ 2869 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2868, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11937", "outputIndexes": [ 2870 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2870 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11938", "outputIndexes": [ 2871 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2866, 2869, 2871, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11941", "outputIndexes": [ 2872 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2872 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11942", "outputIndexes": [ 2873 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2873 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11943", "outputIndexes": [ 2874 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2874, 2873 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11944", "outputIndexes": [ 2875 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2865, 2875, 2868 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_455_output_0", "outputIndexes": [ 2876 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2876, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_90_output_0", "outputIndexes": [ 2877 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2877, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_967_output_0", "outputIndexes": [ 2878 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2863, 2878, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_121_output_0", "outputIndexes": [ 2879 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2879 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_60_output_0", "outputIndexes": [ 2880 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2877, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_966_output_0", "outputIndexes": [ 2881 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2863, 20, 2881, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_120_output_0", "outputIndexes": [ 2882 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2880, 2882 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_393_output_0", "outputIndexes": [ 2883 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2883, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_241_output_0", "outputIndexes": [ 2884 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2864, 2884 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_150_output_0", "outputIndexes": [ 2885 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2838 ], "outputIndexes": [ 6847 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6847 ], "outputIndexes": [ 6848 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/k_proj/Linear", "inputIndexes": [ 6848 ], "outputIndexes": [ 6849 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 8254203426, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6849 ], "outputIndexes": [ 6850 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6850 ], "outputIndexes": [ 2886 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2850, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_962_output_0", "outputIndexes": [ 2887 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2858, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_963_output_0", "outputIndexes": [ 2888 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2887, 2888, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_391_output_0", "outputIndexes": [ 2889 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2886, 2889 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_272_output_0", "outputIndexes": [ 2890 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2890 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_30/Cast_output_0", "outputIndexes": [ 2891 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_30/Mul_1_output_0", "inputIndexes": [ 2891 ], "outputIndexes": [ 2892 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 8257152568, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2892, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_242_output_0", "outputIndexes": [ 2893 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2892 ], "main_type": "NONE", "name": "/Shape_423_output_0", "outputIndexes": [ 2894 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2894 ], "main_type": "NONE", "name": "Shape11826", "outputIndexes": [ 2895 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2894 ], "main_type": "NONE", "name": "Rank11828", "outputIndexes": [ 2896 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2896, 2896 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11830", "outputIndexes": [ 2897 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2897 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11831", "outputIndexes": [ 2898 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2897, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11833", "outputIndexes": [ 2899 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2899 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze11834", "outputIndexes": [ 2900 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2895, 2898, 2900, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice11837", "outputIndexes": [ 2901 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2901 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze11838", "outputIndexes": [ 2902 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2902 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11839", "outputIndexes": [ 2903 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2903, 2902 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp11840", "outputIndexes": [ 2904 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2894, 2904, 2897 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_456_output_0", "outputIndexes": [ 2905 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2905, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_91_output_0", "outputIndexes": [ 2906 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2906, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_969_output_0", "outputIndexes": [ 2907 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2892, 2907, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_123_output_0", "outputIndexes": [ 2908 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2908 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_61_output_0", "outputIndexes": [ 2909 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2906, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_968_output_0", "outputIndexes": [ 2910 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2892, 20, 2910, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_122_output_0", "outputIndexes": [ 2911 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2909, 2911 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_394_output_0", "outputIndexes": [ 2912 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2912, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_243_output_0", "outputIndexes": [ 2913 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2893, 2913 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_151_output_0", "outputIndexes": [ 2914 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2838 ], "outputIndexes": [ 6851 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6851 ], "outputIndexes": [ 6852 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/v_proj/Linear", "inputIndexes": [ 6852 ], "outputIndexes": [ 6853 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 8257153592, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6853 ], "outputIndexes": [ 6854 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6854 ], "outputIndexes": [ 2915 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2850, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_964_output_0", "outputIndexes": [ 2916 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2858, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_965_output_0", "outputIndexes": [ 2917 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2916, 2917, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_392_output_0", "outputIndexes": [ 2918 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2915, 2918 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_273_output_0", "outputIndexes": [ 2919 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2885, 2914, 2919, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_278_output_0", "outputIndexes": [ 2920 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2920 ], "outputIndexes": [ 6855 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6855 ], "outputIndexes": [ 6856 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/o_proj/Linear", "inputIndexes": [ 6856 ], "outputIndexes": [ 6857 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 8260102734, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6857 ], "outputIndexes": [ 6858 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6858 ], "outputIndexes": [ 2921 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2837, 2921 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_153_output_0", "outputIndexes": [ 2922 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_30/Mul_1_output_0", "inputIndexes": [ 2922 ], "outputIndexes": [ 2923 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 8283695716, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2923 ], "outputIndexes": [ 6859 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6859 ], "outputIndexes": [ 6860 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/gate_proj/Linear", "inputIndexes": [ 6860 ], "outputIndexes": [ 6861 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 8283736676, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6861 ], "outputIndexes": [ 6862 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6862 ], "outputIndexes": [ 2924 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2924 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_30/Mul_output_0", "outputIndexes": [ 2925 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2923 ], "outputIndexes": [ 6863 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6863 ], "outputIndexes": [ 6864 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/up_proj/Linear", "inputIndexes": [ 6864 ], "outputIndexes": [ 6865 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 8357464698, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6865 ], "outputIndexes": [ 6866 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6866 ], "outputIndexes": [ 2926 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2925, 2926 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_30/Mul_output_0", "outputIndexes": [ 2927 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2927 ], "outputIndexes": [ 6867 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6867 ], "outputIndexes": [ 6868 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/down_proj/Linear", "inputIndexes": [ 6868 ], "outputIndexes": [ 6869 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 8431192720, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6869 ], "outputIndexes": [ 6870 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.30/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6870 ], "outputIndexes": [ 2928 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2922, 2928 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_154_output_0", "outputIndexes": [ 2929 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2929, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_279_output_0", "outputIndexes": [ 2930 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_31/Mul_1_output_0", "inputIndexes": [ 2930 ], "outputIndexes": [ 2931 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 8504920742, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2931 ], "outputIndexes": [ 6871 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6871 ], "outputIndexes": [ 6872 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/q_proj/Linear", "inputIndexes": [ 6872 ], "outputIndexes": [ 6873 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 8504961702, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6873 ], "outputIndexes": [ 6874 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6874 ], "outputIndexes": [ 2932 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2931 ], "main_type": "NONE", "name": "/Shape_434_output_0", "outputIndexes": [ 2933 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2933 ], "main_type": "NONE", "name": "Shape12169", "outputIndexes": [ 2934 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2933 ], "main_type": "NONE", "name": "Rank12171", "outputIndexes": [ 2935 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2935, 2935 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12173", "outputIndexes": [ 2936 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2936 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12174", "outputIndexes": [ 2937 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2936, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12176", "outputIndexes": [ 2938 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2938 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12177", "outputIndexes": [ 2939 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2934, 2937, 2939, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12180", "outputIndexes": [ 2940 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2940 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12181", "outputIndexes": [ 2941 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2941, 2941 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12183", "outputIndexes": [ 2942 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2933, 2942, 2936 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_468_output_0", "outputIndexes": [ 2943 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2943, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_992_output_0", "outputIndexes": [ 2944 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2936 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12191", "outputIndexes": [ 2945 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2938 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12194", "outputIndexes": [ 2946 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2934, 2945, 2946, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12197", "outputIndexes": [ 2947 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2947 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12198", "outputIndexes": [ 2948 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 2948 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12199", "outputIndexes": [ 2949 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2949, 2948 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12200", "outputIndexes": [ 2950 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2933, 2950, 2936 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_469_output_0", "outputIndexes": [ 2951 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2951, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_993_output_0", "outputIndexes": [ 2952 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2944, 2952, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_403_output_0", "outputIndexes": [ 2953 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2932, 2953 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_280_output_0", "outputIndexes": [ 2954 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2954 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_31/Cast_output_0", "outputIndexes": [ 2955 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_31/Mul_1_output_0", "inputIndexes": [ 2955 ], "outputIndexes": [ 2956 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 8528554684, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2956, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_248_output_0", "outputIndexes": [ 2957 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2956 ], "main_type": "NONE", "name": "/Shape_436_output_0", "outputIndexes": [ 2958 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2958 ], "main_type": "NONE", "name": "Shape12319", "outputIndexes": [ 2959 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2958 ], "main_type": "NONE", "name": "Rank12321", "outputIndexes": [ 2960 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2960, 2960 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12323", "outputIndexes": [ 2961 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2961 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12324", "outputIndexes": [ 2962 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2961, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12326", "outputIndexes": [ 2963 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2963 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12327", "outputIndexes": [ 2964 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2959, 2962, 2964, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12330", "outputIndexes": [ 2965 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2965 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12331", "outputIndexes": [ 2966 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2966 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12332", "outputIndexes": [ 2967 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2967, 2966 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12333", "outputIndexes": [ 2968 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2958, 2968, 2961 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_470_output_0", "outputIndexes": [ 2969 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2969, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_93_output_0", "outputIndexes": [ 2970 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2970, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_999_output_0", "outputIndexes": [ 2971 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2956, 2971, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_125_output_0", "outputIndexes": [ 2972 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2972 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_62_output_0", "outputIndexes": [ 2973 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2970, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_998_output_0", "outputIndexes": [ 2974 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2956, 20, 2974, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_124_output_0", "outputIndexes": [ 2975 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2973, 2975 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_406_output_0", "outputIndexes": [ 2976 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2976, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_249_output_0", "outputIndexes": [ 2977 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2957, 2977 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_155_output_0", "outputIndexes": [ 2978 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2931 ], "outputIndexes": [ 6875 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6875 ], "outputIndexes": [ 6876 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/k_proj/Linear", "inputIndexes": [ 6876 ], "outputIndexes": [ 6877 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 8528555708, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6877 ], "outputIndexes": [ 6878 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6878 ], "outputIndexes": [ 2979 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2943, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_994_output_0", "outputIndexes": [ 2980 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2951, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_995_output_0", "outputIndexes": [ 2981 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2980, 2981, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_404_output_0", "outputIndexes": [ 2982 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2979, 2982 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_281_output_0", "outputIndexes": [ 2983 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2983 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_31/Cast_output_0", "outputIndexes": [ 2984 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_31/Mul_1_output_0", "inputIndexes": [ 2984 ], "outputIndexes": [ 2985 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 8531504850, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2985, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_250_output_0", "outputIndexes": [ 2986 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2985 ], "main_type": "NONE", "name": "/Shape_437_output_0", "outputIndexes": [ 2987 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2987 ], "main_type": "NONE", "name": "Shape12215", "outputIndexes": [ 2988 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 2987 ], "main_type": "NONE", "name": "Rank12217", "outputIndexes": [ 2989 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2989, 2989 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12219", "outputIndexes": [ 2990 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2990 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12220", "outputIndexes": [ 2991 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2990, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12222", "outputIndexes": [ 2992 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2992 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12223", "outputIndexes": [ 2993 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2988, 2991, 2993, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12226", "outputIndexes": [ 2994 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2994 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12227", "outputIndexes": [ 2995 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 2995 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12228", "outputIndexes": [ 2996 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2996, 2995 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12229", "outputIndexes": [ 2997 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2987, 2997, 2990 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_471_output_0", "outputIndexes": [ 2998 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2998, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_94_output_0", "outputIndexes": [ 2999 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2999, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1001_output_0", "outputIndexes": [ 3000 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2985, 3000, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_127_output_0", "outputIndexes": [ 3001 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3001 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_63_output_0", "outputIndexes": [ 3002 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2999, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1000_output_0", "outputIndexes": [ 3003 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2985, 20, 3003, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_126_output_0", "outputIndexes": [ 3004 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3002, 3004 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_407_output_0", "outputIndexes": [ 3005 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3005, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_251_output_0", "outputIndexes": [ 3006 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2986, 3006 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_156_output_0", "outputIndexes": [ 3007 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 2931 ], "outputIndexes": [ 6879 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6879 ], "outputIndexes": [ 6880 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/v_proj/Linear", "inputIndexes": [ 6880 ], "outputIndexes": [ 6881 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 8531505874, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6881 ], "outputIndexes": [ 6882 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6882 ], "outputIndexes": [ 3008 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2943, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_996_output_0", "outputIndexes": [ 3009 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2951, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_997_output_0", "outputIndexes": [ 3010 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3009, 3010, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_405_output_0", "outputIndexes": [ 3011 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3008, 3011 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_282_output_0", "outputIndexes": [ 3012 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2978, 3007, 3012, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_287_output_0", "outputIndexes": [ 3013 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3013 ], "outputIndexes": [ 6883 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6883 ], "outputIndexes": [ 6884 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/o_proj/Linear", "inputIndexes": [ 6884 ], "outputIndexes": [ 6885 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 8534455016, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6885 ], "outputIndexes": [ 6886 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6886 ], "outputIndexes": [ 3014 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 2930, 3014 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_158_output_0", "outputIndexes": [ 3015 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_31/Mul_1_output_0", "inputIndexes": [ 3015 ], "outputIndexes": [ 3016 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 8558047998, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3016 ], "outputIndexes": [ 6887 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6887 ], "outputIndexes": [ 6888 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/gate_proj/Linear", "inputIndexes": [ 6888 ], "outputIndexes": [ 6889 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 8558088958, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6889 ], "outputIndexes": [ 6890 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6890 ], "outputIndexes": [ 3017 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3017 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_31/Mul_output_0", "outputIndexes": [ 3018 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3016 ], "outputIndexes": [ 6891 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6891 ], "outputIndexes": [ 6892 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/up_proj/Linear", "inputIndexes": [ 6892 ], "outputIndexes": [ 6893 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 8631816980, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6893 ], "outputIndexes": [ 6894 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6894 ], "outputIndexes": [ 3019 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3018, 3019 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_31/Mul_output_0", "outputIndexes": [ 3020 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3020 ], "outputIndexes": [ 6895 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6895 ], "outputIndexes": [ 6896 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/down_proj/Linear", "inputIndexes": [ 6896 ], "outputIndexes": [ 6897 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 8705545002, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6897 ], "outputIndexes": [ 6898 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.31/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6898 ], "outputIndexes": [ 3021 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3015, 3021 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_159_output_0", "outputIndexes": [ 3022 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3022, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_288_output_0", "outputIndexes": [ 3023 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_32/Mul_1_output_0", "inputIndexes": [ 3023 ], "outputIndexes": [ 3024 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 8779273024, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3024 ], "outputIndexes": [ 6899 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6899 ], "outputIndexes": [ 6900 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/q_proj/Linear", "inputIndexes": [ 6900 ], "outputIndexes": [ 6901 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 8779313984, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6901 ], "outputIndexes": [ 6902 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6902 ], "outputIndexes": [ 3025 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3024 ], "main_type": "NONE", "name": "/Shape_448_output_0", "outputIndexes": [ 3026 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3026 ], "main_type": "NONE", "name": "Shape12558", "outputIndexes": [ 3027 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3026 ], "main_type": "NONE", "name": "Rank12560", "outputIndexes": [ 3028 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3028, 3028 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12562", "outputIndexes": [ 3029 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3029 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12563", "outputIndexes": [ 3030 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3029, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12565", "outputIndexes": [ 3031 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3031 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12566", "outputIndexes": [ 3032 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3027, 3030, 3032, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12569", "outputIndexes": [ 3033 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3033 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12570", "outputIndexes": [ 3034 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3034, 3034 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12572", "outputIndexes": [ 3035 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3026, 3035, 3029 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_483_output_0", "outputIndexes": [ 3036 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3036, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1024_output_0", "outputIndexes": [ 3037 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3029 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12580", "outputIndexes": [ 3038 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3031 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12583", "outputIndexes": [ 3039 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3027, 3038, 3039, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12586", "outputIndexes": [ 3040 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3040 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12587", "outputIndexes": [ 3041 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 3041 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12588", "outputIndexes": [ 3042 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3042, 3041 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12589", "outputIndexes": [ 3043 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3026, 3043, 3029 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_484_output_0", "outputIndexes": [ 3044 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3044, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1025_output_0", "outputIndexes": [ 3045 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3037, 3045, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_416_output_0", "outputIndexes": [ 3046 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3025, 3046 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_289_output_0", "outputIndexes": [ 3047 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3047 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_32/Cast_output_0", "outputIndexes": [ 3048 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_32/Mul_1_output_0", "inputIndexes": [ 3048 ], "outputIndexes": [ 3049 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 8802906966, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3049, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_256_output_0", "outputIndexes": [ 3050 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3049 ], "main_type": "NONE", "name": "/Shape_450_output_0", "outputIndexes": [ 3051 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3051 ], "main_type": "NONE", "name": "Shape12708", "outputIndexes": [ 3052 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3051 ], "main_type": "NONE", "name": "Rank12710", "outputIndexes": [ 3053 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3053, 3053 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12712", "outputIndexes": [ 3054 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3054 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12713", "outputIndexes": [ 3055 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3054, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12715", "outputIndexes": [ 3056 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3056 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12716", "outputIndexes": [ 3057 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3052, 3055, 3057, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12719", "outputIndexes": [ 3058 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3058 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12720", "outputIndexes": [ 3059 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3059 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12721", "outputIndexes": [ 3060 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3060, 3059 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12722", "outputIndexes": [ 3061 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3051, 3061, 3054 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_485_output_0", "outputIndexes": [ 3062 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3062, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_96_output_0", "outputIndexes": [ 3063 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3063, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1031_output_0", "outputIndexes": [ 3064 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3049, 3064, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_129_output_0", "outputIndexes": [ 3065 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3065 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_64_output_0", "outputIndexes": [ 3066 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3063, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1030_output_0", "outputIndexes": [ 3067 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3049, 20, 3067, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_128_output_0", "outputIndexes": [ 3068 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3066, 3068 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_419_output_0", "outputIndexes": [ 3069 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3069, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_257_output_0", "outputIndexes": [ 3070 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3050, 3070 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_160_output_0", "outputIndexes": [ 3071 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3024 ], "outputIndexes": [ 6903 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6903 ], "outputIndexes": [ 6904 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/k_proj/Linear", "inputIndexes": [ 6904 ], "outputIndexes": [ 6905 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 8802907990, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6905 ], "outputIndexes": [ 6906 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6906 ], "outputIndexes": [ 3072 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3036, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1026_output_0", "outputIndexes": [ 3073 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3044, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1027_output_0", "outputIndexes": [ 3074 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3073, 3074, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_417_output_0", "outputIndexes": [ 3075 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3072, 3075 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_290_output_0", "outputIndexes": [ 3076 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3076 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_32/Cast_output_0", "outputIndexes": [ 3077 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_32/Mul_1_output_0", "inputIndexes": [ 3077 ], "outputIndexes": [ 3078 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 8805857132, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3078, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_258_output_0", "outputIndexes": [ 3079 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3078 ], "main_type": "NONE", "name": "/Shape_451_output_0", "outputIndexes": [ 3080 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3080 ], "main_type": "NONE", "name": "Shape12604", "outputIndexes": [ 3081 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3080 ], "main_type": "NONE", "name": "Rank12606", "outputIndexes": [ 3082 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3082, 3082 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12608", "outputIndexes": [ 3083 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3083 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12609", "outputIndexes": [ 3084 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3083, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12611", "outputIndexes": [ 3085 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3085 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12612", "outputIndexes": [ 3086 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3081, 3084, 3086, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12615", "outputIndexes": [ 3087 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3087 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12616", "outputIndexes": [ 3088 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3088 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12617", "outputIndexes": [ 3089 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3089, 3088 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12618", "outputIndexes": [ 3090 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3080, 3090, 3083 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_486_output_0", "outputIndexes": [ 3091 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3091, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_97_output_0", "outputIndexes": [ 3092 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3092, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1033_output_0", "outputIndexes": [ 3093 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3078, 3093, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_131_output_0", "outputIndexes": [ 3094 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3094 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_65_output_0", "outputIndexes": [ 3095 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3092, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1032_output_0", "outputIndexes": [ 3096 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3078, 20, 3096, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_130_output_0", "outputIndexes": [ 3097 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3095, 3097 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_420_output_0", "outputIndexes": [ 3098 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3098, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_259_output_0", "outputIndexes": [ 3099 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3079, 3099 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_161_output_0", "outputIndexes": [ 3100 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3024 ], "outputIndexes": [ 6907 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6907 ], "outputIndexes": [ 6908 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/v_proj/Linear", "inputIndexes": [ 6908 ], "outputIndexes": [ 6909 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 8805858156, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6909 ], "outputIndexes": [ 6910 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6910 ], "outputIndexes": [ 3101 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3036, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1028_output_0", "outputIndexes": [ 3102 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3044, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1029_output_0", "outputIndexes": [ 3103 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3102, 3103, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_418_output_0", "outputIndexes": [ 3104 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3101, 3104 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_291_output_0", "outputIndexes": [ 3105 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3071, 3100, 3105, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_296_output_0", "outputIndexes": [ 3106 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3106 ], "outputIndexes": [ 6911 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6911 ], "outputIndexes": [ 6912 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/o_proj/Linear", "inputIndexes": [ 6912 ], "outputIndexes": [ 6913 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 8808807298, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6913 ], "outputIndexes": [ 6914 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6914 ], "outputIndexes": [ 3107 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3023, 3107 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_163_output_0", "outputIndexes": [ 3108 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_32/Mul_1_output_0", "inputIndexes": [ 3108 ], "outputIndexes": [ 3109 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 8832400280, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3109 ], "outputIndexes": [ 6915 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6915 ], "outputIndexes": [ 6916 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/gate_proj/Linear", "inputIndexes": [ 6916 ], "outputIndexes": [ 6917 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 8832441240, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6917 ], "outputIndexes": [ 6918 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6918 ], "outputIndexes": [ 3110 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3110 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_32/Mul_output_0", "outputIndexes": [ 3111 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3109 ], "outputIndexes": [ 6919 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6919 ], "outputIndexes": [ 6920 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/up_proj/Linear", "inputIndexes": [ 6920 ], "outputIndexes": [ 6921 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 8906169262, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6921 ], "outputIndexes": [ 6922 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6922 ], "outputIndexes": [ 3112 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3111, 3112 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_32/Mul_output_0", "outputIndexes": [ 3113 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3113 ], "outputIndexes": [ 6923 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6923 ], "outputIndexes": [ 6924 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/down_proj/Linear", "inputIndexes": [ 6924 ], "outputIndexes": [ 6925 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 8979897284, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6925 ], "outputIndexes": [ 6926 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.32/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6926 ], "outputIndexes": [ 3114 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3108, 3114 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_164_output_0", "outputIndexes": [ 3115 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3115, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_297_output_0", "outputIndexes": [ 3116 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_33/Mul_1_output_0", "inputIndexes": [ 3116 ], "outputIndexes": [ 3117 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 9053625306, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3117 ], "outputIndexes": [ 6927 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6927 ], "outputIndexes": [ 6928 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/q_proj/Linear", "inputIndexes": [ 6928 ], "outputIndexes": [ 6929 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 9053666266, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6929 ], "outputIndexes": [ 6930 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6930 ], "outputIndexes": [ 3118 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3117 ], "main_type": "NONE", "name": "/Shape_462_output_0", "outputIndexes": [ 3119 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3119 ], "main_type": "NONE", "name": "Shape12947", "outputIndexes": [ 3120 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3119 ], "main_type": "NONE", "name": "Rank12949", "outputIndexes": [ 3121 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3121, 3121 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12951", "outputIndexes": [ 3122 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3122 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12952", "outputIndexes": [ 3123 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3122, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12954", "outputIndexes": [ 3124 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3124 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12955", "outputIndexes": [ 3125 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3120, 3123, 3125, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12958", "outputIndexes": [ 3126 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3126 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12959", "outputIndexes": [ 3127 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3127, 3127 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12961", "outputIndexes": [ 3128 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3119, 3128, 3122 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_498_output_0", "outputIndexes": [ 3129 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3129, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1056_output_0", "outputIndexes": [ 3130 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3122 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12969", "outputIndexes": [ 3131 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3124 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12972", "outputIndexes": [ 3132 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3120, 3131, 3132, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice12975", "outputIndexes": [ 3133 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3133 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze12976", "outputIndexes": [ 3134 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 3134 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12977", "outputIndexes": [ 3135 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3135, 3134 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12978", "outputIndexes": [ 3136 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3119, 3136, 3122 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_499_output_0", "outputIndexes": [ 3137 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3137, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1057_output_0", "outputIndexes": [ 3138 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3130, 3138, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_429_output_0", "outputIndexes": [ 3139 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3118, 3139 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_298_output_0", "outputIndexes": [ 3140 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3140 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_33/Cast_output_0", "outputIndexes": [ 3141 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_33/Mul_1_output_0", "inputIndexes": [ 3141 ], "outputIndexes": [ 3142 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 9077259248, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3142, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_264_output_0", "outputIndexes": [ 3143 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3142 ], "main_type": "NONE", "name": "/Shape_464_output_0", "outputIndexes": [ 3144 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3144 ], "main_type": "NONE", "name": "Shape13097", "outputIndexes": [ 3145 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3144 ], "main_type": "NONE", "name": "Rank13099", "outputIndexes": [ 3146 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3146, 3146 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13101", "outputIndexes": [ 3147 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3147 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13102", "outputIndexes": [ 3148 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3147, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13104", "outputIndexes": [ 3149 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3149 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13105", "outputIndexes": [ 3150 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3145, 3148, 3150, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice13108", "outputIndexes": [ 3151 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3151 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze13109", "outputIndexes": [ 3152 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3152 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13110", "outputIndexes": [ 3153 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3153, 3152 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13111", "outputIndexes": [ 3154 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3144, 3154, 3147 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_500_output_0", "outputIndexes": [ 3155 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3155, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_99_output_0", "outputIndexes": [ 3156 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3156, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1063_output_0", "outputIndexes": [ 3157 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3142, 3157, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_133_output_0", "outputIndexes": [ 3158 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3158 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_66_output_0", "outputIndexes": [ 3159 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3156, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1062_output_0", "outputIndexes": [ 3160 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3142, 20, 3160, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_132_output_0", "outputIndexes": [ 3161 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3159, 3161 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_432_output_0", "outputIndexes": [ 3162 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3162, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_265_output_0", "outputIndexes": [ 3163 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3143, 3163 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_165_output_0", "outputIndexes": [ 3164 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3117 ], "outputIndexes": [ 6931 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6931 ], "outputIndexes": [ 6932 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/k_proj/Linear", "inputIndexes": [ 6932 ], "outputIndexes": [ 6933 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 9077260272, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6933 ], "outputIndexes": [ 6934 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6934 ], "outputIndexes": [ 3165 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3129, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1058_output_0", "outputIndexes": [ 3166 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3137, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1059_output_0", "outputIndexes": [ 3167 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3166, 3167, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_430_output_0", "outputIndexes": [ 3168 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3165, 3168 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_299_output_0", "outputIndexes": [ 3169 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3169 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_33/Cast_output_0", "outputIndexes": [ 3170 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_33/Mul_1_output_0", "inputIndexes": [ 3170 ], "outputIndexes": [ 3171 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 9080209414, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3171, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_266_output_0", "outputIndexes": [ 3172 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3171 ], "main_type": "NONE", "name": "/Shape_465_output_0", "outputIndexes": [ 3173 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3173 ], "main_type": "NONE", "name": "Shape12993", "outputIndexes": [ 3174 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3173 ], "main_type": "NONE", "name": "Rank12995", "outputIndexes": [ 3175 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3175, 3175 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp12997", "outputIndexes": [ 3176 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3176 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze12998", "outputIndexes": [ 3177 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3176, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13000", "outputIndexes": [ 3178 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3178 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13001", "outputIndexes": [ 3179 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3174, 3177, 3179, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice13004", "outputIndexes": [ 3180 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3180 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze13005", "outputIndexes": [ 3181 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3181 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13006", "outputIndexes": [ 3182 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3182, 3181 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13007", "outputIndexes": [ 3183 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3173, 3183, 3176 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_501_output_0", "outputIndexes": [ 3184 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3184, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_100_output_0", "outputIndexes": [ 3185 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3185, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1065_output_0", "outputIndexes": [ 3186 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3171, 3186, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_135_output_0", "outputIndexes": [ 3187 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3187 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_67_output_0", "outputIndexes": [ 3188 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3185, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1064_output_0", "outputIndexes": [ 3189 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3171, 20, 3189, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_134_output_0", "outputIndexes": [ 3190 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3188, 3190 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_433_output_0", "outputIndexes": [ 3191 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3191, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_267_output_0", "outputIndexes": [ 3192 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3172, 3192 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_166_output_0", "outputIndexes": [ 3193 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3117 ], "outputIndexes": [ 6935 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6935 ], "outputIndexes": [ 6936 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/v_proj/Linear", "inputIndexes": [ 6936 ], "outputIndexes": [ 6937 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 9080210438, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6937 ], "outputIndexes": [ 6938 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6938 ], "outputIndexes": [ 3194 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3129, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1060_output_0", "outputIndexes": [ 3195 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3137, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1061_output_0", "outputIndexes": [ 3196 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3195, 3196, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_431_output_0", "outputIndexes": [ 3197 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3194, 3197 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_300_output_0", "outputIndexes": [ 3198 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3164, 3193, 3198, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_305_output_0", "outputIndexes": [ 3199 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3199 ], "outputIndexes": [ 6939 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6939 ], "outputIndexes": [ 6940 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/o_proj/Linear", "inputIndexes": [ 6940 ], "outputIndexes": [ 6941 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 9083159580, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6941 ], "outputIndexes": [ 6942 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6942 ], "outputIndexes": [ 3200 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3116, 3200 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_168_output_0", "outputIndexes": [ 3201 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_33/Mul_1_output_0", "inputIndexes": [ 3201 ], "outputIndexes": [ 3202 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 9106752562, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3202 ], "outputIndexes": [ 6943 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6943 ], "outputIndexes": [ 6944 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/gate_proj/Linear", "inputIndexes": [ 6944 ], "outputIndexes": [ 6945 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 9106793522, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6945 ], "outputIndexes": [ 6946 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6946 ], "outputIndexes": [ 3203 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3203 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_33/Mul_output_0", "outputIndexes": [ 3204 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3202 ], "outputIndexes": [ 6947 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6947 ], "outputIndexes": [ 6948 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/up_proj/Linear", "inputIndexes": [ 6948 ], "outputIndexes": [ 6949 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 9180521544, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6949 ], "outputIndexes": [ 6950 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6950 ], "outputIndexes": [ 3205 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3204, 3205 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_33/Mul_output_0", "outputIndexes": [ 3206 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3206 ], "outputIndexes": [ 6951 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6951 ], "outputIndexes": [ 6952 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/down_proj/Linear", "inputIndexes": [ 6952 ], "outputIndexes": [ 6953 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 9254249566, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6953 ], "outputIndexes": [ 6954 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.33/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6954 ], "outputIndexes": [ 3207 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3201, 3207 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_169_output_0", "outputIndexes": [ 3208 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3208, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_306_output_0", "outputIndexes": [ 3209 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_34/Mul_1_output_0", "inputIndexes": [ 3209 ], "outputIndexes": [ 3210 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 9327977588, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3210 ], "outputIndexes": [ 6955 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6955 ], "outputIndexes": [ 6956 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/q_proj/Linear", "inputIndexes": [ 6956 ], "outputIndexes": [ 6957 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 9328018548, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6957 ], "outputIndexes": [ 6958 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6958 ], "outputIndexes": [ 3211 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3210 ], "main_type": "NONE", "name": "/Shape_476_output_0", "outputIndexes": [ 3212 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3212 ], "main_type": "NONE", "name": "Shape13336", "outputIndexes": [ 3213 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3212 ], "main_type": "NONE", "name": "Rank13338", "outputIndexes": [ 3214 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3214, 3214 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13340", "outputIndexes": [ 3215 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3215 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13341", "outputIndexes": [ 3216 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3215, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13343", "outputIndexes": [ 3217 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3217 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13344", "outputIndexes": [ 3218 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3213, 3216, 3218, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice13347", "outputIndexes": [ 3219 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3219 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze13348", "outputIndexes": [ 3220 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3220, 3220 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13350", "outputIndexes": [ 3221 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3212, 3221, 3215 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_513_output_0", "outputIndexes": [ 3222 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3222, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1088_output_0", "outputIndexes": [ 3223 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3215 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13358", "outputIndexes": [ 3224 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3217 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13361", "outputIndexes": [ 3225 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3213, 3224, 3225, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice13364", "outputIndexes": [ 3226 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3226 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze13365", "outputIndexes": [ 3227 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 3227 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13366", "outputIndexes": [ 3228 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3228, 3227 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13367", "outputIndexes": [ 3229 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3212, 3229, 3215 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_514_output_0", "outputIndexes": [ 3230 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3230, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1089_output_0", "outputIndexes": [ 3231 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3223, 3231, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_442_output_0", "outputIndexes": [ 3232 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3211, 3232 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_307_output_0", "outputIndexes": [ 3233 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3233 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_34/Cast_output_0", "outputIndexes": [ 3234 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_34/Mul_1_output_0", "inputIndexes": [ 3234 ], "outputIndexes": [ 3235 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 9351611530, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3235, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_272_output_0", "outputIndexes": [ 3236 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3235 ], "main_type": "NONE", "name": "/Shape_478_output_0", "outputIndexes": [ 3237 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3237 ], "main_type": "NONE", "name": "Shape13486", "outputIndexes": [ 3238 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3237 ], "main_type": "NONE", "name": "Rank13488", "outputIndexes": [ 3239 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3239, 3239 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13490", "outputIndexes": [ 3240 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3240 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13491", "outputIndexes": [ 3241 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3240, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13493", "outputIndexes": [ 3242 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3242 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13494", "outputIndexes": [ 3243 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3238, 3241, 3243, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice13497", "outputIndexes": [ 3244 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3244 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze13498", "outputIndexes": [ 3245 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3245 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13499", "outputIndexes": [ 3246 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3246, 3245 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13500", "outputIndexes": [ 3247 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3237, 3247, 3240 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_515_output_0", "outputIndexes": [ 3248 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3248, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_102_output_0", "outputIndexes": [ 3249 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3249, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1095_output_0", "outputIndexes": [ 3250 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3235, 3250, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_137_output_0", "outputIndexes": [ 3251 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3251 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_68_output_0", "outputIndexes": [ 3252 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3249, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1094_output_0", "outputIndexes": [ 3253 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3235, 20, 3253, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_136_output_0", "outputIndexes": [ 3254 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3252, 3254 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_445_output_0", "outputIndexes": [ 3255 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3255, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_273_output_0", "outputIndexes": [ 3256 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3236, 3256 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_170_output_0", "outputIndexes": [ 3257 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3210 ], "outputIndexes": [ 6959 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6959 ], "outputIndexes": [ 6960 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/k_proj/Linear", "inputIndexes": [ 6960 ], "outputIndexes": [ 6961 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 9351612554, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6961 ], "outputIndexes": [ 6962 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6962 ], "outputIndexes": [ 3258 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3222, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1090_output_0", "outputIndexes": [ 3259 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3230, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1091_output_0", "outputIndexes": [ 3260 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3259, 3260, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_443_output_0", "outputIndexes": [ 3261 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3258, 3261 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_308_output_0", "outputIndexes": [ 3262 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3262 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_34/Cast_output_0", "outputIndexes": [ 3263 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_34/Mul_1_output_0", "inputIndexes": [ 3263 ], "outputIndexes": [ 3264 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 9354561696, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3264, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_274_output_0", "outputIndexes": [ 3265 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3264 ], "main_type": "NONE", "name": "/Shape_479_output_0", "outputIndexes": [ 3266 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3266 ], "main_type": "NONE", "name": "Shape13382", "outputIndexes": [ 3267 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3266 ], "main_type": "NONE", "name": "Rank13384", "outputIndexes": [ 3268 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3268, 3268 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13386", "outputIndexes": [ 3269 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3269 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13387", "outputIndexes": [ 3270 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3269, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13389", "outputIndexes": [ 3271 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3271 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13390", "outputIndexes": [ 3272 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3267, 3270, 3272, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice13393", "outputIndexes": [ 3273 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3273 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze13394", "outputIndexes": [ 3274 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3274 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13395", "outputIndexes": [ 3275 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3275, 3274 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13396", "outputIndexes": [ 3276 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3266, 3276, 3269 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_516_output_0", "outputIndexes": [ 3277 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3277, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_103_output_0", "outputIndexes": [ 3278 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3278, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1097_output_0", "outputIndexes": [ 3279 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3264, 3279, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_139_output_0", "outputIndexes": [ 3280 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3280 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_69_output_0", "outputIndexes": [ 3281 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3278, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1096_output_0", "outputIndexes": [ 3282 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3264, 20, 3282, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_138_output_0", "outputIndexes": [ 3283 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3281, 3283 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_446_output_0", "outputIndexes": [ 3284 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3284, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_275_output_0", "outputIndexes": [ 3285 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3265, 3285 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_171_output_0", "outputIndexes": [ 3286 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3210 ], "outputIndexes": [ 6963 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6963 ], "outputIndexes": [ 6964 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/v_proj/Linear", "inputIndexes": [ 6964 ], "outputIndexes": [ 6965 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 9354562720, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6965 ], "outputIndexes": [ 6966 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6966 ], "outputIndexes": [ 3287 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3222, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1092_output_0", "outputIndexes": [ 3288 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3230, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1093_output_0", "outputIndexes": [ 3289 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3288, 3289, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_444_output_0", "outputIndexes": [ 3290 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3287, 3290 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_309_output_0", "outputIndexes": [ 3291 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3257, 3286, 3291, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_314_output_0", "outputIndexes": [ 3292 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3292 ], "outputIndexes": [ 6967 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6967 ], "outputIndexes": [ 6968 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/o_proj/Linear", "inputIndexes": [ 6968 ], "outputIndexes": [ 6969 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 9357511862, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6969 ], "outputIndexes": [ 6970 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6970 ], "outputIndexes": [ 3293 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3209, 3293 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_173_output_0", "outputIndexes": [ 3294 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_34/Mul_1_output_0", "inputIndexes": [ 3294 ], "outputIndexes": [ 3295 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 9381104844, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3295 ], "outputIndexes": [ 6971 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6971 ], "outputIndexes": [ 6972 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/gate_proj/Linear", "inputIndexes": [ 6972 ], "outputIndexes": [ 6973 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 9381145804, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 6973 ], "outputIndexes": [ 6974 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6974 ], "outputIndexes": [ 3296 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3296 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_34/Mul_output_0", "outputIndexes": [ 3297 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3295 ], "outputIndexes": [ 6975 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 6975 ], "outputIndexes": [ 6976 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/up_proj/Linear", "inputIndexes": [ 6976 ], "outputIndexes": [ 6977 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 9454873826, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 6977 ], "outputIndexes": [ 6978 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6978 ], "outputIndexes": [ 3298 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3297, 3298 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_34/Mul_output_0", "outputIndexes": [ 3299 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3299 ], "outputIndexes": [ 6979 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 6979 ], "outputIndexes": [ 6980 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/down_proj/Linear", "inputIndexes": [ 6980 ], "outputIndexes": [ 6981 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 9528601848, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 6981 ], "outputIndexes": [ 6982 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.34/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6982 ], "outputIndexes": [ 3300 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3294, 3300 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_174_output_0", "outputIndexes": [ 3301 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3301, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_315_output_0", "outputIndexes": [ 3302 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_35/Mul_1_output_0", "inputIndexes": [ 3302 ], "outputIndexes": [ 3303 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 9602329870, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3303 ], "outputIndexes": [ 6983 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 6983 ], "outputIndexes": [ 6984 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/q_proj/Linear", "inputIndexes": [ 6984 ], "outputIndexes": [ 6985 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 9602370830, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 6985 ], "outputIndexes": [ 6986 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6986 ], "outputIndexes": [ 3304 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3303 ], "main_type": "NONE", "name": "/Shape_490_output_0", "outputIndexes": [ 3305 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3305 ], "main_type": "NONE", "name": "Shape13725", "outputIndexes": [ 3306 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3305 ], "main_type": "NONE", "name": "Rank13727", "outputIndexes": [ 3307 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3307, 3307 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13729", "outputIndexes": [ 3308 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3308 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13730", "outputIndexes": [ 3309 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3308, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13732", "outputIndexes": [ 3310 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3310 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13733", "outputIndexes": [ 3311 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3306, 3309, 3311, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice13736", "outputIndexes": [ 3312 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3312 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze13737", "outputIndexes": [ 3313 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3313, 3313 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13739", "outputIndexes": [ 3314 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3305, 3314, 3308 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_528_output_0", "outputIndexes": [ 3315 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3315, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1120_output_0", "outputIndexes": [ 3316 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3308 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13747", "outputIndexes": [ 3317 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3310 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13750", "outputIndexes": [ 3318 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3306, 3317, 3318, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice13753", "outputIndexes": [ 3319 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3319 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze13754", "outputIndexes": [ 3320 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 3320 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13755", "outputIndexes": [ 3321 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3321, 3320 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13756", "outputIndexes": [ 3322 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3305, 3322, 3308 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_529_output_0", "outputIndexes": [ 3323 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3323, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1121_output_0", "outputIndexes": [ 3324 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3316, 3324, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_455_output_0", "outputIndexes": [ 3325 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3304, 3325 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_316_output_0", "outputIndexes": [ 3326 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3326 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_35/Cast_output_0", "outputIndexes": [ 3327 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_35/Mul_1_output_0", "inputIndexes": [ 3327 ], "outputIndexes": [ 3328 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 9625963812, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3328, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_280_output_0", "outputIndexes": [ 3329 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3328 ], "main_type": "NONE", "name": "/Shape_492_output_0", "outputIndexes": [ 3330 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3330 ], "main_type": "NONE", "name": "Shape13875", "outputIndexes": [ 3331 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3330 ], "main_type": "NONE", "name": "Rank13877", "outputIndexes": [ 3332 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3332, 3332 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13879", "outputIndexes": [ 3333 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3333 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13880", "outputIndexes": [ 3334 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3333, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13882", "outputIndexes": [ 3335 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3335 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13883", "outputIndexes": [ 3336 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3331, 3334, 3336, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice13886", "outputIndexes": [ 3337 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3337 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze13887", "outputIndexes": [ 3338 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3338 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13888", "outputIndexes": [ 3339 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3339, 3338 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13889", "outputIndexes": [ 3340 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3330, 3340, 3333 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_530_output_0", "outputIndexes": [ 3341 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3341, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_105_output_0", "outputIndexes": [ 3342 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3342, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1127_output_0", "outputIndexes": [ 3343 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3328, 3343, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_141_output_0", "outputIndexes": [ 3344 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3344 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_70_output_0", "outputIndexes": [ 3345 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3342, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1126_output_0", "outputIndexes": [ 3346 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3328, 20, 3346, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_140_output_0", "outputIndexes": [ 3347 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3345, 3347 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_458_output_0", "outputIndexes": [ 3348 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3348, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_281_output_0", "outputIndexes": [ 3349 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3329, 3349 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_175_output_0", "outputIndexes": [ 3350 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3303 ], "outputIndexes": [ 6987 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 6987 ], "outputIndexes": [ 6988 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/k_proj/Linear", "inputIndexes": [ 6988 ], "outputIndexes": [ 6989 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 9625964836, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 6989 ], "outputIndexes": [ 6990 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6990 ], "outputIndexes": [ 3351 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3315, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1122_output_0", "outputIndexes": [ 3352 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3323, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1123_output_0", "outputIndexes": [ 3353 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3352, 3353, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_456_output_0", "outputIndexes": [ 3354 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3351, 3354 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_317_output_0", "outputIndexes": [ 3355 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3355 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_35/Cast_output_0", "outputIndexes": [ 3356 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_35/Mul_1_output_0", "inputIndexes": [ 3356 ], "outputIndexes": [ 3357 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 9628913978, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3357, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_282_output_0", "outputIndexes": [ 3358 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3357 ], "main_type": "NONE", "name": "/Shape_493_output_0", "outputIndexes": [ 3359 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3359 ], "main_type": "NONE", "name": "Shape13771", "outputIndexes": [ 3360 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3359 ], "main_type": "NONE", "name": "Rank13773", "outputIndexes": [ 3361 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3361, 3361 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13775", "outputIndexes": [ 3362 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3362 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13776", "outputIndexes": [ 3363 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3362, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13778", "outputIndexes": [ 3364 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3364 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze13779", "outputIndexes": [ 3365 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3360, 3363, 3365, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice13782", "outputIndexes": [ 3366 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3366 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze13783", "outputIndexes": [ 3367 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3367 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13784", "outputIndexes": [ 3368 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3368, 3367 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp13785", "outputIndexes": [ 3369 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3359, 3369, 3362 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_531_output_0", "outputIndexes": [ 3370 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3370, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_106_output_0", "outputIndexes": [ 3371 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3371, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1129_output_0", "outputIndexes": [ 3372 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3357, 3372, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_143_output_0", "outputIndexes": [ 3373 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3373 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_71_output_0", "outputIndexes": [ 3374 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3371, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1128_output_0", "outputIndexes": [ 3375 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3357, 20, 3375, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_142_output_0", "outputIndexes": [ 3376 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3374, 3376 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_459_output_0", "outputIndexes": [ 3377 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3377, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_283_output_0", "outputIndexes": [ 3378 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3358, 3378 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_176_output_0", "outputIndexes": [ 3379 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3303 ], "outputIndexes": [ 6991 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 6991 ], "outputIndexes": [ 6992 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/v_proj/Linear", "inputIndexes": [ 6992 ], "outputIndexes": [ 6993 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 9628915002, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 6993 ], "outputIndexes": [ 6994 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6994 ], "outputIndexes": [ 3380 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3315, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1124_output_0", "outputIndexes": [ 3381 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3323, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1125_output_0", "outputIndexes": [ 3382 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3381, 3382, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_457_output_0", "outputIndexes": [ 3383 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3380, 3383 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_318_output_0", "outputIndexes": [ 3384 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3350, 3379, 3384, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_323_output_0", "outputIndexes": [ 3385 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3385 ], "outputIndexes": [ 6995 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 6995 ], "outputIndexes": [ 6996 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/o_proj/Linear", "inputIndexes": [ 6996 ], "outputIndexes": [ 6997 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 9631864144, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 6997 ], "outputIndexes": [ 6998 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 6998 ], "outputIndexes": [ 3386 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3302, 3386 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_178_output_0", "outputIndexes": [ 3387 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_35/Mul_1_output_0", "inputIndexes": [ 3387 ], "outputIndexes": [ 3388 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 9655457126, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3388 ], "outputIndexes": [ 6999 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 6999 ], "outputIndexes": [ 7000 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/gate_proj/Linear", "inputIndexes": [ 7000 ], "outputIndexes": [ 7001 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 9655498086, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7001 ], "outputIndexes": [ 7002 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7002 ], "outputIndexes": [ 3389 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3389 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_35/Mul_output_0", "outputIndexes": [ 3390 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3388 ], "outputIndexes": [ 7003 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7003 ], "outputIndexes": [ 7004 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/up_proj/Linear", "inputIndexes": [ 7004 ], "outputIndexes": [ 7005 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 9729226108, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7005 ], "outputIndexes": [ 7006 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7006 ], "outputIndexes": [ 3391 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3390, 3391 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_35/Mul_output_0", "outputIndexes": [ 3392 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3392 ], "outputIndexes": [ 7007 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7007 ], "outputIndexes": [ 7008 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/down_proj/Linear", "inputIndexes": [ 7008 ], "outputIndexes": [ 7009 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 9802954130, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7009 ], "outputIndexes": [ 7010 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.35/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7010 ], "outputIndexes": [ 3393 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3387, 3393 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_179_output_0", "outputIndexes": [ 3394 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3394, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_324_output_0", "outputIndexes": [ 3395 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_36/Mul_1_output_0", "inputIndexes": [ 3395 ], "outputIndexes": [ 3396 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 9876682152, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3396 ], "outputIndexes": [ 7011 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7011 ], "outputIndexes": [ 7012 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/q_proj/Linear", "inputIndexes": [ 7012 ], "outputIndexes": [ 7013 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 9876723112, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7013 ], "outputIndexes": [ 7014 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7014 ], "outputIndexes": [ 3397 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3396 ], "main_type": "NONE", "name": "/Shape_504_output_0", "outputIndexes": [ 3398 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3398 ], "main_type": "NONE", "name": "Shape14114", "outputIndexes": [ 3399 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3398 ], "main_type": "NONE", "name": "Rank14116", "outputIndexes": [ 3400 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3400, 3400 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14118", "outputIndexes": [ 3401 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3401 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14119", "outputIndexes": [ 3402 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3401, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14121", "outputIndexes": [ 3403 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3403 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14122", "outputIndexes": [ 3404 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3399, 3402, 3404, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice14125", "outputIndexes": [ 3405 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3405 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze14126", "outputIndexes": [ 3406 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3406, 3406 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14128", "outputIndexes": [ 3407 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3398, 3407, 3401 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_543_output_0", "outputIndexes": [ 3408 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3408, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1152_output_0", "outputIndexes": [ 3409 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3401 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14136", "outputIndexes": [ 3410 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3403 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14139", "outputIndexes": [ 3411 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3399, 3410, 3411, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice14142", "outputIndexes": [ 3412 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3412 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze14143", "outputIndexes": [ 3413 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 3413 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14144", "outputIndexes": [ 3414 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3414, 3413 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14145", "outputIndexes": [ 3415 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3398, 3415, 3401 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_544_output_0", "outputIndexes": [ 3416 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3416, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1153_output_0", "outputIndexes": [ 3417 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3409, 3417, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_468_output_0", "outputIndexes": [ 3418 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3397, 3418 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_325_output_0", "outputIndexes": [ 3419 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3419 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_36/Cast_output_0", "outputIndexes": [ 3420 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_36/Mul_1_output_0", "inputIndexes": [ 3420 ], "outputIndexes": [ 3421 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 9900316094, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3421, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_288_output_0", "outputIndexes": [ 3422 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3421 ], "main_type": "NONE", "name": "/Shape_506_output_0", "outputIndexes": [ 3423 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3423 ], "main_type": "NONE", "name": "Shape14264", "outputIndexes": [ 3424 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3423 ], "main_type": "NONE", "name": "Rank14266", "outputIndexes": [ 3425 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3425, 3425 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14268", "outputIndexes": [ 3426 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3426 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14269", "outputIndexes": [ 3427 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3426, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14271", "outputIndexes": [ 3428 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3428 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14272", "outputIndexes": [ 3429 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3424, 3427, 3429, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice14275", "outputIndexes": [ 3430 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3430 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze14276", "outputIndexes": [ 3431 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3431 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14277", "outputIndexes": [ 3432 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3432, 3431 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14278", "outputIndexes": [ 3433 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3423, 3433, 3426 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_545_output_0", "outputIndexes": [ 3434 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3434, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_108_output_0", "outputIndexes": [ 3435 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3435, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1159_output_0", "outputIndexes": [ 3436 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3421, 3436, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_145_output_0", "outputIndexes": [ 3437 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3437 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_72_output_0", "outputIndexes": [ 3438 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3435, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1158_output_0", "outputIndexes": [ 3439 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3421, 20, 3439, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_144_output_0", "outputIndexes": [ 3440 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3438, 3440 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_471_output_0", "outputIndexes": [ 3441 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3441, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_289_output_0", "outputIndexes": [ 3442 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3422, 3442 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_180_output_0", "outputIndexes": [ 3443 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3396 ], "outputIndexes": [ 7015 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7015 ], "outputIndexes": [ 7016 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/k_proj/Linear", "inputIndexes": [ 7016 ], "outputIndexes": [ 7017 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 9900317118, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7017 ], "outputIndexes": [ 7018 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7018 ], "outputIndexes": [ 3444 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3408, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1154_output_0", "outputIndexes": [ 3445 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3416, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1155_output_0", "outputIndexes": [ 3446 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3445, 3446, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_469_output_0", "outputIndexes": [ 3447 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3444, 3447 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_326_output_0", "outputIndexes": [ 3448 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3448 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_36/Cast_output_0", "outputIndexes": [ 3449 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_36/Mul_1_output_0", "inputIndexes": [ 3449 ], "outputIndexes": [ 3450 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 9903266260, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3450, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_290_output_0", "outputIndexes": [ 3451 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3450 ], "main_type": "NONE", "name": "/Shape_507_output_0", "outputIndexes": [ 3452 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3452 ], "main_type": "NONE", "name": "Shape14160", "outputIndexes": [ 3453 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3452 ], "main_type": "NONE", "name": "Rank14162", "outputIndexes": [ 3454 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3454, 3454 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14164", "outputIndexes": [ 3455 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3455 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14165", "outputIndexes": [ 3456 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3455, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14167", "outputIndexes": [ 3457 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3457 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14168", "outputIndexes": [ 3458 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3453, 3456, 3458, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice14171", "outputIndexes": [ 3459 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3459 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze14172", "outputIndexes": [ 3460 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3460 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14173", "outputIndexes": [ 3461 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3461, 3460 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14174", "outputIndexes": [ 3462 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3452, 3462, 3455 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_546_output_0", "outputIndexes": [ 3463 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3463, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_109_output_0", "outputIndexes": [ 3464 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3464, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1161_output_0", "outputIndexes": [ 3465 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3450, 3465, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_147_output_0", "outputIndexes": [ 3466 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3466 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_73_output_0", "outputIndexes": [ 3467 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3464, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1160_output_0", "outputIndexes": [ 3468 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3450, 20, 3468, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_146_output_0", "outputIndexes": [ 3469 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3467, 3469 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_472_output_0", "outputIndexes": [ 3470 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3470, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_291_output_0", "outputIndexes": [ 3471 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3451, 3471 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_181_output_0", "outputIndexes": [ 3472 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3396 ], "outputIndexes": [ 7019 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7019 ], "outputIndexes": [ 7020 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/v_proj/Linear", "inputIndexes": [ 7020 ], "outputIndexes": [ 7021 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 9903267284, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7021 ], "outputIndexes": [ 7022 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7022 ], "outputIndexes": [ 3473 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3408, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1156_output_0", "outputIndexes": [ 3474 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3416, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1157_output_0", "outputIndexes": [ 3475 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3474, 3475, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_470_output_0", "outputIndexes": [ 3476 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3473, 3476 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_327_output_0", "outputIndexes": [ 3477 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3443, 3472, 3477, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_332_output_0", "outputIndexes": [ 3478 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3478 ], "outputIndexes": [ 7023 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7023 ], "outputIndexes": [ 7024 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/o_proj/Linear", "inputIndexes": [ 7024 ], "outputIndexes": [ 7025 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 9906216426, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7025 ], "outputIndexes": [ 7026 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7026 ], "outputIndexes": [ 3479 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3395, 3479 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_183_output_0", "outputIndexes": [ 3480 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_36/Mul_1_output_0", "inputIndexes": [ 3480 ], "outputIndexes": [ 3481 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 9929809408, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3481 ], "outputIndexes": [ 7027 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7027 ], "outputIndexes": [ 7028 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/mlp/gate_proj/Linear", "inputIndexes": [ 7028 ], "outputIndexes": [ 7029 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 9929850368, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7029 ], "outputIndexes": [ 7030 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7030 ], "outputIndexes": [ 3482 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3482 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_36/Mul_output_0", "outputIndexes": [ 3483 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3481 ], "outputIndexes": [ 7031 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7031 ], "outputIndexes": [ 7032 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/mlp/up_proj/Linear", "inputIndexes": [ 7032 ], "outputIndexes": [ 7033 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 10003578390, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7033 ], "outputIndexes": [ 7034 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7034 ], "outputIndexes": [ 3484 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3483, 3484 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_36/Mul_output_0", "outputIndexes": [ 3485 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3485 ], "outputIndexes": [ 7035 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7035 ], "outputIndexes": [ 7036 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/mlp/down_proj/Linear", "inputIndexes": [ 7036 ], "outputIndexes": [ 7037 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 10077306412, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7037 ], "outputIndexes": [ 7038 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.36/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7038 ], "outputIndexes": [ 3486 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3480, 3486 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_184_output_0", "outputIndexes": [ 3487 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3487, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_333_output_0", "outputIndexes": [ 3488 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_37/Mul_1_output_0", "inputIndexes": [ 3488 ], "outputIndexes": [ 3489 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 10151034434, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3489 ], "outputIndexes": [ 7039 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7039 ], "outputIndexes": [ 7040 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/q_proj/Linear", "inputIndexes": [ 7040 ], "outputIndexes": [ 7041 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 10151075394, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7041 ], "outputIndexes": [ 7042 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7042 ], "outputIndexes": [ 3490 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3489 ], "main_type": "NONE", "name": "/Shape_518_output_0", "outputIndexes": [ 3491 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3491 ], "main_type": "NONE", "name": "Shape14503", "outputIndexes": [ 3492 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3491 ], "main_type": "NONE", "name": "Rank14505", "outputIndexes": [ 3493 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3493, 3493 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14507", "outputIndexes": [ 3494 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3494 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14508", "outputIndexes": [ 3495 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3494, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14510", "outputIndexes": [ 3496 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3496 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14511", "outputIndexes": [ 3497 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3492, 3495, 3497, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice14514", "outputIndexes": [ 3498 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3498 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze14515", "outputIndexes": [ 3499 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3499, 3499 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14517", "outputIndexes": [ 3500 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3491, 3500, 3494 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_558_output_0", "outputIndexes": [ 3501 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3501, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1184_output_0", "outputIndexes": [ 3502 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3494 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14525", "outputIndexes": [ 3503 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3496 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14528", "outputIndexes": [ 3504 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3492, 3503, 3504, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice14531", "outputIndexes": [ 3505 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3505 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze14532", "outputIndexes": [ 3506 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 3506 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14533", "outputIndexes": [ 3507 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3507, 3506 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14534", "outputIndexes": [ 3508 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3491, 3508, 3494 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_559_output_0", "outputIndexes": [ 3509 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3509, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1185_output_0", "outputIndexes": [ 3510 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3502, 3510, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_481_output_0", "outputIndexes": [ 3511 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3490, 3511 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_334_output_0", "outputIndexes": [ 3512 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3512 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_37/Cast_output_0", "outputIndexes": [ 3513 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_37/Mul_1_output_0", "inputIndexes": [ 3513 ], "outputIndexes": [ 3514 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 10174668376, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3514, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_296_output_0", "outputIndexes": [ 3515 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3514 ], "main_type": "NONE", "name": "/Shape_520_output_0", "outputIndexes": [ 3516 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3516 ], "main_type": "NONE", "name": "Shape14653", "outputIndexes": [ 3517 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3516 ], "main_type": "NONE", "name": "Rank14655", "outputIndexes": [ 3518 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3518, 3518 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14657", "outputIndexes": [ 3519 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3519 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14658", "outputIndexes": [ 3520 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3519, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14660", "outputIndexes": [ 3521 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3521 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14661", "outputIndexes": [ 3522 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3517, 3520, 3522, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice14664", "outputIndexes": [ 3523 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3523 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze14665", "outputIndexes": [ 3524 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3524 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14666", "outputIndexes": [ 3525 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3525, 3524 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14667", "outputIndexes": [ 3526 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3516, 3526, 3519 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_560_output_0", "outputIndexes": [ 3527 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3527, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_111_output_0", "outputIndexes": [ 3528 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3528, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1191_output_0", "outputIndexes": [ 3529 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3514, 3529, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_149_output_0", "outputIndexes": [ 3530 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3530 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_74_output_0", "outputIndexes": [ 3531 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3528, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1190_output_0", "outputIndexes": [ 3532 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3514, 20, 3532, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_148_output_0", "outputIndexes": [ 3533 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3531, 3533 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_484_output_0", "outputIndexes": [ 3534 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3534, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_297_output_0", "outputIndexes": [ 3535 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3515, 3535 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_185_output_0", "outputIndexes": [ 3536 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3489 ], "outputIndexes": [ 7043 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7043 ], "outputIndexes": [ 7044 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/k_proj/Linear", "inputIndexes": [ 7044 ], "outputIndexes": [ 7045 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 10174669400, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7045 ], "outputIndexes": [ 7046 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7046 ], "outputIndexes": [ 3537 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3501, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1186_output_0", "outputIndexes": [ 3538 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3509, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1187_output_0", "outputIndexes": [ 3539 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3538, 3539, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_482_output_0", "outputIndexes": [ 3540 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3537, 3540 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_335_output_0", "outputIndexes": [ 3541 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3541 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_37/Cast_output_0", "outputIndexes": [ 3542 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_37/Mul_1_output_0", "inputIndexes": [ 3542 ], "outputIndexes": [ 3543 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 10177618542, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3543, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_298_output_0", "outputIndexes": [ 3544 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3543 ], "main_type": "NONE", "name": "/Shape_521_output_0", "outputIndexes": [ 3545 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3545 ], "main_type": "NONE", "name": "Shape14549", "outputIndexes": [ 3546 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3545 ], "main_type": "NONE", "name": "Rank14551", "outputIndexes": [ 3547 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3547, 3547 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14553", "outputIndexes": [ 3548 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3548 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14554", "outputIndexes": [ 3549 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3548, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14556", "outputIndexes": [ 3550 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3550 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14557", "outputIndexes": [ 3551 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3546, 3549, 3551, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice14560", "outputIndexes": [ 3552 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3552 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze14561", "outputIndexes": [ 3553 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3553 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14562", "outputIndexes": [ 3554 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3554, 3553 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14563", "outputIndexes": [ 3555 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3545, 3555, 3548 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_561_output_0", "outputIndexes": [ 3556 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3556, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_112_output_0", "outputIndexes": [ 3557 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3557, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1193_output_0", "outputIndexes": [ 3558 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3543, 3558, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_151_output_0", "outputIndexes": [ 3559 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3559 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_75_output_0", "outputIndexes": [ 3560 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3557, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1192_output_0", "outputIndexes": [ 3561 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3543, 20, 3561, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_150_output_0", "outputIndexes": [ 3562 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3560, 3562 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_485_output_0", "outputIndexes": [ 3563 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3563, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_299_output_0", "outputIndexes": [ 3564 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3544, 3564 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_186_output_0", "outputIndexes": [ 3565 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3489 ], "outputIndexes": [ 7047 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7047 ], "outputIndexes": [ 7048 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/v_proj/Linear", "inputIndexes": [ 7048 ], "outputIndexes": [ 7049 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 10177619566, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7049 ], "outputIndexes": [ 7050 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7050 ], "outputIndexes": [ 3566 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3501, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1188_output_0", "outputIndexes": [ 3567 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3509, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1189_output_0", "outputIndexes": [ 3568 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3567, 3568, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_483_output_0", "outputIndexes": [ 3569 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3566, 3569 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_336_output_0", "outputIndexes": [ 3570 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3536, 3565, 3570, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_341_output_0", "outputIndexes": [ 3571 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3571 ], "outputIndexes": [ 7051 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7051 ], "outputIndexes": [ 7052 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/o_proj/Linear", "inputIndexes": [ 7052 ], "outputIndexes": [ 7053 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 10180568708, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7053 ], "outputIndexes": [ 7054 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7054 ], "outputIndexes": [ 3572 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3488, 3572 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_188_output_0", "outputIndexes": [ 3573 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_37/Mul_1_output_0", "inputIndexes": [ 3573 ], "outputIndexes": [ 3574 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 10204161690, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3574 ], "outputIndexes": [ 7055 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7055 ], "outputIndexes": [ 7056 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/mlp/gate_proj/Linear", "inputIndexes": [ 7056 ], "outputIndexes": [ 7057 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 10204202650, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7057 ], "outputIndexes": [ 7058 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7058 ], "outputIndexes": [ 3575 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3575 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_37/Mul_output_0", "outputIndexes": [ 3576 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3574 ], "outputIndexes": [ 7059 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7059 ], "outputIndexes": [ 7060 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/mlp/up_proj/Linear", "inputIndexes": [ 7060 ], "outputIndexes": [ 7061 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 10277930672, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7061 ], "outputIndexes": [ 7062 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7062 ], "outputIndexes": [ 3577 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3576, 3577 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_37/Mul_output_0", "outputIndexes": [ 3578 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3578 ], "outputIndexes": [ 7063 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7063 ], "outputIndexes": [ 7064 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/mlp/down_proj/Linear", "inputIndexes": [ 7064 ], "outputIndexes": [ 7065 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 10351658694, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7065 ], "outputIndexes": [ 7066 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.37/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7066 ], "outputIndexes": [ 3579 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3573, 3579 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_189_output_0", "outputIndexes": [ 3580 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3580, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_342_output_0", "outputIndexes": [ 3581 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_38/Mul_1_output_0", "inputIndexes": [ 3581 ], "outputIndexes": [ 3582 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 10425386716, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3582 ], "outputIndexes": [ 7067 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7067 ], "outputIndexes": [ 7068 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/q_proj/Linear", "inputIndexes": [ 7068 ], "outputIndexes": [ 7069 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 10425427676, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7069 ], "outputIndexes": [ 7070 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7070 ], "outputIndexes": [ 3583 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3582 ], "main_type": "NONE", "name": "/Shape_532_output_0", "outputIndexes": [ 3584 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3584 ], "main_type": "NONE", "name": "Shape14892", "outputIndexes": [ 3585 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3584 ], "main_type": "NONE", "name": "Rank14894", "outputIndexes": [ 3586 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3586, 3586 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14896", "outputIndexes": [ 3587 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3587 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14897", "outputIndexes": [ 3588 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3587, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14899", "outputIndexes": [ 3589 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3589 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14900", "outputIndexes": [ 3590 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3585, 3588, 3590, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice14903", "outputIndexes": [ 3591 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3591 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze14904", "outputIndexes": [ 3592 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3592, 3592 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14906", "outputIndexes": [ 3593 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3584, 3593, 3587 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_573_output_0", "outputIndexes": [ 3594 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3594, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1216_output_0", "outputIndexes": [ 3595 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3587 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14914", "outputIndexes": [ 3596 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3589 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14917", "outputIndexes": [ 3597 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3585, 3596, 3597, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice14920", "outputIndexes": [ 3598 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3598 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze14921", "outputIndexes": [ 3599 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 3599 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14922", "outputIndexes": [ 3600 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3600, 3599 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14923", "outputIndexes": [ 3601 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3584, 3601, 3587 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_574_output_0", "outputIndexes": [ 3602 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3602, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1217_output_0", "outputIndexes": [ 3603 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3595, 3603, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_494_output_0", "outputIndexes": [ 3604 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3583, 3604 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_343_output_0", "outputIndexes": [ 3605 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3605 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_38/Cast_output_0", "outputIndexes": [ 3606 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_38/Mul_1_output_0", "inputIndexes": [ 3606 ], "outputIndexes": [ 3607 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 10449020658, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3607, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_304_output_0", "outputIndexes": [ 3608 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3607 ], "main_type": "NONE", "name": "/Shape_534_output_0", "outputIndexes": [ 3609 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3609 ], "main_type": "NONE", "name": "Shape15042", "outputIndexes": [ 3610 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3609 ], "main_type": "NONE", "name": "Rank15044", "outputIndexes": [ 3611 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3611, 3611 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15046", "outputIndexes": [ 3612 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3612 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15047", "outputIndexes": [ 3613 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3612, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15049", "outputIndexes": [ 3614 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3614 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15050", "outputIndexes": [ 3615 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3610, 3613, 3615, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice15053", "outputIndexes": [ 3616 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3616 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze15054", "outputIndexes": [ 3617 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3617 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15055", "outputIndexes": [ 3618 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3618, 3617 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15056", "outputIndexes": [ 3619 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3609, 3619, 3612 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_575_output_0", "outputIndexes": [ 3620 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3620, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_114_output_0", "outputIndexes": [ 3621 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3621, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1223_output_0", "outputIndexes": [ 3622 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3607, 3622, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_153_output_0", "outputIndexes": [ 3623 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3623 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_76_output_0", "outputIndexes": [ 3624 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3621, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1222_output_0", "outputIndexes": [ 3625 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3607, 20, 3625, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_152_output_0", "outputIndexes": [ 3626 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3624, 3626 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_497_output_0", "outputIndexes": [ 3627 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3627, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_305_output_0", "outputIndexes": [ 3628 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3608, 3628 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_190_output_0", "outputIndexes": [ 3629 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3582 ], "outputIndexes": [ 7071 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7071 ], "outputIndexes": [ 7072 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/k_proj/Linear", "inputIndexes": [ 7072 ], "outputIndexes": [ 7073 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 10449021682, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7073 ], "outputIndexes": [ 7074 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7074 ], "outputIndexes": [ 3630 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3594, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1218_output_0", "outputIndexes": [ 3631 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3602, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1219_output_0", "outputIndexes": [ 3632 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3631, 3632, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_495_output_0", "outputIndexes": [ 3633 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3630, 3633 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_344_output_0", "outputIndexes": [ 3634 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3634 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_38/Cast_output_0", "outputIndexes": [ 3635 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_38/Mul_1_output_0", "inputIndexes": [ 3635 ], "outputIndexes": [ 3636 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 10451970824, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3636, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_306_output_0", "outputIndexes": [ 3637 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3636 ], "main_type": "NONE", "name": "/Shape_535_output_0", "outputIndexes": [ 3638 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3638 ], "main_type": "NONE", "name": "Shape14938", "outputIndexes": [ 3639 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3638 ], "main_type": "NONE", "name": "Rank14940", "outputIndexes": [ 3640 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3640, 3640 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14942", "outputIndexes": [ 3641 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3641 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14943", "outputIndexes": [ 3642 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3641, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14945", "outputIndexes": [ 3643 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3643 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze14946", "outputIndexes": [ 3644 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3639, 3642, 3644, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice14949", "outputIndexes": [ 3645 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3645 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze14950", "outputIndexes": [ 3646 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3646 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14951", "outputIndexes": [ 3647 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3647, 3646 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp14952", "outputIndexes": [ 3648 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3638, 3648, 3641 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_576_output_0", "outputIndexes": [ 3649 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3649, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_115_output_0", "outputIndexes": [ 3650 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3650, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1225_output_0", "outputIndexes": [ 3651 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3636, 3651, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_155_output_0", "outputIndexes": [ 3652 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3652 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_77_output_0", "outputIndexes": [ 3653 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3650, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1224_output_0", "outputIndexes": [ 3654 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3636, 20, 3654, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_154_output_0", "outputIndexes": [ 3655 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3653, 3655 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_498_output_0", "outputIndexes": [ 3656 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3656, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_307_output_0", "outputIndexes": [ 3657 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3637, 3657 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_191_output_0", "outputIndexes": [ 3658 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3582 ], "outputIndexes": [ 7075 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7075 ], "outputIndexes": [ 7076 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/v_proj/Linear", "inputIndexes": [ 7076 ], "outputIndexes": [ 7077 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 10451971848, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7077 ], "outputIndexes": [ 7078 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7078 ], "outputIndexes": [ 3659 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3594, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1220_output_0", "outputIndexes": [ 3660 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3602, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1221_output_0", "outputIndexes": [ 3661 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3660, 3661, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_496_output_0", "outputIndexes": [ 3662 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3659, 3662 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_345_output_0", "outputIndexes": [ 3663 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3629, 3658, 3663, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_350_output_0", "outputIndexes": [ 3664 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3664 ], "outputIndexes": [ 7079 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7079 ], "outputIndexes": [ 7080 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/o_proj/Linear", "inputIndexes": [ 7080 ], "outputIndexes": [ 7081 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 10454920990, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7081 ], "outputIndexes": [ 7082 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7082 ], "outputIndexes": [ 3665 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3581, 3665 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_193_output_0", "outputIndexes": [ 3666 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_38/Mul_1_output_0", "inputIndexes": [ 3666 ], "outputIndexes": [ 3667 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 10478513972, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3667 ], "outputIndexes": [ 7083 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7083 ], "outputIndexes": [ 7084 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/mlp/gate_proj/Linear", "inputIndexes": [ 7084 ], "outputIndexes": [ 7085 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 10478554932, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7085 ], "outputIndexes": [ 7086 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7086 ], "outputIndexes": [ 3668 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3668 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_38/Mul_output_0", "outputIndexes": [ 3669 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3667 ], "outputIndexes": [ 7087 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7087 ], "outputIndexes": [ 7088 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/mlp/up_proj/Linear", "inputIndexes": [ 7088 ], "outputIndexes": [ 7089 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 10552282954, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7089 ], "outputIndexes": [ 7090 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7090 ], "outputIndexes": [ 3670 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3669, 3670 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_38/Mul_output_0", "outputIndexes": [ 3671 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3671 ], "outputIndexes": [ 7091 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7091 ], "outputIndexes": [ 7092 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/mlp/down_proj/Linear", "inputIndexes": [ 7092 ], "outputIndexes": [ 7093 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 10626010976, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7093 ], "outputIndexes": [ 7094 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.38/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7094 ], "outputIndexes": [ 3672 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3666, 3672 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_194_output_0", "outputIndexes": [ 3673 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3673, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_351_output_0", "outputIndexes": [ 3674 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_39/Mul_1_output_0", "inputIndexes": [ 3674 ], "outputIndexes": [ 3675 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 10699738998, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3675 ], "outputIndexes": [ 7095 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7095 ], "outputIndexes": [ 7096 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/q_proj/Linear", "inputIndexes": [ 7096 ], "outputIndexes": [ 7097 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 10699779958, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7097 ], "outputIndexes": [ 7098 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7098 ], "outputIndexes": [ 3676 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3675 ], "main_type": "NONE", "name": "/Shape_546_output_0", "outputIndexes": [ 3677 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3677 ], "main_type": "NONE", "name": "Shape15281", "outputIndexes": [ 3678 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3677 ], "main_type": "NONE", "name": "Rank15283", "outputIndexes": [ 3679 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3679, 3679 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15285", "outputIndexes": [ 3680 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3680 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15286", "outputIndexes": [ 3681 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3680, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15288", "outputIndexes": [ 3682 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3682 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15289", "outputIndexes": [ 3683 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3678, 3681, 3683, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice15292", "outputIndexes": [ 3684 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3684 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze15293", "outputIndexes": [ 3685 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3685, 3685 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15295", "outputIndexes": [ 3686 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3677, 3686, 3680 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_588_output_0", "outputIndexes": [ 3687 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3687, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1248_output_0", "outputIndexes": [ 3688 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3680 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15303", "outputIndexes": [ 3689 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3682 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15306", "outputIndexes": [ 3690 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3678, 3689, 3690, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice15309", "outputIndexes": [ 3691 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3691 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze15310", "outputIndexes": [ 3692 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 3692 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15311", "outputIndexes": [ 3693 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3693, 3692 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15312", "outputIndexes": [ 3694 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3677, 3694, 3680 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_589_output_0", "outputIndexes": [ 3695 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3695, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1249_output_0", "outputIndexes": [ 3696 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3688, 3696, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_507_output_0", "outputIndexes": [ 3697 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3676, 3697 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_352_output_0", "outputIndexes": [ 3698 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3698 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_39/Cast_output_0", "outputIndexes": [ 3699 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_39/Mul_1_output_0", "inputIndexes": [ 3699 ], "outputIndexes": [ 3700 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 10723372940, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3700, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_312_output_0", "outputIndexes": [ 3701 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3700 ], "main_type": "NONE", "name": "/Shape_548_output_0", "outputIndexes": [ 3702 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3702 ], "main_type": "NONE", "name": "Shape15431", "outputIndexes": [ 3703 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3702 ], "main_type": "NONE", "name": "Rank15433", "outputIndexes": [ 3704 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3704, 3704 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15435", "outputIndexes": [ 3705 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3705 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15436", "outputIndexes": [ 3706 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3705, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15438", "outputIndexes": [ 3707 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3707 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15439", "outputIndexes": [ 3708 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3703, 3706, 3708, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice15442", "outputIndexes": [ 3709 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3709 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze15443", "outputIndexes": [ 3710 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3710 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15444", "outputIndexes": [ 3711 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3711, 3710 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15445", "outputIndexes": [ 3712 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3702, 3712, 3705 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_590_output_0", "outputIndexes": [ 3713 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3713, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_117_output_0", "outputIndexes": [ 3714 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3714, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1255_output_0", "outputIndexes": [ 3715 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3700, 3715, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_157_output_0", "outputIndexes": [ 3716 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3716 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_78_output_0", "outputIndexes": [ 3717 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3714, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1254_output_0", "outputIndexes": [ 3718 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3700, 20, 3718, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_156_output_0", "outputIndexes": [ 3719 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3717, 3719 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_510_output_0", "outputIndexes": [ 3720 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3720, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_313_output_0", "outputIndexes": [ 3721 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3701, 3721 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_195_output_0", "outputIndexes": [ 3722 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3675 ], "outputIndexes": [ 7099 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7099 ], "outputIndexes": [ 7100 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/k_proj/Linear", "inputIndexes": [ 7100 ], "outputIndexes": [ 7101 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 10723373964, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7101 ], "outputIndexes": [ 7102 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7102 ], "outputIndexes": [ 3723 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3687, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1250_output_0", "outputIndexes": [ 3724 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3695, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1251_output_0", "outputIndexes": [ 3725 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3724, 3725, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_508_output_0", "outputIndexes": [ 3726 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3723, 3726 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_353_output_0", "outputIndexes": [ 3727 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3727 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_39/Cast_output_0", "outputIndexes": [ 3728 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_39/Mul_1_output_0", "inputIndexes": [ 3728 ], "outputIndexes": [ 3729 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 10726323106, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3729, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_314_output_0", "outputIndexes": [ 3730 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3729 ], "main_type": "NONE", "name": "/Shape_549_output_0", "outputIndexes": [ 3731 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3731 ], "main_type": "NONE", "name": "Shape15327", "outputIndexes": [ 3732 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3731 ], "main_type": "NONE", "name": "Rank15329", "outputIndexes": [ 3733 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3733, 3733 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15331", "outputIndexes": [ 3734 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3734 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15332", "outputIndexes": [ 3735 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3734, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15334", "outputIndexes": [ 3736 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3736 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15335", "outputIndexes": [ 3737 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3732, 3735, 3737, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice15338", "outputIndexes": [ 3738 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3738 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze15339", "outputIndexes": [ 3739 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3739 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15340", "outputIndexes": [ 3740 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3740, 3739 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15341", "outputIndexes": [ 3741 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3731, 3741, 3734 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_591_output_0", "outputIndexes": [ 3742 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3742, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_118_output_0", "outputIndexes": [ 3743 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3743, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1257_output_0", "outputIndexes": [ 3744 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3729, 3744, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_159_output_0", "outputIndexes": [ 3745 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3745 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_79_output_0", "outputIndexes": [ 3746 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3743, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1256_output_0", "outputIndexes": [ 3747 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3729, 20, 3747, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_158_output_0", "outputIndexes": [ 3748 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3746, 3748 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_511_output_0", "outputIndexes": [ 3749 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3749, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_315_output_0", "outputIndexes": [ 3750 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3730, 3750 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_196_output_0", "outputIndexes": [ 3751 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3675 ], "outputIndexes": [ 7103 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7103 ], "outputIndexes": [ 7104 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/v_proj/Linear", "inputIndexes": [ 7104 ], "outputIndexes": [ 7105 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 10726324130, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7105 ], "outputIndexes": [ 7106 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7106 ], "outputIndexes": [ 3752 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3687, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1252_output_0", "outputIndexes": [ 3753 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3695, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1253_output_0", "outputIndexes": [ 3754 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3753, 3754, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_509_output_0", "outputIndexes": [ 3755 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3752, 3755 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_354_output_0", "outputIndexes": [ 3756 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3722, 3751, 3756, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_359_output_0", "outputIndexes": [ 3757 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3757 ], "outputIndexes": [ 7107 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7107 ], "outputIndexes": [ 7108 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/o_proj/Linear", "inputIndexes": [ 7108 ], "outputIndexes": [ 7109 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 10729273272, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7109 ], "outputIndexes": [ 7110 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7110 ], "outputIndexes": [ 3758 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3674, 3758 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_198_output_0", "outputIndexes": [ 3759 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_39/Mul_1_output_0", "inputIndexes": [ 3759 ], "outputIndexes": [ 3760 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 10752866254, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3760 ], "outputIndexes": [ 7111 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7111 ], "outputIndexes": [ 7112 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/mlp/gate_proj/Linear", "inputIndexes": [ 7112 ], "outputIndexes": [ 7113 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 10752907214, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7113 ], "outputIndexes": [ 7114 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7114 ], "outputIndexes": [ 3761 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3761 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_39/Mul_output_0", "outputIndexes": [ 3762 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3760 ], "outputIndexes": [ 7115 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7115 ], "outputIndexes": [ 7116 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/mlp/up_proj/Linear", "inputIndexes": [ 7116 ], "outputIndexes": [ 7117 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 10826635236, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7117 ], "outputIndexes": [ 7118 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7118 ], "outputIndexes": [ 3763 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3762, 3763 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_39/Mul_output_0", "outputIndexes": [ 3764 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3764 ], "outputIndexes": [ 7119 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7119 ], "outputIndexes": [ 7120 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/mlp/down_proj/Linear", "inputIndexes": [ 7120 ], "outputIndexes": [ 7121 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 10900363258, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7121 ], "outputIndexes": [ 7122 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.39/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7122 ], "outputIndexes": [ 3765 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3759, 3765 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_199_output_0", "outputIndexes": [ 3766 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3766, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_360_output_0", "outputIndexes": [ 3767 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_40/Mul_1_output_0", "inputIndexes": [ 3767 ], "outputIndexes": [ 3768 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 10974091280, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3768 ], "outputIndexes": [ 7123 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7123 ], "outputIndexes": [ 7124 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/q_proj/Linear", "inputIndexes": [ 7124 ], "outputIndexes": [ 7125 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 10974132240, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7125 ], "outputIndexes": [ 7126 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7126 ], "outputIndexes": [ 3769 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3768 ], "main_type": "NONE", "name": "/Shape_560_output_0", "outputIndexes": [ 3770 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3770 ], "main_type": "NONE", "name": "Shape15670", "outputIndexes": [ 3771 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3770 ], "main_type": "NONE", "name": "Rank15672", "outputIndexes": [ 3772 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3772, 3772 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15674", "outputIndexes": [ 3773 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3773 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15675", "outputIndexes": [ 3774 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3773, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15677", "outputIndexes": [ 3775 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3775 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15678", "outputIndexes": [ 3776 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3771, 3774, 3776, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice15681", "outputIndexes": [ 3777 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3777 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze15682", "outputIndexes": [ 3778 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3778, 3778 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15684", "outputIndexes": [ 3779 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3770, 3779, 3773 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_603_output_0", "outputIndexes": [ 3780 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3780, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1280_output_0", "outputIndexes": [ 3781 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3773 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15692", "outputIndexes": [ 3782 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3775 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15695", "outputIndexes": [ 3783 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3771, 3782, 3783, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice15698", "outputIndexes": [ 3784 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3784 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze15699", "outputIndexes": [ 3785 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 3785 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15700", "outputIndexes": [ 3786 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3786, 3785 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15701", "outputIndexes": [ 3787 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3770, 3787, 3773 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_604_output_0", "outputIndexes": [ 3788 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3788, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1281_output_0", "outputIndexes": [ 3789 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3781, 3789, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_520_output_0", "outputIndexes": [ 3790 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3769, 3790 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_361_output_0", "outputIndexes": [ 3791 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3791 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_40/Cast_output_0", "outputIndexes": [ 3792 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_40/Mul_1_output_0", "inputIndexes": [ 3792 ], "outputIndexes": [ 3793 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 10997725222, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3793, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_320_output_0", "outputIndexes": [ 3794 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3793 ], "main_type": "NONE", "name": "/Shape_562_output_0", "outputIndexes": [ 3795 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3795 ], "main_type": "NONE", "name": "Shape15820", "outputIndexes": [ 3796 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3795 ], "main_type": "NONE", "name": "Rank15822", "outputIndexes": [ 3797 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3797, 3797 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15824", "outputIndexes": [ 3798 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3798 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15825", "outputIndexes": [ 3799 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3798, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15827", "outputIndexes": [ 3800 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3800 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15828", "outputIndexes": [ 3801 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3796, 3799, 3801, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice15831", "outputIndexes": [ 3802 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3802 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze15832", "outputIndexes": [ 3803 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3803 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15833", "outputIndexes": [ 3804 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3804, 3803 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15834", "outputIndexes": [ 3805 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3795, 3805, 3798 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_605_output_0", "outputIndexes": [ 3806 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3806, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_120_output_0", "outputIndexes": [ 3807 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3807, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1287_output_0", "outputIndexes": [ 3808 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3793, 3808, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_161_output_0", "outputIndexes": [ 3809 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3809 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_80_output_0", "outputIndexes": [ 3810 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3807, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1286_output_0", "outputIndexes": [ 3811 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3793, 20, 3811, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_160_output_0", "outputIndexes": [ 3812 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3810, 3812 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_523_output_0", "outputIndexes": [ 3813 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3813, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_321_output_0", "outputIndexes": [ 3814 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3794, 3814 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_200_output_0", "outputIndexes": [ 3815 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3768 ], "outputIndexes": [ 7127 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7127 ], "outputIndexes": [ 7128 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/k_proj/Linear", "inputIndexes": [ 7128 ], "outputIndexes": [ 7129 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 10997726246, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7129 ], "outputIndexes": [ 7130 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7130 ], "outputIndexes": [ 3816 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3780, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1282_output_0", "outputIndexes": [ 3817 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3788, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1283_output_0", "outputIndexes": [ 3818 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3817, 3818, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_521_output_0", "outputIndexes": [ 3819 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3816, 3819 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_362_output_0", "outputIndexes": [ 3820 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3820 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_40/Cast_output_0", "outputIndexes": [ 3821 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_40/Mul_1_output_0", "inputIndexes": [ 3821 ], "outputIndexes": [ 3822 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 11000675388, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3822, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_322_output_0", "outputIndexes": [ 3823 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3822 ], "main_type": "NONE", "name": "/Shape_563_output_0", "outputIndexes": [ 3824 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3824 ], "main_type": "NONE", "name": "Shape15716", "outputIndexes": [ 3825 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3824 ], "main_type": "NONE", "name": "Rank15718", "outputIndexes": [ 3826 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3826, 3826 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15720", "outputIndexes": [ 3827 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3827 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15721", "outputIndexes": [ 3828 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3827, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15723", "outputIndexes": [ 3829 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3829 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze15724", "outputIndexes": [ 3830 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3825, 3828, 3830, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice15727", "outputIndexes": [ 3831 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3831 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze15728", "outputIndexes": [ 3832 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3832 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15729", "outputIndexes": [ 3833 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3833, 3832 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp15730", "outputIndexes": [ 3834 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3824, 3834, 3827 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_606_output_0", "outputIndexes": [ 3835 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3835, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_121_output_0", "outputIndexes": [ 3836 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3836, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1289_output_0", "outputIndexes": [ 3837 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3822, 3837, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_163_output_0", "outputIndexes": [ 3838 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3838 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_81_output_0", "outputIndexes": [ 3839 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3836, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1288_output_0", "outputIndexes": [ 3840 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3822, 20, 3840, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_162_output_0", "outputIndexes": [ 3841 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3839, 3841 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_524_output_0", "outputIndexes": [ 3842 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3842, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_323_output_0", "outputIndexes": [ 3843 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3823, 3843 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_201_output_0", "outputIndexes": [ 3844 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3768 ], "outputIndexes": [ 7131 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7131 ], "outputIndexes": [ 7132 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/v_proj/Linear", "inputIndexes": [ 7132 ], "outputIndexes": [ 7133 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 11000676412, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7133 ], "outputIndexes": [ 7134 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7134 ], "outputIndexes": [ 3845 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3780, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1284_output_0", "outputIndexes": [ 3846 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3788, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1285_output_0", "outputIndexes": [ 3847 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3846, 3847, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_522_output_0", "outputIndexes": [ 3848 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3845, 3848 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_363_output_0", "outputIndexes": [ 3849 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3815, 3844, 3849, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_368_output_0", "outputIndexes": [ 3850 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3850 ], "outputIndexes": [ 7135 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7135 ], "outputIndexes": [ 7136 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/o_proj/Linear", "inputIndexes": [ 7136 ], "outputIndexes": [ 7137 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 11003625554, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7137 ], "outputIndexes": [ 7138 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7138 ], "outputIndexes": [ 3851 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3767, 3851 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_203_output_0", "outputIndexes": [ 3852 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_40/Mul_1_output_0", "inputIndexes": [ 3852 ], "outputIndexes": [ 3853 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 11027218536, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3853 ], "outputIndexes": [ 7139 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7139 ], "outputIndexes": [ 7140 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/mlp/gate_proj/Linear", "inputIndexes": [ 7140 ], "outputIndexes": [ 7141 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 11027259496, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7141 ], "outputIndexes": [ 7142 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7142 ], "outputIndexes": [ 3854 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3854 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_40/Mul_output_0", "outputIndexes": [ 3855 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3853 ], "outputIndexes": [ 7143 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7143 ], "outputIndexes": [ 7144 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/mlp/up_proj/Linear", "inputIndexes": [ 7144 ], "outputIndexes": [ 7145 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 11100987518, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7145 ], "outputIndexes": [ 7146 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7146 ], "outputIndexes": [ 3856 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3855, 3856 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_40/Mul_output_0", "outputIndexes": [ 3857 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3857 ], "outputIndexes": [ 7147 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7147 ], "outputIndexes": [ 7148 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/mlp/down_proj/Linear", "inputIndexes": [ 7148 ], "outputIndexes": [ 7149 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 11174715540, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7149 ], "outputIndexes": [ 7150 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.40/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7150 ], "outputIndexes": [ 3858 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3852, 3858 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_204_output_0", "outputIndexes": [ 3859 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3859, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_369_output_0", "outputIndexes": [ 3860 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_41/Mul_1_output_0", "inputIndexes": [ 3860 ], "outputIndexes": [ 3861 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 11248443562, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3861 ], "outputIndexes": [ 7151 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7151 ], "outputIndexes": [ 7152 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/q_proj/Linear", "inputIndexes": [ 7152 ], "outputIndexes": [ 7153 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 11248484522, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7153 ], "outputIndexes": [ 7154 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7154 ], "outputIndexes": [ 3862 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3861 ], "main_type": "NONE", "name": "/Shape_574_output_0", "outputIndexes": [ 3863 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3863 ], "main_type": "NONE", "name": "Shape16059", "outputIndexes": [ 3864 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3863 ], "main_type": "NONE", "name": "Rank16061", "outputIndexes": [ 3865 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3865, 3865 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16063", "outputIndexes": [ 3866 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3866 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16064", "outputIndexes": [ 3867 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3866, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16066", "outputIndexes": [ 3868 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3868 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16067", "outputIndexes": [ 3869 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3864, 3867, 3869, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice16070", "outputIndexes": [ 3870 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3870 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze16071", "outputIndexes": [ 3871 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3871, 3871 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16073", "outputIndexes": [ 3872 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3863, 3872, 3866 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_618_output_0", "outputIndexes": [ 3873 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3873, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1312_output_0", "outputIndexes": [ 3874 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3866 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16081", "outputIndexes": [ 3875 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3868 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16084", "outputIndexes": [ 3876 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3864, 3875, 3876, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice16087", "outputIndexes": [ 3877 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3877 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze16088", "outputIndexes": [ 3878 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 3878 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16089", "outputIndexes": [ 3879 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3879, 3878 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16090", "outputIndexes": [ 3880 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3863, 3880, 3866 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_619_output_0", "outputIndexes": [ 3881 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3881, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1313_output_0", "outputIndexes": [ 3882 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3874, 3882, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_533_output_0", "outputIndexes": [ 3883 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3862, 3883 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_370_output_0", "outputIndexes": [ 3884 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3884 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_41/Cast_output_0", "outputIndexes": [ 3885 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_41/Mul_1_output_0", "inputIndexes": [ 3885 ], "outputIndexes": [ 3886 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 11272077504, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3886, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_328_output_0", "outputIndexes": [ 3887 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3886 ], "main_type": "NONE", "name": "/Shape_576_output_0", "outputIndexes": [ 3888 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3888 ], "main_type": "NONE", "name": "Shape16209", "outputIndexes": [ 3889 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3888 ], "main_type": "NONE", "name": "Rank16211", "outputIndexes": [ 3890 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3890, 3890 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16213", "outputIndexes": [ 3891 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3891 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16214", "outputIndexes": [ 3892 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3891, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16216", "outputIndexes": [ 3893 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3893 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16217", "outputIndexes": [ 3894 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3889, 3892, 3894, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice16220", "outputIndexes": [ 3895 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3895 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze16221", "outputIndexes": [ 3896 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3896 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16222", "outputIndexes": [ 3897 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3897, 3896 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16223", "outputIndexes": [ 3898 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3888, 3898, 3891 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_620_output_0", "outputIndexes": [ 3899 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3899, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_123_output_0", "outputIndexes": [ 3900 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3900, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1319_output_0", "outputIndexes": [ 3901 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3886, 3901, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_165_output_0", "outputIndexes": [ 3902 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3902 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_82_output_0", "outputIndexes": [ 3903 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3900, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1318_output_0", "outputIndexes": [ 3904 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3886, 20, 3904, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_164_output_0", "outputIndexes": [ 3905 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3903, 3905 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_536_output_0", "outputIndexes": [ 3906 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3906, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_329_output_0", "outputIndexes": [ 3907 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3887, 3907 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_205_output_0", "outputIndexes": [ 3908 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3861 ], "outputIndexes": [ 7155 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7155 ], "outputIndexes": [ 7156 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/k_proj/Linear", "inputIndexes": [ 7156 ], "outputIndexes": [ 7157 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 11272078528, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7157 ], "outputIndexes": [ 7158 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7158 ], "outputIndexes": [ 3909 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3873, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1314_output_0", "outputIndexes": [ 3910 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3881, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1315_output_0", "outputIndexes": [ 3911 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3910, 3911, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_534_output_0", "outputIndexes": [ 3912 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3909, 3912 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_371_output_0", "outputIndexes": [ 3913 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3913 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_41/Cast_output_0", "outputIndexes": [ 3914 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_41/Mul_1_output_0", "inputIndexes": [ 3914 ], "outputIndexes": [ 3915 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 11275027670, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3915, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_330_output_0", "outputIndexes": [ 3916 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3915 ], "main_type": "NONE", "name": "/Shape_577_output_0", "outputIndexes": [ 3917 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3917 ], "main_type": "NONE", "name": "Shape16105", "outputIndexes": [ 3918 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3917 ], "main_type": "NONE", "name": "Rank16107", "outputIndexes": [ 3919 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3919, 3919 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16109", "outputIndexes": [ 3920 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3920 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16110", "outputIndexes": [ 3921 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3920, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16112", "outputIndexes": [ 3922 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3922 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16113", "outputIndexes": [ 3923 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3918, 3921, 3923, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice16116", "outputIndexes": [ 3924 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3924 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze16117", "outputIndexes": [ 3925 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3925 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16118", "outputIndexes": [ 3926 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3926, 3925 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16119", "outputIndexes": [ 3927 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3917, 3927, 3920 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_621_output_0", "outputIndexes": [ 3928 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3928, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_124_output_0", "outputIndexes": [ 3929 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3929, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1321_output_0", "outputIndexes": [ 3930 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3915, 3930, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_167_output_0", "outputIndexes": [ 3931 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3931 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_83_output_0", "outputIndexes": [ 3932 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3929, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1320_output_0", "outputIndexes": [ 3933 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3915, 20, 3933, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_166_output_0", "outputIndexes": [ 3934 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3932, 3934 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_537_output_0", "outputIndexes": [ 3935 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3935, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_331_output_0", "outputIndexes": [ 3936 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3916, 3936 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_206_output_0", "outputIndexes": [ 3937 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3861 ], "outputIndexes": [ 7159 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7159 ], "outputIndexes": [ 7160 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/v_proj/Linear", "inputIndexes": [ 7160 ], "outputIndexes": [ 7161 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 11275028694, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7161 ], "outputIndexes": [ 7162 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7162 ], "outputIndexes": [ 3938 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3873, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1316_output_0", "outputIndexes": [ 3939 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3881, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1317_output_0", "outputIndexes": [ 3940 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3939, 3940, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_535_output_0", "outputIndexes": [ 3941 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3938, 3941 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_372_output_0", "outputIndexes": [ 3942 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3908, 3937, 3942, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_377_output_0", "outputIndexes": [ 3943 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3943 ], "outputIndexes": [ 7163 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7163 ], "outputIndexes": [ 7164 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/o_proj/Linear", "inputIndexes": [ 7164 ], "outputIndexes": [ 7165 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 11277977836, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7165 ], "outputIndexes": [ 7166 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7166 ], "outputIndexes": [ 3944 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3860, 3944 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_208_output_0", "outputIndexes": [ 3945 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_41/Mul_1_output_0", "inputIndexes": [ 3945 ], "outputIndexes": [ 3946 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 11301570818, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3946 ], "outputIndexes": [ 7167 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7167 ], "outputIndexes": [ 7168 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/mlp/gate_proj/Linear", "inputIndexes": [ 7168 ], "outputIndexes": [ 7169 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 11301611778, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7169 ], "outputIndexes": [ 7170 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7170 ], "outputIndexes": [ 3947 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3947 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_41/Mul_output_0", "outputIndexes": [ 3948 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3946 ], "outputIndexes": [ 7171 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7171 ], "outputIndexes": [ 7172 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/mlp/up_proj/Linear", "inputIndexes": [ 7172 ], "outputIndexes": [ 7173 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 11375339800, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7173 ], "outputIndexes": [ 7174 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7174 ], "outputIndexes": [ 3949 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3948, 3949 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_41/Mul_output_0", "outputIndexes": [ 3950 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3950 ], "outputIndexes": [ 7175 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7175 ], "outputIndexes": [ 7176 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/mlp/down_proj/Linear", "inputIndexes": [ 7176 ], "outputIndexes": [ 7177 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 11449067822, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7177 ], "outputIndexes": [ 7178 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.41/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7178 ], "outputIndexes": [ 3951 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3945, 3951 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_209_output_0", "outputIndexes": [ 3952 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3952, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_378_output_0", "outputIndexes": [ 3953 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_42/Mul_1_output_0", "inputIndexes": [ 3953 ], "outputIndexes": [ 3954 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 11522795844, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3954 ], "outputIndexes": [ 7179 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7179 ], "outputIndexes": [ 7180 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/q_proj/Linear", "inputIndexes": [ 7180 ], "outputIndexes": [ 7181 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 11522836804, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7181 ], "outputIndexes": [ 7182 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7182 ], "outputIndexes": [ 3955 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3954 ], "main_type": "NONE", "name": "/Shape_588_output_0", "outputIndexes": [ 3956 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3956 ], "main_type": "NONE", "name": "Shape16448", "outputIndexes": [ 3957 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3956 ], "main_type": "NONE", "name": "Rank16450", "outputIndexes": [ 3958 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3958, 3958 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16452", "outputIndexes": [ 3959 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3959 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16453", "outputIndexes": [ 3960 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3959, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16455", "outputIndexes": [ 3961 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3961 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16456", "outputIndexes": [ 3962 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3957, 3960, 3962, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice16459", "outputIndexes": [ 3963 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3963 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze16460", "outputIndexes": [ 3964 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3964, 3964 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16462", "outputIndexes": [ 3965 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3956, 3965, 3959 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_633_output_0", "outputIndexes": [ 3966 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3966, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1344_output_0", "outputIndexes": [ 3967 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3959 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16470", "outputIndexes": [ 3968 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3961 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16473", "outputIndexes": [ 3969 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3957, 3968, 3969, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice16476", "outputIndexes": [ 3970 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3970 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze16477", "outputIndexes": [ 3971 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 3971 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16478", "outputIndexes": [ 3972 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3972, 3971 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16479", "outputIndexes": [ 3973 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3956, 3973, 3959 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_634_output_0", "outputIndexes": [ 3974 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3974, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1345_output_0", "outputIndexes": [ 3975 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3967, 3975, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_546_output_0", "outputIndexes": [ 3976 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3955, 3976 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_379_output_0", "outputIndexes": [ 3977 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3977 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_42/Cast_output_0", "outputIndexes": [ 3978 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_42/Mul_1_output_0", "inputIndexes": [ 3978 ], "outputIndexes": [ 3979 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 11546429786, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3979, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_336_output_0", "outputIndexes": [ 3980 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3979 ], "main_type": "NONE", "name": "/Shape_590_output_0", "outputIndexes": [ 3981 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3981 ], "main_type": "NONE", "name": "Shape16598", "outputIndexes": [ 3982 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 3981 ], "main_type": "NONE", "name": "Rank16600", "outputIndexes": [ 3983 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3983, 3983 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16602", "outputIndexes": [ 3984 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3984 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16603", "outputIndexes": [ 3985 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3984, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16605", "outputIndexes": [ 3986 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3986 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16606", "outputIndexes": [ 3987 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3982, 3985, 3987, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice16609", "outputIndexes": [ 3988 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3988 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze16610", "outputIndexes": [ 3989 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 3989 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16611", "outputIndexes": [ 3990 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3990, 3989 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16612", "outputIndexes": [ 3991 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3981, 3991, 3984 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_635_output_0", "outputIndexes": [ 3992 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3992, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_126_output_0", "outputIndexes": [ 3993 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3993, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1351_output_0", "outputIndexes": [ 3994 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3979, 3994, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_169_output_0", "outputIndexes": [ 3995 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3995 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_84_output_0", "outputIndexes": [ 3996 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3993, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1350_output_0", "outputIndexes": [ 3997 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3979, 20, 3997, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_168_output_0", "outputIndexes": [ 3998 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3996, 3998 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_549_output_0", "outputIndexes": [ 3999 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3999, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_337_output_0", "outputIndexes": [ 4000 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3980, 4000 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_210_output_0", "outputIndexes": [ 4001 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3954 ], "outputIndexes": [ 7183 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7183 ], "outputIndexes": [ 7184 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/k_proj/Linear", "inputIndexes": [ 7184 ], "outputIndexes": [ 7185 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 11546430810, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7185 ], "outputIndexes": [ 7186 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7186 ], "outputIndexes": [ 4002 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3966, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1346_output_0", "outputIndexes": [ 4003 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3974, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1347_output_0", "outputIndexes": [ 4004 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4003, 4004, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_547_output_0", "outputIndexes": [ 4005 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4002, 4005 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_380_output_0", "outputIndexes": [ 4006 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4006 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_42/Cast_output_0", "outputIndexes": [ 4007 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_42/Mul_1_output_0", "inputIndexes": [ 4007 ], "outputIndexes": [ 4008 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 11549379952, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4008, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_338_output_0", "outputIndexes": [ 4009 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4008 ], "main_type": "NONE", "name": "/Shape_591_output_0", "outputIndexes": [ 4010 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4010 ], "main_type": "NONE", "name": "Shape16494", "outputIndexes": [ 4011 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4010 ], "main_type": "NONE", "name": "Rank16496", "outputIndexes": [ 4012 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4012, 4012 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16498", "outputIndexes": [ 4013 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4013 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16499", "outputIndexes": [ 4014 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4013, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16501", "outputIndexes": [ 4015 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4015 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16502", "outputIndexes": [ 4016 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4011, 4014, 4016, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice16505", "outputIndexes": [ 4017 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4017 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze16506", "outputIndexes": [ 4018 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4018 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16507", "outputIndexes": [ 4019 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4019, 4018 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16508", "outputIndexes": [ 4020 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4010, 4020, 4013 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_636_output_0", "outputIndexes": [ 4021 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4021, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_127_output_0", "outputIndexes": [ 4022 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4022, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1353_output_0", "outputIndexes": [ 4023 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4008, 4023, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_171_output_0", "outputIndexes": [ 4024 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4024 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_85_output_0", "outputIndexes": [ 4025 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4022, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1352_output_0", "outputIndexes": [ 4026 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4008, 20, 4026, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_170_output_0", "outputIndexes": [ 4027 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4025, 4027 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_550_output_0", "outputIndexes": [ 4028 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4028, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_339_output_0", "outputIndexes": [ 4029 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4009, 4029 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_211_output_0", "outputIndexes": [ 4030 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 3954 ], "outputIndexes": [ 7187 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7187 ], "outputIndexes": [ 7188 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/v_proj/Linear", "inputIndexes": [ 7188 ], "outputIndexes": [ 7189 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 11549380976, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7189 ], "outputIndexes": [ 7190 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7190 ], "outputIndexes": [ 4031 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3966, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1348_output_0", "outputIndexes": [ 4032 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3974, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1349_output_0", "outputIndexes": [ 4033 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4032, 4033, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_548_output_0", "outputIndexes": [ 4034 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4031, 4034 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_381_output_0", "outputIndexes": [ 4035 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4001, 4030, 4035, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_386_output_0", "outputIndexes": [ 4036 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4036 ], "outputIndexes": [ 7191 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7191 ], "outputIndexes": [ 7192 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/o_proj/Linear", "inputIndexes": [ 7192 ], "outputIndexes": [ 7193 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 11552330118, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7193 ], "outputIndexes": [ 7194 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7194 ], "outputIndexes": [ 4037 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 3953, 4037 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_213_output_0", "outputIndexes": [ 4038 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_42/Mul_1_output_0", "inputIndexes": [ 4038 ], "outputIndexes": [ 4039 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 11575923100, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4039 ], "outputIndexes": [ 7195 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7195 ], "outputIndexes": [ 7196 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/mlp/gate_proj/Linear", "inputIndexes": [ 7196 ], "outputIndexes": [ 7197 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 11575964060, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7197 ], "outputIndexes": [ 7198 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7198 ], "outputIndexes": [ 4040 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4040 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_42/Mul_output_0", "outputIndexes": [ 4041 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4039 ], "outputIndexes": [ 7199 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7199 ], "outputIndexes": [ 7200 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/mlp/up_proj/Linear", "inputIndexes": [ 7200 ], "outputIndexes": [ 7201 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 11649692082, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7201 ], "outputIndexes": [ 7202 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7202 ], "outputIndexes": [ 4042 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4041, 4042 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_42/Mul_output_0", "outputIndexes": [ 4043 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4043 ], "outputIndexes": [ 7203 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7203 ], "outputIndexes": [ 7204 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/mlp/down_proj/Linear", "inputIndexes": [ 7204 ], "outputIndexes": [ 7205 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 11723420104, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7205 ], "outputIndexes": [ 7206 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.42/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7206 ], "outputIndexes": [ 4044 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4038, 4044 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_214_output_0", "outputIndexes": [ 4045 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4045, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_387_output_0", "outputIndexes": [ 4046 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_43/Mul_1_output_0", "inputIndexes": [ 4046 ], "outputIndexes": [ 4047 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 11797148126, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4047 ], "outputIndexes": [ 7207 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7207 ], "outputIndexes": [ 7208 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/q_proj/Linear", "inputIndexes": [ 7208 ], "outputIndexes": [ 7209 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 11797189086, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7209 ], "outputIndexes": [ 7210 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7210 ], "outputIndexes": [ 4048 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4047 ], "main_type": "NONE", "name": "/Shape_602_output_0", "outputIndexes": [ 4049 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4049 ], "main_type": "NONE", "name": "Shape16837", "outputIndexes": [ 4050 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4049 ], "main_type": "NONE", "name": "Rank16839", "outputIndexes": [ 4051 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4051, 4051 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16841", "outputIndexes": [ 4052 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4052 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16842", "outputIndexes": [ 4053 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4052, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16844", "outputIndexes": [ 4054 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4054 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16845", "outputIndexes": [ 4055 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4050, 4053, 4055, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice16848", "outputIndexes": [ 4056 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4056 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze16849", "outputIndexes": [ 4057 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4057, 4057 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16851", "outputIndexes": [ 4058 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4049, 4058, 4052 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_648_output_0", "outputIndexes": [ 4059 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4059, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1376_output_0", "outputIndexes": [ 4060 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4052 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16859", "outputIndexes": [ 4061 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4054 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16862", "outputIndexes": [ 4062 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4050, 4061, 4062, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice16865", "outputIndexes": [ 4063 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4063 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze16866", "outputIndexes": [ 4064 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 4064 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16867", "outputIndexes": [ 4065 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4065, 4064 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16868", "outputIndexes": [ 4066 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4049, 4066, 4052 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_649_output_0", "outputIndexes": [ 4067 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4067, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1377_output_0", "outputIndexes": [ 4068 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4060, 4068, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_559_output_0", "outputIndexes": [ 4069 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4048, 4069 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_388_output_0", "outputIndexes": [ 4070 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4070 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_43/Cast_output_0", "outputIndexes": [ 4071 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_43/Mul_1_output_0", "inputIndexes": [ 4071 ], "outputIndexes": [ 4072 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 11820782068, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4072, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_344_output_0", "outputIndexes": [ 4073 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4072 ], "main_type": "NONE", "name": "/Shape_604_output_0", "outputIndexes": [ 4074 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4074 ], "main_type": "NONE", "name": "Shape16987", "outputIndexes": [ 4075 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4074 ], "main_type": "NONE", "name": "Rank16989", "outputIndexes": [ 4076 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4076, 4076 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16991", "outputIndexes": [ 4077 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4077 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16992", "outputIndexes": [ 4078 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4077, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16994", "outputIndexes": [ 4079 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4079 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16995", "outputIndexes": [ 4080 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4075, 4078, 4080, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice16998", "outputIndexes": [ 4081 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4081 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze16999", "outputIndexes": [ 4082 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4082 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17000", "outputIndexes": [ 4083 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4083, 4082 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17001", "outputIndexes": [ 4084 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4074, 4084, 4077 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_650_output_0", "outputIndexes": [ 4085 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4085, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_129_output_0", "outputIndexes": [ 4086 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4086, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1383_output_0", "outputIndexes": [ 4087 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4072, 4087, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_173_output_0", "outputIndexes": [ 4088 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4088 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_86_output_0", "outputIndexes": [ 4089 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4086, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1382_output_0", "outputIndexes": [ 4090 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4072, 20, 4090, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_172_output_0", "outputIndexes": [ 4091 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4089, 4091 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_562_output_0", "outputIndexes": [ 4092 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4092, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_345_output_0", "outputIndexes": [ 4093 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4073, 4093 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_215_output_0", "outputIndexes": [ 4094 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4047 ], "outputIndexes": [ 7211 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7211 ], "outputIndexes": [ 7212 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/k_proj/Linear", "inputIndexes": [ 7212 ], "outputIndexes": [ 7213 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 11820783092, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7213 ], "outputIndexes": [ 7214 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7214 ], "outputIndexes": [ 4095 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4059, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1378_output_0", "outputIndexes": [ 4096 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4067, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1379_output_0", "outputIndexes": [ 4097 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4096, 4097, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_560_output_0", "outputIndexes": [ 4098 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4095, 4098 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_389_output_0", "outputIndexes": [ 4099 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4099 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_43/Cast_output_0", "outputIndexes": [ 4100 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_43/Mul_1_output_0", "inputIndexes": [ 4100 ], "outputIndexes": [ 4101 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 11823732234, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4101, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_346_output_0", "outputIndexes": [ 4102 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4101 ], "main_type": "NONE", "name": "/Shape_605_output_0", "outputIndexes": [ 4103 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4103 ], "main_type": "NONE", "name": "Shape16883", "outputIndexes": [ 4104 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4103 ], "main_type": "NONE", "name": "Rank16885", "outputIndexes": [ 4105 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4105, 4105 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16887", "outputIndexes": [ 4106 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4106 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16888", "outputIndexes": [ 4107 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4106, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16890", "outputIndexes": [ 4108 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4108 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze16891", "outputIndexes": [ 4109 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4104, 4107, 4109, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice16894", "outputIndexes": [ 4110 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4110 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze16895", "outputIndexes": [ 4111 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4111 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16896", "outputIndexes": [ 4112 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4112, 4111 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp16897", "outputIndexes": [ 4113 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4103, 4113, 4106 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_651_output_0", "outputIndexes": [ 4114 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4114, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_130_output_0", "outputIndexes": [ 4115 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4115, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1385_output_0", "outputIndexes": [ 4116 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4101, 4116, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_175_output_0", "outputIndexes": [ 4117 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4117 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_87_output_0", "outputIndexes": [ 4118 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4115, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1384_output_0", "outputIndexes": [ 4119 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4101, 20, 4119, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_174_output_0", "outputIndexes": [ 4120 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4118, 4120 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_563_output_0", "outputIndexes": [ 4121 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4121, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_347_output_0", "outputIndexes": [ 4122 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4102, 4122 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_216_output_0", "outputIndexes": [ 4123 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4047 ], "outputIndexes": [ 7215 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7215 ], "outputIndexes": [ 7216 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/v_proj/Linear", "inputIndexes": [ 7216 ], "outputIndexes": [ 7217 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 11823733258, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7217 ], "outputIndexes": [ 7218 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7218 ], "outputIndexes": [ 4124 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4059, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1380_output_0", "outputIndexes": [ 4125 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4067, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1381_output_0", "outputIndexes": [ 4126 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4125, 4126, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_561_output_0", "outputIndexes": [ 4127 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4124, 4127 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_390_output_0", "outputIndexes": [ 4128 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4094, 4123, 4128, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_395_output_0", "outputIndexes": [ 4129 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4129 ], "outputIndexes": [ 7219 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7219 ], "outputIndexes": [ 7220 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/o_proj/Linear", "inputIndexes": [ 7220 ], "outputIndexes": [ 7221 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 11826682400, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7221 ], "outputIndexes": [ 7222 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7222 ], "outputIndexes": [ 4130 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4046, 4130 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_218_output_0", "outputIndexes": [ 4131 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_43/Mul_1_output_0", "inputIndexes": [ 4131 ], "outputIndexes": [ 4132 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 11850275382, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4132 ], "outputIndexes": [ 7223 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7223 ], "outputIndexes": [ 7224 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/mlp/gate_proj/Linear", "inputIndexes": [ 7224 ], "outputIndexes": [ 7225 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 11850316342, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7225 ], "outputIndexes": [ 7226 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7226 ], "outputIndexes": [ 4133 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4133 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_43/Mul_output_0", "outputIndexes": [ 4134 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4132 ], "outputIndexes": [ 7227 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7227 ], "outputIndexes": [ 7228 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/mlp/up_proj/Linear", "inputIndexes": [ 7228 ], "outputIndexes": [ 7229 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 11924044364, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7229 ], "outputIndexes": [ 7230 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7230 ], "outputIndexes": [ 4135 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4134, 4135 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_43/Mul_output_0", "outputIndexes": [ 4136 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4136 ], "outputIndexes": [ 7231 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7231 ], "outputIndexes": [ 7232 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/mlp/down_proj/Linear", "inputIndexes": [ 7232 ], "outputIndexes": [ 7233 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 11997772386, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7233 ], "outputIndexes": [ 7234 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.43/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7234 ], "outputIndexes": [ 4137 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4131, 4137 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_219_output_0", "outputIndexes": [ 4138 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4138, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_396_output_0", "outputIndexes": [ 4139 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_44/Mul_1_output_0", "inputIndexes": [ 4139 ], "outputIndexes": [ 4140 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 12071500408, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4140 ], "outputIndexes": [ 7235 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7235 ], "outputIndexes": [ 7236 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/q_proj/Linear", "inputIndexes": [ 7236 ], "outputIndexes": [ 7237 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 12071541368, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7237 ], "outputIndexes": [ 7238 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7238 ], "outputIndexes": [ 4141 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4140 ], "main_type": "NONE", "name": "/Shape_616_output_0", "outputIndexes": [ 4142 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4142 ], "main_type": "NONE", "name": "Shape17226", "outputIndexes": [ 4143 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4142 ], "main_type": "NONE", "name": "Rank17228", "outputIndexes": [ 4144 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4144, 4144 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17230", "outputIndexes": [ 4145 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4145 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze17231", "outputIndexes": [ 4146 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4145, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17233", "outputIndexes": [ 4147 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4147 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze17234", "outputIndexes": [ 4148 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4143, 4146, 4148, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice17237", "outputIndexes": [ 4149 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4149 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze17238", "outputIndexes": [ 4150 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4150, 4150 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17240", "outputIndexes": [ 4151 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4142, 4151, 4145 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_663_output_0", "outputIndexes": [ 4152 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4152, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1408_output_0", "outputIndexes": [ 4153 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4145 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze17248", "outputIndexes": [ 4154 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4147 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze17251", "outputIndexes": [ 4155 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4143, 4154, 4155, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice17254", "outputIndexes": [ 4156 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4156 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze17255", "outputIndexes": [ 4157 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 4157 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17256", "outputIndexes": [ 4158 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4158, 4157 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17257", "outputIndexes": [ 4159 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4142, 4159, 4145 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_664_output_0", "outputIndexes": [ 4160 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4160, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1409_output_0", "outputIndexes": [ 4161 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4153, 4161, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_572_output_0", "outputIndexes": [ 4162 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4141, 4162 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_397_output_0", "outputIndexes": [ 4163 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4163 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_44/Cast_output_0", "outputIndexes": [ 4164 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_44/Mul_1_output_0", "inputIndexes": [ 4164 ], "outputIndexes": [ 4165 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 12095134350, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4165, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_352_output_0", "outputIndexes": [ 4166 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4165 ], "main_type": "NONE", "name": "/Shape_618_output_0", "outputIndexes": [ 4167 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4167 ], "main_type": "NONE", "name": "Shape17376", "outputIndexes": [ 4168 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4167 ], "main_type": "NONE", "name": "Rank17378", "outputIndexes": [ 4169 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4169, 4169 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17380", "outputIndexes": [ 4170 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4170 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze17381", "outputIndexes": [ 4171 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4170, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17383", "outputIndexes": [ 4172 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4172 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze17384", "outputIndexes": [ 4173 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4168, 4171, 4173, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice17387", "outputIndexes": [ 4174 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4174 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze17388", "outputIndexes": [ 4175 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4175 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17389", "outputIndexes": [ 4176 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4176, 4175 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17390", "outputIndexes": [ 4177 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4167, 4177, 4170 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_665_output_0", "outputIndexes": [ 4178 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4178, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_132_output_0", "outputIndexes": [ 4179 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4179, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1415_output_0", "outputIndexes": [ 4180 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4165, 4180, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_177_output_0", "outputIndexes": [ 4181 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4181 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_88_output_0", "outputIndexes": [ 4182 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4179, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1414_output_0", "outputIndexes": [ 4183 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4165, 20, 4183, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_176_output_0", "outputIndexes": [ 4184 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4182, 4184 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_575_output_0", "outputIndexes": [ 4185 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4185, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_353_output_0", "outputIndexes": [ 4186 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4166, 4186 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_220_output_0", "outputIndexes": [ 4187 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4140 ], "outputIndexes": [ 7239 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7239 ], "outputIndexes": [ 7240 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/k_proj/Linear", "inputIndexes": [ 7240 ], "outputIndexes": [ 7241 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 12095135374, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7241 ], "outputIndexes": [ 7242 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7242 ], "outputIndexes": [ 4188 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4152, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1410_output_0", "outputIndexes": [ 4189 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4160, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1411_output_0", "outputIndexes": [ 4190 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4189, 4190, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_573_output_0", "outputIndexes": [ 4191 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4188, 4191 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_398_output_0", "outputIndexes": [ 4192 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4192 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_44/Cast_output_0", "outputIndexes": [ 4193 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_44/Mul_1_output_0", "inputIndexes": [ 4193 ], "outputIndexes": [ 4194 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 12098084516, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4194, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_354_output_0", "outputIndexes": [ 4195 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4194 ], "main_type": "NONE", "name": "/Shape_619_output_0", "outputIndexes": [ 4196 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4196 ], "main_type": "NONE", "name": "Shape17272", "outputIndexes": [ 4197 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4196 ], "main_type": "NONE", "name": "Rank17274", "outputIndexes": [ 4198 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4198, 4198 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17276", "outputIndexes": [ 4199 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4199 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze17277", "outputIndexes": [ 4200 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4199, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17279", "outputIndexes": [ 4201 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4201 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze17280", "outputIndexes": [ 4202 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4197, 4200, 4202, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice17283", "outputIndexes": [ 4203 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4203 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze17284", "outputIndexes": [ 4204 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4204 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17285", "outputIndexes": [ 4205 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4205, 4204 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17286", "outputIndexes": [ 4206 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4196, 4206, 4199 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_666_output_0", "outputIndexes": [ 4207 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4207, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_133_output_0", "outputIndexes": [ 4208 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4208, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1417_output_0", "outputIndexes": [ 4209 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4194, 4209, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_179_output_0", "outputIndexes": [ 4210 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4210 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_89_output_0", "outputIndexes": [ 4211 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4208, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1416_output_0", "outputIndexes": [ 4212 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4194, 20, 4212, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_178_output_0", "outputIndexes": [ 4213 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4211, 4213 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_576_output_0", "outputIndexes": [ 4214 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4214, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_355_output_0", "outputIndexes": [ 4215 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4195, 4215 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_221_output_0", "outputIndexes": [ 4216 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4140 ], "outputIndexes": [ 7243 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7243 ], "outputIndexes": [ 7244 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/v_proj/Linear", "inputIndexes": [ 7244 ], "outputIndexes": [ 7245 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 12098085540, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7245 ], "outputIndexes": [ 7246 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7246 ], "outputIndexes": [ 4217 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4152, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1412_output_0", "outputIndexes": [ 4218 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4160, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1413_output_0", "outputIndexes": [ 4219 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4218, 4219, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_574_output_0", "outputIndexes": [ 4220 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4217, 4220 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_399_output_0", "outputIndexes": [ 4221 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4187, 4216, 4221, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_404_output_0", "outputIndexes": [ 4222 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4222 ], "outputIndexes": [ 7247 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7247 ], "outputIndexes": [ 7248 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/o_proj/Linear", "inputIndexes": [ 7248 ], "outputIndexes": [ 7249 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 12101034682, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7249 ], "outputIndexes": [ 7250 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7250 ], "outputIndexes": [ 4223 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4139, 4223 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_223_output_0", "outputIndexes": [ 4224 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_44/Mul_1_output_0", "inputIndexes": [ 4224 ], "outputIndexes": [ 4225 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 12124627664, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4225 ], "outputIndexes": [ 7251 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7251 ], "outputIndexes": [ 7252 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/mlp/gate_proj/Linear", "inputIndexes": [ 7252 ], "outputIndexes": [ 7253 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 12124668624, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7253 ], "outputIndexes": [ 7254 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7254 ], "outputIndexes": [ 4226 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4226 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_44/Mul_output_0", "outputIndexes": [ 4227 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4225 ], "outputIndexes": [ 7255 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7255 ], "outputIndexes": [ 7256 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/mlp/up_proj/Linear", "inputIndexes": [ 7256 ], "outputIndexes": [ 7257 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 12198396646, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7257 ], "outputIndexes": [ 7258 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7258 ], "outputIndexes": [ 4228 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4227, 4228 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_44/Mul_output_0", "outputIndexes": [ 4229 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4229 ], "outputIndexes": [ 7259 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7259 ], "outputIndexes": [ 7260 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/mlp/down_proj/Linear", "inputIndexes": [ 7260 ], "outputIndexes": [ 7261 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 12272124668, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7261 ], "outputIndexes": [ 7262 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.44/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7262 ], "outputIndexes": [ 4230 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4224, 4230 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_224_output_0", "outputIndexes": [ 4231 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4231, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_405_output_0", "outputIndexes": [ 4232 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_45/Mul_1_output_0", "inputIndexes": [ 4232 ], "outputIndexes": [ 4233 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 12345852690, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4233 ], "outputIndexes": [ 7263 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7263 ], "outputIndexes": [ 7264 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/q_proj/Linear", "inputIndexes": [ 7264 ], "outputIndexes": [ 7265 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 12345893650, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7265 ], "outputIndexes": [ 7266 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7266 ], "outputIndexes": [ 4234 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4233 ], "main_type": "NONE", "name": "/Shape_630_output_0", "outputIndexes": [ 4235 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4235 ], "main_type": "NONE", "name": "Shape17615", "outputIndexes": [ 4236 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4235 ], "main_type": "NONE", "name": "Rank17617", "outputIndexes": [ 4237 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4237, 4237 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17619", "outputIndexes": [ 4238 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4238 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze17620", "outputIndexes": [ 4239 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4238, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17622", "outputIndexes": [ 4240 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4240 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze17623", "outputIndexes": [ 4241 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4236, 4239, 4241, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice17626", "outputIndexes": [ 4242 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4242 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze17627", "outputIndexes": [ 4243 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4243, 4243 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17629", "outputIndexes": [ 4244 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4235, 4244, 4238 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_678_output_0", "outputIndexes": [ 4245 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4245, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1440_output_0", "outputIndexes": [ 4246 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4238 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze17637", "outputIndexes": [ 4247 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4240 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze17640", "outputIndexes": [ 4248 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4236, 4247, 4248, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice17643", "outputIndexes": [ 4249 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4249 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze17644", "outputIndexes": [ 4250 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 4250 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17645", "outputIndexes": [ 4251 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4251, 4250 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17646", "outputIndexes": [ 4252 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4235, 4252, 4238 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_679_output_0", "outputIndexes": [ 4253 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4253, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1441_output_0", "outputIndexes": [ 4254 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4246, 4254, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_585_output_0", "outputIndexes": [ 4255 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4234, 4255 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_406_output_0", "outputIndexes": [ 4256 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4256 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_45/Cast_output_0", "outputIndexes": [ 4257 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_45/Mul_1_output_0", "inputIndexes": [ 4257 ], "outputIndexes": [ 4258 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 12369486632, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4258, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_360_output_0", "outputIndexes": [ 4259 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4258 ], "main_type": "NONE", "name": "/Shape_632_output_0", "outputIndexes": [ 4260 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4260 ], "main_type": "NONE", "name": "Shape17765", "outputIndexes": [ 4261 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4260 ], "main_type": "NONE", "name": "Rank17767", "outputIndexes": [ 4262 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4262, 4262 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17769", "outputIndexes": [ 4263 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4263 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze17770", "outputIndexes": [ 4264 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4263, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17772", "outputIndexes": [ 4265 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4265 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze17773", "outputIndexes": [ 4266 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4261, 4264, 4266, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice17776", "outputIndexes": [ 4267 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4267 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze17777", "outputIndexes": [ 4268 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4268 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17778", "outputIndexes": [ 4269 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4269, 4268 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17779", "outputIndexes": [ 4270 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4260, 4270, 4263 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_680_output_0", "outputIndexes": [ 4271 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4271, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_135_output_0", "outputIndexes": [ 4272 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4272, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1447_output_0", "outputIndexes": [ 4273 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4258, 4273, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_181_output_0", "outputIndexes": [ 4274 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4274 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_90_output_0", "outputIndexes": [ 4275 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4272, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1446_output_0", "outputIndexes": [ 4276 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4258, 20, 4276, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_180_output_0", "outputIndexes": [ 4277 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4275, 4277 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_588_output_0", "outputIndexes": [ 4278 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4278, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_361_output_0", "outputIndexes": [ 4279 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4259, 4279 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_225_output_0", "outputIndexes": [ 4280 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4233 ], "outputIndexes": [ 7267 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7267 ], "outputIndexes": [ 7268 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/k_proj/Linear", "inputIndexes": [ 7268 ], "outputIndexes": [ 7269 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 12369487656, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7269 ], "outputIndexes": [ 7270 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7270 ], "outputIndexes": [ 4281 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4245, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1442_output_0", "outputIndexes": [ 4282 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4253, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1443_output_0", "outputIndexes": [ 4283 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4282, 4283, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_586_output_0", "outputIndexes": [ 4284 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4281, 4284 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_407_output_0", "outputIndexes": [ 4285 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4285 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_45/Cast_output_0", "outputIndexes": [ 4286 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_45/Mul_1_output_0", "inputIndexes": [ 4286 ], "outputIndexes": [ 4287 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 12372436798, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4287, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_362_output_0", "outputIndexes": [ 4288 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4287 ], "main_type": "NONE", "name": "/Shape_633_output_0", "outputIndexes": [ 4289 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4289 ], "main_type": "NONE", "name": "Shape17661", "outputIndexes": [ 4290 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4289 ], "main_type": "NONE", "name": "Rank17663", "outputIndexes": [ 4291 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4291, 4291 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17665", "outputIndexes": [ 4292 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4292 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze17666", "outputIndexes": [ 4293 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4292, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17668", "outputIndexes": [ 4294 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4294 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze17669", "outputIndexes": [ 4295 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4290, 4293, 4295, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice17672", "outputIndexes": [ 4296 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4296 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze17673", "outputIndexes": [ 4297 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4297 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17674", "outputIndexes": [ 4298 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4298, 4297 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp17675", "outputIndexes": [ 4299 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4289, 4299, 4292 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_681_output_0", "outputIndexes": [ 4300 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4300, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_136_output_0", "outputIndexes": [ 4301 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4301, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1449_output_0", "outputIndexes": [ 4302 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4287, 4302, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_183_output_0", "outputIndexes": [ 4303 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4303 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_91_output_0", "outputIndexes": [ 4304 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4301, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1448_output_0", "outputIndexes": [ 4305 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4287, 20, 4305, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_182_output_0", "outputIndexes": [ 4306 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4304, 4306 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_589_output_0", "outputIndexes": [ 4307 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4307, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_363_output_0", "outputIndexes": [ 4308 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4288, 4308 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_226_output_0", "outputIndexes": [ 4309 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4233 ], "outputIndexes": [ 7271 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7271 ], "outputIndexes": [ 7272 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/v_proj/Linear", "inputIndexes": [ 7272 ], "outputIndexes": [ 7273 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 12372437822, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7273 ], "outputIndexes": [ 7274 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7274 ], "outputIndexes": [ 4310 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4245, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1444_output_0", "outputIndexes": [ 4311 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4253, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1445_output_0", "outputIndexes": [ 4312 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4311, 4312, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_587_output_0", "outputIndexes": [ 4313 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4310, 4313 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_408_output_0", "outputIndexes": [ 4314 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4280, 4309, 4314, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_413_output_0", "outputIndexes": [ 4315 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4315 ], "outputIndexes": [ 7275 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7275 ], "outputIndexes": [ 7276 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/o_proj/Linear", "inputIndexes": [ 7276 ], "outputIndexes": [ 7277 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 12375386964, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7277 ], "outputIndexes": [ 7278 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7278 ], "outputIndexes": [ 4316 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4232, 4316 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_228_output_0", "outputIndexes": [ 4317 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_45/Mul_1_output_0", "inputIndexes": [ 4317 ], "outputIndexes": [ 4318 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 12398979946, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4318 ], "outputIndexes": [ 7279 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7279 ], "outputIndexes": [ 7280 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/mlp/gate_proj/Linear", "inputIndexes": [ 7280 ], "outputIndexes": [ 7281 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 12399020906, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7281 ], "outputIndexes": [ 7282 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7282 ], "outputIndexes": [ 4319 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4319 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_45/Mul_output_0", "outputIndexes": [ 4320 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4318 ], "outputIndexes": [ 7283 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7283 ], "outputIndexes": [ 7284 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/mlp/up_proj/Linear", "inputIndexes": [ 7284 ], "outputIndexes": [ 7285 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 12472748928, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7285 ], "outputIndexes": [ 7286 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7286 ], "outputIndexes": [ 4321 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4320, 4321 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_45/Mul_output_0", "outputIndexes": [ 4322 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4322 ], "outputIndexes": [ 7287 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7287 ], "outputIndexes": [ 7288 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/mlp/down_proj/Linear", "inputIndexes": [ 7288 ], "outputIndexes": [ 7289 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 12546476950, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7289 ], "outputIndexes": [ 7290 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.45/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7290 ], "outputIndexes": [ 4323 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4317, 4323 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_229_output_0", "outputIndexes": [ 4324 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4324, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_414_output_0", "outputIndexes": [ 4325 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_46/Mul_1_output_0", "inputIndexes": [ 4325 ], "outputIndexes": [ 4326 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 12620204972, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4326 ], "outputIndexes": [ 7291 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7291 ], "outputIndexes": [ 7292 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/q_proj/Linear", "inputIndexes": [ 7292 ], "outputIndexes": [ 7293 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 12620245932, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7293 ], "outputIndexes": [ 7294 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7294 ], "outputIndexes": [ 4327 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4326 ], "main_type": "NONE", "name": "/Shape_644_output_0", "outputIndexes": [ 4328 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4328 ], "main_type": "NONE", "name": "Shape18004", "outputIndexes": [ 4329 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4328 ], "main_type": "NONE", "name": "Rank18006", "outputIndexes": [ 4330 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4330, 4330 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18008", "outputIndexes": [ 4331 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4331 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18009", "outputIndexes": [ 4332 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4331, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18011", "outputIndexes": [ 4333 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4333 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18012", "outputIndexes": [ 4334 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4329, 4332, 4334, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice18015", "outputIndexes": [ 4335 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4335 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze18016", "outputIndexes": [ 4336 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4336, 4336 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18018", "outputIndexes": [ 4337 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4328, 4337, 4331 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_693_output_0", "outputIndexes": [ 4338 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4338, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1472_output_0", "outputIndexes": [ 4339 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4331 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18026", "outputIndexes": [ 4340 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4333 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18029", "outputIndexes": [ 4341 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4329, 4340, 4341, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice18032", "outputIndexes": [ 4342 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4342 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze18033", "outputIndexes": [ 4343 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 4343 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18034", "outputIndexes": [ 4344 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4344, 4343 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18035", "outputIndexes": [ 4345 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4328, 4345, 4331 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_694_output_0", "outputIndexes": [ 4346 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4346, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1473_output_0", "outputIndexes": [ 4347 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4339, 4347, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_598_output_0", "outputIndexes": [ 4348 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4327, 4348 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_415_output_0", "outputIndexes": [ 4349 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4349 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_46/Cast_output_0", "outputIndexes": [ 4350 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_46/Mul_1_output_0", "inputIndexes": [ 4350 ], "outputIndexes": [ 4351 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 12643838914, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4351, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_368_output_0", "outputIndexes": [ 4352 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4351 ], "main_type": "NONE", "name": "/Shape_646_output_0", "outputIndexes": [ 4353 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4353 ], "main_type": "NONE", "name": "Shape18154", "outputIndexes": [ 4354 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4353 ], "main_type": "NONE", "name": "Rank18156", "outputIndexes": [ 4355 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4355, 4355 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18158", "outputIndexes": [ 4356 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4356 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18159", "outputIndexes": [ 4357 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4356, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18161", "outputIndexes": [ 4358 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4358 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18162", "outputIndexes": [ 4359 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4354, 4357, 4359, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice18165", "outputIndexes": [ 4360 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4360 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze18166", "outputIndexes": [ 4361 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4361 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18167", "outputIndexes": [ 4362 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4362, 4361 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18168", "outputIndexes": [ 4363 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4353, 4363, 4356 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_695_output_0", "outputIndexes": [ 4364 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4364, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_138_output_0", "outputIndexes": [ 4365 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4365, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1479_output_0", "outputIndexes": [ 4366 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4351, 4366, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_185_output_0", "outputIndexes": [ 4367 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4367 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_92_output_0", "outputIndexes": [ 4368 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4365, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1478_output_0", "outputIndexes": [ 4369 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4351, 20, 4369, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_184_output_0", "outputIndexes": [ 4370 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4368, 4370 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_601_output_0", "outputIndexes": [ 4371 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4371, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_369_output_0", "outputIndexes": [ 4372 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4352, 4372 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_230_output_0", "outputIndexes": [ 4373 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4326 ], "outputIndexes": [ 7295 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7295 ], "outputIndexes": [ 7296 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/k_proj/Linear", "inputIndexes": [ 7296 ], "outputIndexes": [ 7297 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 12643839938, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7297 ], "outputIndexes": [ 7298 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7298 ], "outputIndexes": [ 4374 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4338, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1474_output_0", "outputIndexes": [ 4375 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4346, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1475_output_0", "outputIndexes": [ 4376 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4375, 4376, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_599_output_0", "outputIndexes": [ 4377 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4374, 4377 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_416_output_0", "outputIndexes": [ 4378 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4378 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_46/Cast_output_0", "outputIndexes": [ 4379 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_46/Mul_1_output_0", "inputIndexes": [ 4379 ], "outputIndexes": [ 4380 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 12646789080, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4380, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_370_output_0", "outputIndexes": [ 4381 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4380 ], "main_type": "NONE", "name": "/Shape_647_output_0", "outputIndexes": [ 4382 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4382 ], "main_type": "NONE", "name": "Shape18050", "outputIndexes": [ 4383 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4382 ], "main_type": "NONE", "name": "Rank18052", "outputIndexes": [ 4384 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4384, 4384 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18054", "outputIndexes": [ 4385 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4385 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18055", "outputIndexes": [ 4386 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4385, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18057", "outputIndexes": [ 4387 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4387 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18058", "outputIndexes": [ 4388 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4383, 4386, 4388, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice18061", "outputIndexes": [ 4389 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4389 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze18062", "outputIndexes": [ 4390 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4390 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18063", "outputIndexes": [ 4391 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4391, 4390 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18064", "outputIndexes": [ 4392 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4382, 4392, 4385 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_696_output_0", "outputIndexes": [ 4393 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4393, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_139_output_0", "outputIndexes": [ 4394 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4394, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1481_output_0", "outputIndexes": [ 4395 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4380, 4395, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_187_output_0", "outputIndexes": [ 4396 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4396 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_93_output_0", "outputIndexes": [ 4397 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4394, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1480_output_0", "outputIndexes": [ 4398 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4380, 20, 4398, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_186_output_0", "outputIndexes": [ 4399 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4397, 4399 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_602_output_0", "outputIndexes": [ 4400 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4400, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_371_output_0", "outputIndexes": [ 4401 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4381, 4401 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_231_output_0", "outputIndexes": [ 4402 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4326 ], "outputIndexes": [ 7299 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7299 ], "outputIndexes": [ 7300 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/v_proj/Linear", "inputIndexes": [ 7300 ], "outputIndexes": [ 7301 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 12646790104, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7301 ], "outputIndexes": [ 7302 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7302 ], "outputIndexes": [ 4403 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4338, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1476_output_0", "outputIndexes": [ 4404 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4346, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1477_output_0", "outputIndexes": [ 4405 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4404, 4405, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_600_output_0", "outputIndexes": [ 4406 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4403, 4406 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_417_output_0", "outputIndexes": [ 4407 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4373, 4402, 4407, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_422_output_0", "outputIndexes": [ 4408 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4408 ], "outputIndexes": [ 7303 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7303 ], "outputIndexes": [ 7304 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/o_proj/Linear", "inputIndexes": [ 7304 ], "outputIndexes": [ 7305 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 12649739246, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7305 ], "outputIndexes": [ 7306 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7306 ], "outputIndexes": [ 4409 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4325, 4409 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_233_output_0", "outputIndexes": [ 4410 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_46/Mul_1_output_0", "inputIndexes": [ 4410 ], "outputIndexes": [ 4411 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 12673332228, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4411 ], "outputIndexes": [ 7307 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7307 ], "outputIndexes": [ 7308 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/mlp/gate_proj/Linear", "inputIndexes": [ 7308 ], "outputIndexes": [ 7309 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 12673373188, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7309 ], "outputIndexes": [ 7310 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7310 ], "outputIndexes": [ 4412 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4412 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_46/Mul_output_0", "outputIndexes": [ 4413 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4411 ], "outputIndexes": [ 7311 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7311 ], "outputIndexes": [ 7312 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/mlp/up_proj/Linear", "inputIndexes": [ 7312 ], "outputIndexes": [ 7313 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 12747101210, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7313 ], "outputIndexes": [ 7314 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7314 ], "outputIndexes": [ 4414 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4413, 4414 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_46/Mul_output_0", "outputIndexes": [ 4415 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4415 ], "outputIndexes": [ 7315 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7315 ], "outputIndexes": [ 7316 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/mlp/down_proj/Linear", "inputIndexes": [ 7316 ], "outputIndexes": [ 7317 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 12820829232, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7317 ], "outputIndexes": [ 7318 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.46/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7318 ], "outputIndexes": [ 4416 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4410, 4416 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_234_output_0", "outputIndexes": [ 4417 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4417, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_423_output_0", "outputIndexes": [ 4418 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_47/Mul_1_output_0", "inputIndexes": [ 4418 ], "outputIndexes": [ 4419 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 12894557254, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4419 ], "outputIndexes": [ 7319 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7319 ], "outputIndexes": [ 7320 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/q_proj/Linear", "inputIndexes": [ 7320 ], "outputIndexes": [ 7321 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 12894598214, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7321 ], "outputIndexes": [ 7322 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7322 ], "outputIndexes": [ 4420 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4419 ], "main_type": "NONE", "name": "/Shape_658_output_0", "outputIndexes": [ 4421 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4421 ], "main_type": "NONE", "name": "Shape18393", "outputIndexes": [ 4422 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4421 ], "main_type": "NONE", "name": "Rank18395", "outputIndexes": [ 4423 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4423, 4423 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18397", "outputIndexes": [ 4424 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4424 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18398", "outputIndexes": [ 4425 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4424, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18400", "outputIndexes": [ 4426 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4426 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18401", "outputIndexes": [ 4427 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4422, 4425, 4427, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice18404", "outputIndexes": [ 4428 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4428 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze18405", "outputIndexes": [ 4429 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4429, 4429 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18407", "outputIndexes": [ 4430 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4421, 4430, 4424 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_708_output_0", "outputIndexes": [ 4431 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4431, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1504_output_0", "outputIndexes": [ 4432 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4424 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18415", "outputIndexes": [ 4433 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4426 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18418", "outputIndexes": [ 4434 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4422, 4433, 4434, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice18421", "outputIndexes": [ 4435 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4435 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze18422", "outputIndexes": [ 4436 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 4436 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18423", "outputIndexes": [ 4437 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4437, 4436 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18424", "outputIndexes": [ 4438 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4421, 4438, 4424 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_709_output_0", "outputIndexes": [ 4439 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4439, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1505_output_0", "outputIndexes": [ 4440 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4432, 4440, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_611_output_0", "outputIndexes": [ 4441 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4420, 4441 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_424_output_0", "outputIndexes": [ 4442 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4442 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_47/Cast_output_0", "outputIndexes": [ 4443 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_47/Mul_1_output_0", "inputIndexes": [ 4443 ], "outputIndexes": [ 4444 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 12918191196, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4444, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_376_output_0", "outputIndexes": [ 4445 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4444 ], "main_type": "NONE", "name": "/Shape_660_output_0", "outputIndexes": [ 4446 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4446 ], "main_type": "NONE", "name": "Shape18543", "outputIndexes": [ 4447 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4446 ], "main_type": "NONE", "name": "Rank18545", "outputIndexes": [ 4448 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4448, 4448 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18547", "outputIndexes": [ 4449 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4449 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18548", "outputIndexes": [ 4450 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4449, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18550", "outputIndexes": [ 4451 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4451 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18551", "outputIndexes": [ 4452 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4447, 4450, 4452, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice18554", "outputIndexes": [ 4453 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4453 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze18555", "outputIndexes": [ 4454 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4454 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18556", "outputIndexes": [ 4455 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4455, 4454 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18557", "outputIndexes": [ 4456 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4446, 4456, 4449 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_710_output_0", "outputIndexes": [ 4457 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4457, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_141_output_0", "outputIndexes": [ 4458 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4458, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1511_output_0", "outputIndexes": [ 4459 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4444, 4459, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_189_output_0", "outputIndexes": [ 4460 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4460 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_94_output_0", "outputIndexes": [ 4461 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4458, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1510_output_0", "outputIndexes": [ 4462 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4444, 20, 4462, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_188_output_0", "outputIndexes": [ 4463 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4461, 4463 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_614_output_0", "outputIndexes": [ 4464 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4464, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_377_output_0", "outputIndexes": [ 4465 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4445, 4465 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_235_output_0", "outputIndexes": [ 4466 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4419 ], "outputIndexes": [ 7323 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7323 ], "outputIndexes": [ 7324 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/k_proj/Linear", "inputIndexes": [ 7324 ], "outputIndexes": [ 7325 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 12918192220, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7325 ], "outputIndexes": [ 7326 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7326 ], "outputIndexes": [ 4467 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4431, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1506_output_0", "outputIndexes": [ 4468 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4439, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1507_output_0", "outputIndexes": [ 4469 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4468, 4469, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_612_output_0", "outputIndexes": [ 4470 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4467, 4470 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_425_output_0", "outputIndexes": [ 4471 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4471 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_47/Cast_output_0", "outputIndexes": [ 4472 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_47/Mul_1_output_0", "inputIndexes": [ 4472 ], "outputIndexes": [ 4473 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 12921141362, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4473, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_378_output_0", "outputIndexes": [ 4474 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4473 ], "main_type": "NONE", "name": "/Shape_661_output_0", "outputIndexes": [ 4475 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4475 ], "main_type": "NONE", "name": "Shape18439", "outputIndexes": [ 4476 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4475 ], "main_type": "NONE", "name": "Rank18441", "outputIndexes": [ 4477 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4477, 4477 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18443", "outputIndexes": [ 4478 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4478 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18444", "outputIndexes": [ 4479 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4478, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18446", "outputIndexes": [ 4480 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4480 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18447", "outputIndexes": [ 4481 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4476, 4479, 4481, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice18450", "outputIndexes": [ 4482 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4482 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze18451", "outputIndexes": [ 4483 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4483 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18452", "outputIndexes": [ 4484 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4484, 4483 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18453", "outputIndexes": [ 4485 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4475, 4485, 4478 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_711_output_0", "outputIndexes": [ 4486 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4486, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_142_output_0", "outputIndexes": [ 4487 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4487, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1513_output_0", "outputIndexes": [ 4488 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4473, 4488, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_191_output_0", "outputIndexes": [ 4489 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4489 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_95_output_0", "outputIndexes": [ 4490 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4487, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1512_output_0", "outputIndexes": [ 4491 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4473, 20, 4491, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_190_output_0", "outputIndexes": [ 4492 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4490, 4492 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_615_output_0", "outputIndexes": [ 4493 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4493, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_379_output_0", "outputIndexes": [ 4494 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4474, 4494 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_236_output_0", "outputIndexes": [ 4495 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4419 ], "outputIndexes": [ 7327 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7327 ], "outputIndexes": [ 7328 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/v_proj/Linear", "inputIndexes": [ 7328 ], "outputIndexes": [ 7329 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 12921142386, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7329 ], "outputIndexes": [ 7330 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7330 ], "outputIndexes": [ 4496 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4431, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1508_output_0", "outputIndexes": [ 4497 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4439, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1509_output_0", "outputIndexes": [ 4498 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4497, 4498, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_613_output_0", "outputIndexes": [ 4499 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4496, 4499 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_426_output_0", "outputIndexes": [ 4500 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4466, 4495, 4500, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_431_output_0", "outputIndexes": [ 4501 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4501 ], "outputIndexes": [ 7331 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7331 ], "outputIndexes": [ 7332 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/o_proj/Linear", "inputIndexes": [ 7332 ], "outputIndexes": [ 7333 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 12924091528, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7333 ], "outputIndexes": [ 7334 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7334 ], "outputIndexes": [ 4502 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4418, 4502 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_238_output_0", "outputIndexes": [ 4503 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_47/Mul_1_output_0", "inputIndexes": [ 4503 ], "outputIndexes": [ 4504 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 12947684510, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4504 ], "outputIndexes": [ 7335 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7335 ], "outputIndexes": [ 7336 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/mlp/gate_proj/Linear", "inputIndexes": [ 7336 ], "outputIndexes": [ 7337 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 12947725470, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7337 ], "outputIndexes": [ 7338 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7338 ], "outputIndexes": [ 4505 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4505 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_47/Mul_output_0", "outputIndexes": [ 4506 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4504 ], "outputIndexes": [ 7339 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7339 ], "outputIndexes": [ 7340 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/mlp/up_proj/Linear", "inputIndexes": [ 7340 ], "outputIndexes": [ 7341 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 13021453492, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7341 ], "outputIndexes": [ 7342 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7342 ], "outputIndexes": [ 4507 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4506, 4507 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_47/Mul_output_0", "outputIndexes": [ 4508 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4508 ], "outputIndexes": [ 7343 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7343 ], "outputIndexes": [ 7344 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/mlp/down_proj/Linear", "inputIndexes": [ 7344 ], "outputIndexes": [ 7345 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 13095181514, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7345 ], "outputIndexes": [ 7346 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.47/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7346 ], "outputIndexes": [ 4509 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4503, 4509 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_239_output_0", "outputIndexes": [ 4510 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4510, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_432_output_0", "outputIndexes": [ 4511 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_48/Mul_1_output_0", "inputIndexes": [ 4511 ], "outputIndexes": [ 4512 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 13168909536, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4512 ], "outputIndexes": [ 7347 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7347 ], "outputIndexes": [ 7348 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/q_proj/Linear", "inputIndexes": [ 7348 ], "outputIndexes": [ 7349 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 13168950496, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7349 ], "outputIndexes": [ 7350 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7350 ], "outputIndexes": [ 4513 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4512 ], "main_type": "NONE", "name": "/Shape_672_output_0", "outputIndexes": [ 4514 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4514 ], "main_type": "NONE", "name": "Shape18782", "outputIndexes": [ 4515 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4514 ], "main_type": "NONE", "name": "Rank18784", "outputIndexes": [ 4516 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4516, 4516 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18786", "outputIndexes": [ 4517 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4517 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18787", "outputIndexes": [ 4518 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4517, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18789", "outputIndexes": [ 4519 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4519 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18790", "outputIndexes": [ 4520 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4515, 4518, 4520, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice18793", "outputIndexes": [ 4521 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4521 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze18794", "outputIndexes": [ 4522 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4522, 4522 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18796", "outputIndexes": [ 4523 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4514, 4523, 4517 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_723_output_0", "outputIndexes": [ 4524 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4524, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1536_output_0", "outputIndexes": [ 4525 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4517 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18804", "outputIndexes": [ 4526 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4519 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18807", "outputIndexes": [ 4527 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4515, 4526, 4527, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice18810", "outputIndexes": [ 4528 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4528 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze18811", "outputIndexes": [ 4529 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 4529 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18812", "outputIndexes": [ 4530 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4530, 4529 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18813", "outputIndexes": [ 4531 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4514, 4531, 4517 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_724_output_0", "outputIndexes": [ 4532 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4532, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1537_output_0", "outputIndexes": [ 4533 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4525, 4533, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_624_output_0", "outputIndexes": [ 4534 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4513, 4534 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_433_output_0", "outputIndexes": [ 4535 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4535 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_48/Cast_output_0", "outputIndexes": [ 4536 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_48/Mul_1_output_0", "inputIndexes": [ 4536 ], "outputIndexes": [ 4537 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 13192543478, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4537, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_384_output_0", "outputIndexes": [ 4538 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4537 ], "main_type": "NONE", "name": "/Shape_674_output_0", "outputIndexes": [ 4539 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4539 ], "main_type": "NONE", "name": "Shape18932", "outputIndexes": [ 4540 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4539 ], "main_type": "NONE", "name": "Rank18934", "outputIndexes": [ 4541 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4541, 4541 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18936", "outputIndexes": [ 4542 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4542 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18937", "outputIndexes": [ 4543 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4542, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18939", "outputIndexes": [ 4544 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4544 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18940", "outputIndexes": [ 4545 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4540, 4543, 4545, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice18943", "outputIndexes": [ 4546 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4546 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze18944", "outputIndexes": [ 4547 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4547 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18945", "outputIndexes": [ 4548 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4548, 4547 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18946", "outputIndexes": [ 4549 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4539, 4549, 4542 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_725_output_0", "outputIndexes": [ 4550 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4550, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_144_output_0", "outputIndexes": [ 4551 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4551, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1543_output_0", "outputIndexes": [ 4552 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4537, 4552, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_193_output_0", "outputIndexes": [ 4553 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4553 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_96_output_0", "outputIndexes": [ 4554 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4551, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1542_output_0", "outputIndexes": [ 4555 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4537, 20, 4555, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_192_output_0", "outputIndexes": [ 4556 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4554, 4556 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_627_output_0", "outputIndexes": [ 4557 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4557, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_385_output_0", "outputIndexes": [ 4558 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4538, 4558 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_240_output_0", "outputIndexes": [ 4559 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4512 ], "outputIndexes": [ 7351 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7351 ], "outputIndexes": [ 7352 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/k_proj/Linear", "inputIndexes": [ 7352 ], "outputIndexes": [ 7353 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 13192544502, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7353 ], "outputIndexes": [ 7354 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7354 ], "outputIndexes": [ 4560 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4524, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1538_output_0", "outputIndexes": [ 4561 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4532, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1539_output_0", "outputIndexes": [ 4562 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4561, 4562, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_625_output_0", "outputIndexes": [ 4563 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4560, 4563 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_434_output_0", "outputIndexes": [ 4564 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4564 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_48/Cast_output_0", "outputIndexes": [ 4565 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_48/Mul_1_output_0", "inputIndexes": [ 4565 ], "outputIndexes": [ 4566 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 13195493644, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4566, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_386_output_0", "outputIndexes": [ 4567 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4566 ], "main_type": "NONE", "name": "/Shape_675_output_0", "outputIndexes": [ 4568 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4568 ], "main_type": "NONE", "name": "Shape18828", "outputIndexes": [ 4569 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4568 ], "main_type": "NONE", "name": "Rank18830", "outputIndexes": [ 4570 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4570, 4570 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18832", "outputIndexes": [ 4571 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4571 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18833", "outputIndexes": [ 4572 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4571, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18835", "outputIndexes": [ 4573 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4573 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze18836", "outputIndexes": [ 4574 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4569, 4572, 4574, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice18839", "outputIndexes": [ 4575 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4575 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze18840", "outputIndexes": [ 4576 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4576 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18841", "outputIndexes": [ 4577 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4577, 4576 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp18842", "outputIndexes": [ 4578 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4568, 4578, 4571 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_726_output_0", "outputIndexes": [ 4579 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4579, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_145_output_0", "outputIndexes": [ 4580 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4580, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1545_output_0", "outputIndexes": [ 4581 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4566, 4581, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_195_output_0", "outputIndexes": [ 4582 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4582 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_97_output_0", "outputIndexes": [ 4583 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4580, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1544_output_0", "outputIndexes": [ 4584 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4566, 20, 4584, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_194_output_0", "outputIndexes": [ 4585 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4583, 4585 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_628_output_0", "outputIndexes": [ 4586 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4586, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_387_output_0", "outputIndexes": [ 4587 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4567, 4587 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_241_output_0", "outputIndexes": [ 4588 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4512 ], "outputIndexes": [ 7355 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7355 ], "outputIndexes": [ 7356 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/v_proj/Linear", "inputIndexes": [ 7356 ], "outputIndexes": [ 7357 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 13195494668, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7357 ], "outputIndexes": [ 7358 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7358 ], "outputIndexes": [ 4589 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4524, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1540_output_0", "outputIndexes": [ 4590 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4532, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1541_output_0", "outputIndexes": [ 4591 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4590, 4591, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_626_output_0", "outputIndexes": [ 4592 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4589, 4592 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_435_output_0", "outputIndexes": [ 4593 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4559, 4588, 4593, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_440_output_0", "outputIndexes": [ 4594 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4594 ], "outputIndexes": [ 7359 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7359 ], "outputIndexes": [ 7360 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/o_proj/Linear", "inputIndexes": [ 7360 ], "outputIndexes": [ 7361 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 13198443810, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7361 ], "outputIndexes": [ 7362 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7362 ], "outputIndexes": [ 4595 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4511, 4595 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_243_output_0", "outputIndexes": [ 4596 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_48/Mul_1_output_0", "inputIndexes": [ 4596 ], "outputIndexes": [ 4597 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 13222036792, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4597 ], "outputIndexes": [ 7363 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7363 ], "outputIndexes": [ 7364 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/mlp/gate_proj/Linear", "inputIndexes": [ 7364 ], "outputIndexes": [ 7365 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 13222077752, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7365 ], "outputIndexes": [ 7366 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7366 ], "outputIndexes": [ 4598 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4598 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_48/Mul_output_0", "outputIndexes": [ 4599 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4597 ], "outputIndexes": [ 7367 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7367 ], "outputIndexes": [ 7368 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/mlp/up_proj/Linear", "inputIndexes": [ 7368 ], "outputIndexes": [ 7369 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 13295805774, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7369 ], "outputIndexes": [ 7370 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7370 ], "outputIndexes": [ 4600 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4599, 4600 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_48/Mul_output_0", "outputIndexes": [ 4601 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4601 ], "outputIndexes": [ 7371 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7371 ], "outputIndexes": [ 7372 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/mlp/down_proj/Linear", "inputIndexes": [ 7372 ], "outputIndexes": [ 7373 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 13369533796, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7373 ], "outputIndexes": [ 7374 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.48/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7374 ], "outputIndexes": [ 4602 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4596, 4602 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_244_output_0", "outputIndexes": [ 4603 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4603, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_441_output_0", "outputIndexes": [ 4604 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_49/Mul_1_output_0", "inputIndexes": [ 4604 ], "outputIndexes": [ 4605 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 13443261818, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4605 ], "outputIndexes": [ 7375 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7375 ], "outputIndexes": [ 7376 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/q_proj/Linear", "inputIndexes": [ 7376 ], "outputIndexes": [ 7377 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 13443302778, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7377 ], "outputIndexes": [ 7378 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7378 ], "outputIndexes": [ 4606 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4605 ], "main_type": "NONE", "name": "/Shape_686_output_0", "outputIndexes": [ 4607 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4607 ], "main_type": "NONE", "name": "Shape19171", "outputIndexes": [ 4608 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4607 ], "main_type": "NONE", "name": "Rank19173", "outputIndexes": [ 4609 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4609, 4609 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19175", "outputIndexes": [ 4610 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4610 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19176", "outputIndexes": [ 4611 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4610, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19178", "outputIndexes": [ 4612 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4612 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19179", "outputIndexes": [ 4613 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4608, 4611, 4613, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice19182", "outputIndexes": [ 4614 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4614 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze19183", "outputIndexes": [ 4615 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4615, 4615 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19185", "outputIndexes": [ 4616 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4607, 4616, 4610 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_738_output_0", "outputIndexes": [ 4617 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4617, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1568_output_0", "outputIndexes": [ 4618 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4610 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19193", "outputIndexes": [ 4619 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4612 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19196", "outputIndexes": [ 4620 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4608, 4619, 4620, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice19199", "outputIndexes": [ 4621 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4621 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze19200", "outputIndexes": [ 4622 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 4622 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19201", "outputIndexes": [ 4623 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4623, 4622 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19202", "outputIndexes": [ 4624 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4607, 4624, 4610 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_739_output_0", "outputIndexes": [ 4625 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4625, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1569_output_0", "outputIndexes": [ 4626 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4618, 4626, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_637_output_0", "outputIndexes": [ 4627 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4606, 4627 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_442_output_0", "outputIndexes": [ 4628 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4628 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_49/Cast_output_0", "outputIndexes": [ 4629 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_49/Mul_1_output_0", "inputIndexes": [ 4629 ], "outputIndexes": [ 4630 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 13466895760, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4630, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_392_output_0", "outputIndexes": [ 4631 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4630 ], "main_type": "NONE", "name": "/Shape_688_output_0", "outputIndexes": [ 4632 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4632 ], "main_type": "NONE", "name": "Shape19321", "outputIndexes": [ 4633 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4632 ], "main_type": "NONE", "name": "Rank19323", "outputIndexes": [ 4634 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4634, 4634 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19325", "outputIndexes": [ 4635 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4635 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19326", "outputIndexes": [ 4636 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4635, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19328", "outputIndexes": [ 4637 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4637 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19329", "outputIndexes": [ 4638 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4633, 4636, 4638, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice19332", "outputIndexes": [ 4639 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4639 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze19333", "outputIndexes": [ 4640 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4640 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19334", "outputIndexes": [ 4641 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4641, 4640 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19335", "outputIndexes": [ 4642 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4632, 4642, 4635 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_740_output_0", "outputIndexes": [ 4643 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4643, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_147_output_0", "outputIndexes": [ 4644 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4644, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1575_output_0", "outputIndexes": [ 4645 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4630, 4645, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_197_output_0", "outputIndexes": [ 4646 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4646 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_98_output_0", "outputIndexes": [ 4647 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4644, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1574_output_0", "outputIndexes": [ 4648 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4630, 20, 4648, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_196_output_0", "outputIndexes": [ 4649 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4647, 4649 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_640_output_0", "outputIndexes": [ 4650 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4650, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_393_output_0", "outputIndexes": [ 4651 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4631, 4651 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_245_output_0", "outputIndexes": [ 4652 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4605 ], "outputIndexes": [ 7379 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7379 ], "outputIndexes": [ 7380 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/k_proj/Linear", "inputIndexes": [ 7380 ], "outputIndexes": [ 7381 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 13466896784, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7381 ], "outputIndexes": [ 7382 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7382 ], "outputIndexes": [ 4653 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4617, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1570_output_0", "outputIndexes": [ 4654 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4625, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1571_output_0", "outputIndexes": [ 4655 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4654, 4655, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_638_output_0", "outputIndexes": [ 4656 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4653, 4656 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_443_output_0", "outputIndexes": [ 4657 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4657 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_49/Cast_output_0", "outputIndexes": [ 4658 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_49/Mul_1_output_0", "inputIndexes": [ 4658 ], "outputIndexes": [ 4659 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 13469845926, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4659, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_394_output_0", "outputIndexes": [ 4660 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4659 ], "main_type": "NONE", "name": "/Shape_689_output_0", "outputIndexes": [ 4661 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4661 ], "main_type": "NONE", "name": "Shape19217", "outputIndexes": [ 4662 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4661 ], "main_type": "NONE", "name": "Rank19219", "outputIndexes": [ 4663 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4663, 4663 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19221", "outputIndexes": [ 4664 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4664 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19222", "outputIndexes": [ 4665 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4664, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19224", "outputIndexes": [ 4666 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4666 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19225", "outputIndexes": [ 4667 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4662, 4665, 4667, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice19228", "outputIndexes": [ 4668 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4668 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze19229", "outputIndexes": [ 4669 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4669 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19230", "outputIndexes": [ 4670 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4670, 4669 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19231", "outputIndexes": [ 4671 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4661, 4671, 4664 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_741_output_0", "outputIndexes": [ 4672 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4672, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_148_output_0", "outputIndexes": [ 4673 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4673, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1577_output_0", "outputIndexes": [ 4674 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4659, 4674, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_199_output_0", "outputIndexes": [ 4675 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4675 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_99_output_0", "outputIndexes": [ 4676 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4673, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1576_output_0", "outputIndexes": [ 4677 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4659, 20, 4677, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_198_output_0", "outputIndexes": [ 4678 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4676, 4678 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_641_output_0", "outputIndexes": [ 4679 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4679, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_395_output_0", "outputIndexes": [ 4680 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4660, 4680 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_246_output_0", "outputIndexes": [ 4681 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4605 ], "outputIndexes": [ 7383 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7383 ], "outputIndexes": [ 7384 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/v_proj/Linear", "inputIndexes": [ 7384 ], "outputIndexes": [ 7385 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 13469846950, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7385 ], "outputIndexes": [ 7386 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7386 ], "outputIndexes": [ 4682 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4617, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1572_output_0", "outputIndexes": [ 4683 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4625, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1573_output_0", "outputIndexes": [ 4684 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4683, 4684, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_639_output_0", "outputIndexes": [ 4685 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4682, 4685 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_444_output_0", "outputIndexes": [ 4686 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4652, 4681, 4686, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_449_output_0", "outputIndexes": [ 4687 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4687 ], "outputIndexes": [ 7387 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7387 ], "outputIndexes": [ 7388 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/o_proj/Linear", "inputIndexes": [ 7388 ], "outputIndexes": [ 7389 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 13472796092, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7389 ], "outputIndexes": [ 7390 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7390 ], "outputIndexes": [ 4688 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4604, 4688 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_248_output_0", "outputIndexes": [ 4689 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_49/Mul_1_output_0", "inputIndexes": [ 4689 ], "outputIndexes": [ 4690 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 13496389074, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4690 ], "outputIndexes": [ 7391 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7391 ], "outputIndexes": [ 7392 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/mlp/gate_proj/Linear", "inputIndexes": [ 7392 ], "outputIndexes": [ 7393 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 13496430034, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7393 ], "outputIndexes": [ 7394 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7394 ], "outputIndexes": [ 4691 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4691 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_49/Mul_output_0", "outputIndexes": [ 4692 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4690 ], "outputIndexes": [ 7395 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7395 ], "outputIndexes": [ 7396 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/mlp/up_proj/Linear", "inputIndexes": [ 7396 ], "outputIndexes": [ 7397 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 13570158056, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7397 ], "outputIndexes": [ 7398 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7398 ], "outputIndexes": [ 4693 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4692, 4693 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_49/Mul_output_0", "outputIndexes": [ 4694 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4694 ], "outputIndexes": [ 7399 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7399 ], "outputIndexes": [ 7400 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/mlp/down_proj/Linear", "inputIndexes": [ 7400 ], "outputIndexes": [ 7401 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 13643886078, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7401 ], "outputIndexes": [ 7402 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.49/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7402 ], "outputIndexes": [ 4695 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4689, 4695 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_249_output_0", "outputIndexes": [ 4696 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4696, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_450_output_0", "outputIndexes": [ 4697 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_50/Mul_1_output_0", "inputIndexes": [ 4697 ], "outputIndexes": [ 4698 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 13717614100, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4698 ], "outputIndexes": [ 7403 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7403 ], "outputIndexes": [ 7404 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/q_proj/Linear", "inputIndexes": [ 7404 ], "outputIndexes": [ 7405 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 13717655060, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7405 ], "outputIndexes": [ 7406 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7406 ], "outputIndexes": [ 4699 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4698 ], "main_type": "NONE", "name": "/Shape_700_output_0", "outputIndexes": [ 4700 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4700 ], "main_type": "NONE", "name": "Shape19560", "outputIndexes": [ 4701 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4700 ], "main_type": "NONE", "name": "Rank19562", "outputIndexes": [ 4702 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4702, 4702 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19564", "outputIndexes": [ 4703 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4703 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19565", "outputIndexes": [ 4704 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4703, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19567", "outputIndexes": [ 4705 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4705 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19568", "outputIndexes": [ 4706 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4701, 4704, 4706, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice19571", "outputIndexes": [ 4707 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4707 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze19572", "outputIndexes": [ 4708 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4708, 4708 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19574", "outputIndexes": [ 4709 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4700, 4709, 4703 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_753_output_0", "outputIndexes": [ 4710 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4710, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1600_output_0", "outputIndexes": [ 4711 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4703 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19582", "outputIndexes": [ 4712 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4705 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19585", "outputIndexes": [ 4713 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4701, 4712, 4713, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice19588", "outputIndexes": [ 4714 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4714 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze19589", "outputIndexes": [ 4715 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 4715 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19590", "outputIndexes": [ 4716 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4716, 4715 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19591", "outputIndexes": [ 4717 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4700, 4717, 4703 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_754_output_0", "outputIndexes": [ 4718 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4718, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1601_output_0", "outputIndexes": [ 4719 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4711, 4719, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_650_output_0", "outputIndexes": [ 4720 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4699, 4720 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_451_output_0", "outputIndexes": [ 4721 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4721 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_50/Cast_output_0", "outputIndexes": [ 4722 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_50/Mul_1_output_0", "inputIndexes": [ 4722 ], "outputIndexes": [ 4723 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 13741248042, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4723, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_400_output_0", "outputIndexes": [ 4724 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4723 ], "main_type": "NONE", "name": "/Shape_702_output_0", "outputIndexes": [ 4725 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4725 ], "main_type": "NONE", "name": "Shape19710", "outputIndexes": [ 4726 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4725 ], "main_type": "NONE", "name": "Rank19712", "outputIndexes": [ 4727 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4727, 4727 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19714", "outputIndexes": [ 4728 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4728 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19715", "outputIndexes": [ 4729 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4728, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19717", "outputIndexes": [ 4730 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4730 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19718", "outputIndexes": [ 4731 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4726, 4729, 4731, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice19721", "outputIndexes": [ 4732 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4732 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze19722", "outputIndexes": [ 4733 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4733 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19723", "outputIndexes": [ 4734 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4734, 4733 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19724", "outputIndexes": [ 4735 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4725, 4735, 4728 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_755_output_0", "outputIndexes": [ 4736 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4736, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_150_output_0", "outputIndexes": [ 4737 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4737, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1607_output_0", "outputIndexes": [ 4738 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4723, 4738, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_201_output_0", "outputIndexes": [ 4739 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4739 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_100_output_0", "outputIndexes": [ 4740 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4737, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1606_output_0", "outputIndexes": [ 4741 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4723, 20, 4741, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_200_output_0", "outputIndexes": [ 4742 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4740, 4742 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_653_output_0", "outputIndexes": [ 4743 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4743, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_401_output_0", "outputIndexes": [ 4744 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4724, 4744 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_250_output_0", "outputIndexes": [ 4745 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4698 ], "outputIndexes": [ 7407 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7407 ], "outputIndexes": [ 7408 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/k_proj/Linear", "inputIndexes": [ 7408 ], "outputIndexes": [ 7409 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 13741249066, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7409 ], "outputIndexes": [ 7410 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7410 ], "outputIndexes": [ 4746 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4710, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1602_output_0", "outputIndexes": [ 4747 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4718, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1603_output_0", "outputIndexes": [ 4748 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4747, 4748, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_651_output_0", "outputIndexes": [ 4749 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4746, 4749 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_452_output_0", "outputIndexes": [ 4750 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4750 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_50/Cast_output_0", "outputIndexes": [ 4751 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_50/Mul_1_output_0", "inputIndexes": [ 4751 ], "outputIndexes": [ 4752 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 13744198208, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4752, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_402_output_0", "outputIndexes": [ 4753 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4752 ], "main_type": "NONE", "name": "/Shape_703_output_0", "outputIndexes": [ 4754 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4754 ], "main_type": "NONE", "name": "Shape19606", "outputIndexes": [ 4755 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4754 ], "main_type": "NONE", "name": "Rank19608", "outputIndexes": [ 4756 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4756, 4756 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19610", "outputIndexes": [ 4757 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4757 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19611", "outputIndexes": [ 4758 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4757, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19613", "outputIndexes": [ 4759 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4759 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19614", "outputIndexes": [ 4760 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4755, 4758, 4760, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice19617", "outputIndexes": [ 4761 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4761 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze19618", "outputIndexes": [ 4762 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4762 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19619", "outputIndexes": [ 4763 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4763, 4762 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19620", "outputIndexes": [ 4764 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4754, 4764, 4757 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_756_output_0", "outputIndexes": [ 4765 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4765, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_151_output_0", "outputIndexes": [ 4766 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4766, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1609_output_0", "outputIndexes": [ 4767 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4752, 4767, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_203_output_0", "outputIndexes": [ 4768 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4768 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_101_output_0", "outputIndexes": [ 4769 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4766, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1608_output_0", "outputIndexes": [ 4770 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4752, 20, 4770, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_202_output_0", "outputIndexes": [ 4771 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4769, 4771 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_654_output_0", "outputIndexes": [ 4772 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4772, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_403_output_0", "outputIndexes": [ 4773 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4753, 4773 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_251_output_0", "outputIndexes": [ 4774 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4698 ], "outputIndexes": [ 7411 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7411 ], "outputIndexes": [ 7412 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/v_proj/Linear", "inputIndexes": [ 7412 ], "outputIndexes": [ 7413 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 13744199232, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7413 ], "outputIndexes": [ 7414 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7414 ], "outputIndexes": [ 4775 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4710, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1604_output_0", "outputIndexes": [ 4776 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4718, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1605_output_0", "outputIndexes": [ 4777 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4776, 4777, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_652_output_0", "outputIndexes": [ 4778 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4775, 4778 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_453_output_0", "outputIndexes": [ 4779 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4745, 4774, 4779, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_458_output_0", "outputIndexes": [ 4780 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4780 ], "outputIndexes": [ 7415 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7415 ], "outputIndexes": [ 7416 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/o_proj/Linear", "inputIndexes": [ 7416 ], "outputIndexes": [ 7417 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 13747148374, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7417 ], "outputIndexes": [ 7418 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7418 ], "outputIndexes": [ 4781 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4697, 4781 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_253_output_0", "outputIndexes": [ 4782 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_50/Mul_1_output_0", "inputIndexes": [ 4782 ], "outputIndexes": [ 4783 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 13770741356, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4783 ], "outputIndexes": [ 7419 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7419 ], "outputIndexes": [ 7420 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/mlp/gate_proj/Linear", "inputIndexes": [ 7420 ], "outputIndexes": [ 7421 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 13770782316, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7421 ], "outputIndexes": [ 7422 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7422 ], "outputIndexes": [ 4784 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4784 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_50/Mul_output_0", "outputIndexes": [ 4785 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4783 ], "outputIndexes": [ 7423 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7423 ], "outputIndexes": [ 7424 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/mlp/up_proj/Linear", "inputIndexes": [ 7424 ], "outputIndexes": [ 7425 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 13844510338, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7425 ], "outputIndexes": [ 7426 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7426 ], "outputIndexes": [ 4786 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4785, 4786 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_50/Mul_output_0", "outputIndexes": [ 4787 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4787 ], "outputIndexes": [ 7427 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7427 ], "outputIndexes": [ 7428 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/mlp/down_proj/Linear", "inputIndexes": [ 7428 ], "outputIndexes": [ 7429 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 13918238360, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7429 ], "outputIndexes": [ 7430 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.50/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7430 ], "outputIndexes": [ 4788 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4782, 4788 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_254_output_0", "outputIndexes": [ 4789 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4789, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_459_output_0", "outputIndexes": [ 4790 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_51/Mul_1_output_0", "inputIndexes": [ 4790 ], "outputIndexes": [ 4791 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 13991966382, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4791 ], "outputIndexes": [ 7431 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7431 ], "outputIndexes": [ 7432 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/q_proj/Linear", "inputIndexes": [ 7432 ], "outputIndexes": [ 7433 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 13992007342, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7433 ], "outputIndexes": [ 7434 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7434 ], "outputIndexes": [ 4792 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4791 ], "main_type": "NONE", "name": "/Shape_714_output_0", "outputIndexes": [ 4793 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4793 ], "main_type": "NONE", "name": "Shape19949", "outputIndexes": [ 4794 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4793 ], "main_type": "NONE", "name": "Rank19951", "outputIndexes": [ 4795 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4795, 4795 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19953", "outputIndexes": [ 4796 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4796 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19954", "outputIndexes": [ 4797 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4796, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19956", "outputIndexes": [ 4798 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4798 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19957", "outputIndexes": [ 4799 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4794, 4797, 4799, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice19960", "outputIndexes": [ 4800 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4800 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze19961", "outputIndexes": [ 4801 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4801, 4801 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19963", "outputIndexes": [ 4802 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4793, 4802, 4796 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_768_output_0", "outputIndexes": [ 4803 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4803, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1632_output_0", "outputIndexes": [ 4804 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4796 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19971", "outputIndexes": [ 4805 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4798 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze19974", "outputIndexes": [ 4806 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4794, 4805, 4806, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice19977", "outputIndexes": [ 4807 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4807 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze19978", "outputIndexes": [ 4808 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 4808 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19979", "outputIndexes": [ 4809 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4809, 4808 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19980", "outputIndexes": [ 4810 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4793, 4810, 4796 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_769_output_0", "outputIndexes": [ 4811 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4811, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1633_output_0", "outputIndexes": [ 4812 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4804, 4812, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_663_output_0", "outputIndexes": [ 4813 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4792, 4813 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_460_output_0", "outputIndexes": [ 4814 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4814 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_51/Cast_output_0", "outputIndexes": [ 4815 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_51/Mul_1_output_0", "inputIndexes": [ 4815 ], "outputIndexes": [ 4816 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 14015600324, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4816, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_408_output_0", "outputIndexes": [ 4817 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4816 ], "main_type": "NONE", "name": "/Shape_716_output_0", "outputIndexes": [ 4818 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4818 ], "main_type": "NONE", "name": "Shape20099", "outputIndexes": [ 4819 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4818 ], "main_type": "NONE", "name": "Rank20101", "outputIndexes": [ 4820 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4820, 4820 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20103", "outputIndexes": [ 4821 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4821 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20104", "outputIndexes": [ 4822 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4821, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20106", "outputIndexes": [ 4823 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4823 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20107", "outputIndexes": [ 4824 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4819, 4822, 4824, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice20110", "outputIndexes": [ 4825 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4825 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze20111", "outputIndexes": [ 4826 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4826 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20112", "outputIndexes": [ 4827 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4827, 4826 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20113", "outputIndexes": [ 4828 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4818, 4828, 4821 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_770_output_0", "outputIndexes": [ 4829 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4829, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_153_output_0", "outputIndexes": [ 4830 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4830, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1639_output_0", "outputIndexes": [ 4831 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4816, 4831, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_205_output_0", "outputIndexes": [ 4832 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4832 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_102_output_0", "outputIndexes": [ 4833 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4830, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1638_output_0", "outputIndexes": [ 4834 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4816, 20, 4834, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_204_output_0", "outputIndexes": [ 4835 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4833, 4835 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_666_output_0", "outputIndexes": [ 4836 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4836, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_409_output_0", "outputIndexes": [ 4837 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4817, 4837 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_255_output_0", "outputIndexes": [ 4838 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4791 ], "outputIndexes": [ 7435 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7435 ], "outputIndexes": [ 7436 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/k_proj/Linear", "inputIndexes": [ 7436 ], "outputIndexes": [ 7437 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 14015601348, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7437 ], "outputIndexes": [ 7438 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7438 ], "outputIndexes": [ 4839 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4803, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1634_output_0", "outputIndexes": [ 4840 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4811, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1635_output_0", "outputIndexes": [ 4841 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4840, 4841, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_664_output_0", "outputIndexes": [ 4842 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4839, 4842 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_461_output_0", "outputIndexes": [ 4843 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4843 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_51/Cast_output_0", "outputIndexes": [ 4844 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_51/Mul_1_output_0", "inputIndexes": [ 4844 ], "outputIndexes": [ 4845 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 14018550490, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4845, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_410_output_0", "outputIndexes": [ 4846 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4845 ], "main_type": "NONE", "name": "/Shape_717_output_0", "outputIndexes": [ 4847 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4847 ], "main_type": "NONE", "name": "Shape19995", "outputIndexes": [ 4848 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4847 ], "main_type": "NONE", "name": "Rank19997", "outputIndexes": [ 4849 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4849, 4849 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp19999", "outputIndexes": [ 4850 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4850 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20000", "outputIndexes": [ 4851 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4850, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20002", "outputIndexes": [ 4852 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4852 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20003", "outputIndexes": [ 4853 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4848, 4851, 4853, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice20006", "outputIndexes": [ 4854 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4854 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze20007", "outputIndexes": [ 4855 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4855 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20008", "outputIndexes": [ 4856 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4856, 4855 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20009", "outputIndexes": [ 4857 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4847, 4857, 4850 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_771_output_0", "outputIndexes": [ 4858 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4858, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_154_output_0", "outputIndexes": [ 4859 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4859, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1641_output_0", "outputIndexes": [ 4860 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4845, 4860, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_207_output_0", "outputIndexes": [ 4861 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4861 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_103_output_0", "outputIndexes": [ 4862 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4859, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1640_output_0", "outputIndexes": [ 4863 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4845, 20, 4863, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_206_output_0", "outputIndexes": [ 4864 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4862, 4864 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_667_output_0", "outputIndexes": [ 4865 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4865, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_411_output_0", "outputIndexes": [ 4866 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4846, 4866 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_256_output_0", "outputIndexes": [ 4867 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4791 ], "outputIndexes": [ 7439 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7439 ], "outputIndexes": [ 7440 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/v_proj/Linear", "inputIndexes": [ 7440 ], "outputIndexes": [ 7441 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 14018551514, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7441 ], "outputIndexes": [ 7442 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7442 ], "outputIndexes": [ 4868 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4803, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1636_output_0", "outputIndexes": [ 4869 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4811, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1637_output_0", "outputIndexes": [ 4870 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4869, 4870, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_665_output_0", "outputIndexes": [ 4871 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4868, 4871 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_462_output_0", "outputIndexes": [ 4872 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4838, 4867, 4872, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_467_output_0", "outputIndexes": [ 4873 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4873 ], "outputIndexes": [ 7443 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7443 ], "outputIndexes": [ 7444 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/o_proj/Linear", "inputIndexes": [ 7444 ], "outputIndexes": [ 7445 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 14021500656, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7445 ], "outputIndexes": [ 7446 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7446 ], "outputIndexes": [ 4874 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4790, 4874 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_258_output_0", "outputIndexes": [ 4875 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_51/Mul_1_output_0", "inputIndexes": [ 4875 ], "outputIndexes": [ 4876 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 14045093638, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4876 ], "outputIndexes": [ 7447 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7447 ], "outputIndexes": [ 7448 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/mlp/gate_proj/Linear", "inputIndexes": [ 7448 ], "outputIndexes": [ 7449 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 14045134598, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7449 ], "outputIndexes": [ 7450 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7450 ], "outputIndexes": [ 4877 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4877 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_51/Mul_output_0", "outputIndexes": [ 4878 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4876 ], "outputIndexes": [ 7451 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7451 ], "outputIndexes": [ 7452 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/mlp/up_proj/Linear", "inputIndexes": [ 7452 ], "outputIndexes": [ 7453 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 14118862620, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7453 ], "outputIndexes": [ 7454 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7454 ], "outputIndexes": [ 4879 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4878, 4879 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_51/Mul_output_0", "outputIndexes": [ 4880 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4880 ], "outputIndexes": [ 7455 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7455 ], "outputIndexes": [ 7456 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/mlp/down_proj/Linear", "inputIndexes": [ 7456 ], "outputIndexes": [ 7457 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 14192590642, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7457 ], "outputIndexes": [ 7458 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.51/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7458 ], "outputIndexes": [ 4881 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4875, 4881 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_259_output_0", "outputIndexes": [ 4882 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4882, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_468_output_0", "outputIndexes": [ 4883 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_52/Mul_1_output_0", "inputIndexes": [ 4883 ], "outputIndexes": [ 4884 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 14266318664, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4884 ], "outputIndexes": [ 7459 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7459 ], "outputIndexes": [ 7460 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/q_proj/Linear", "inputIndexes": [ 7460 ], "outputIndexes": [ 7461 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 14266359624, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7461 ], "outputIndexes": [ 7462 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7462 ], "outputIndexes": [ 4885 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4884 ], "main_type": "NONE", "name": "/Shape_728_output_0", "outputIndexes": [ 4886 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4886 ], "main_type": "NONE", "name": "Shape20338", "outputIndexes": [ 4887 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4886 ], "main_type": "NONE", "name": "Rank20340", "outputIndexes": [ 4888 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4888, 4888 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20342", "outputIndexes": [ 4889 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4889 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20343", "outputIndexes": [ 4890 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4889, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20345", "outputIndexes": [ 4891 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4891 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20346", "outputIndexes": [ 4892 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4887, 4890, 4892, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice20349", "outputIndexes": [ 4893 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4893 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze20350", "outputIndexes": [ 4894 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4894, 4894 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20352", "outputIndexes": [ 4895 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4886, 4895, 4889 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_783_output_0", "outputIndexes": [ 4896 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4896, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1664_output_0", "outputIndexes": [ 4897 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4889 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20360", "outputIndexes": [ 4898 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4891 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20363", "outputIndexes": [ 4899 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4887, 4898, 4899, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice20366", "outputIndexes": [ 4900 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4900 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze20367", "outputIndexes": [ 4901 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 4901 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20368", "outputIndexes": [ 4902 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4902, 4901 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20369", "outputIndexes": [ 4903 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4886, 4903, 4889 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_784_output_0", "outputIndexes": [ 4904 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4904, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1665_output_0", "outputIndexes": [ 4905 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4897, 4905, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_676_output_0", "outputIndexes": [ 4906 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4885, 4906 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_469_output_0", "outputIndexes": [ 4907 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4907 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_52/Cast_output_0", "outputIndexes": [ 4908 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_52/Mul_1_output_0", "inputIndexes": [ 4908 ], "outputIndexes": [ 4909 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 14289952606, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4909, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_416_output_0", "outputIndexes": [ 4910 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4909 ], "main_type": "NONE", "name": "/Shape_730_output_0", "outputIndexes": [ 4911 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4911 ], "main_type": "NONE", "name": "Shape20488", "outputIndexes": [ 4912 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4911 ], "main_type": "NONE", "name": "Rank20490", "outputIndexes": [ 4913 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4913, 4913 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20492", "outputIndexes": [ 4914 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4914 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20493", "outputIndexes": [ 4915 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4914, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20495", "outputIndexes": [ 4916 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4916 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20496", "outputIndexes": [ 4917 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4912, 4915, 4917, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice20499", "outputIndexes": [ 4918 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4918 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze20500", "outputIndexes": [ 4919 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4919 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20501", "outputIndexes": [ 4920 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4920, 4919 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20502", "outputIndexes": [ 4921 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4911, 4921, 4914 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_785_output_0", "outputIndexes": [ 4922 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4922, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_156_output_0", "outputIndexes": [ 4923 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4923, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1671_output_0", "outputIndexes": [ 4924 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4909, 4924, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_209_output_0", "outputIndexes": [ 4925 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4925 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_104_output_0", "outputIndexes": [ 4926 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4923, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1670_output_0", "outputIndexes": [ 4927 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4909, 20, 4927, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_208_output_0", "outputIndexes": [ 4928 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4926, 4928 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_679_output_0", "outputIndexes": [ 4929 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4929, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_417_output_0", "outputIndexes": [ 4930 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4910, 4930 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_260_output_0", "outputIndexes": [ 4931 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4884 ], "outputIndexes": [ 7463 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7463 ], "outputIndexes": [ 7464 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/k_proj/Linear", "inputIndexes": [ 7464 ], "outputIndexes": [ 7465 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 14289953630, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7465 ], "outputIndexes": [ 7466 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7466 ], "outputIndexes": [ 4932 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4896, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1666_output_0", "outputIndexes": [ 4933 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4904, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1667_output_0", "outputIndexes": [ 4934 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4933, 4934, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_677_output_0", "outputIndexes": [ 4935 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4932, 4935 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_470_output_0", "outputIndexes": [ 4936 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4936 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_52/Cast_output_0", "outputIndexes": [ 4937 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_52/Mul_1_output_0", "inputIndexes": [ 4937 ], "outputIndexes": [ 4938 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 14292902772, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4938, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_418_output_0", "outputIndexes": [ 4939 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4938 ], "main_type": "NONE", "name": "/Shape_731_output_0", "outputIndexes": [ 4940 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4940 ], "main_type": "NONE", "name": "Shape20384", "outputIndexes": [ 4941 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4940 ], "main_type": "NONE", "name": "Rank20386", "outputIndexes": [ 4942 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4942, 4942 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20388", "outputIndexes": [ 4943 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4943 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20389", "outputIndexes": [ 4944 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4943, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20391", "outputIndexes": [ 4945 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4945 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20392", "outputIndexes": [ 4946 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4941, 4944, 4946, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice20395", "outputIndexes": [ 4947 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4947 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze20396", "outputIndexes": [ 4948 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 4948 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20397", "outputIndexes": [ 4949 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4949, 4948 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20398", "outputIndexes": [ 4950 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4940, 4950, 4943 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_786_output_0", "outputIndexes": [ 4951 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4951, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_157_output_0", "outputIndexes": [ 4952 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4952, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1673_output_0", "outputIndexes": [ 4953 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4938, 4953, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_211_output_0", "outputIndexes": [ 4954 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4954 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_105_output_0", "outputIndexes": [ 4955 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4952, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1672_output_0", "outputIndexes": [ 4956 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4938, 20, 4956, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_210_output_0", "outputIndexes": [ 4957 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4955, 4957 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_680_output_0", "outputIndexes": [ 4958 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4958, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_419_output_0", "outputIndexes": [ 4959 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4939, 4959 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_261_output_0", "outputIndexes": [ 4960 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4884 ], "outputIndexes": [ 7467 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7467 ], "outputIndexes": [ 7468 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/v_proj/Linear", "inputIndexes": [ 7468 ], "outputIndexes": [ 7469 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 14292903796, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7469 ], "outputIndexes": [ 7470 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7470 ], "outputIndexes": [ 4961 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4896, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1668_output_0", "outputIndexes": [ 4962 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4904, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1669_output_0", "outputIndexes": [ 4963 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4962, 4963, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_678_output_0", "outputIndexes": [ 4964 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4961, 4964 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_471_output_0", "outputIndexes": [ 4965 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4931, 4960, 4965, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_476_output_0", "outputIndexes": [ 4966 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4966 ], "outputIndexes": [ 7471 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7471 ], "outputIndexes": [ 7472 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/o_proj/Linear", "inputIndexes": [ 7472 ], "outputIndexes": [ 7473 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 14295852938, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7473 ], "outputIndexes": [ 7474 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7474 ], "outputIndexes": [ 4967 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4883, 4967 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_263_output_0", "outputIndexes": [ 4968 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_52/Mul_1_output_0", "inputIndexes": [ 4968 ], "outputIndexes": [ 4969 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 14319445920, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4969 ], "outputIndexes": [ 7475 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7475 ], "outputIndexes": [ 7476 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/mlp/gate_proj/Linear", "inputIndexes": [ 7476 ], "outputIndexes": [ 7477 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 14319486880, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7477 ], "outputIndexes": [ 7478 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7478 ], "outputIndexes": [ 4970 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4970 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_52/Mul_output_0", "outputIndexes": [ 4971 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4969 ], "outputIndexes": [ 7479 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7479 ], "outputIndexes": [ 7480 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/mlp/up_proj/Linear", "inputIndexes": [ 7480 ], "outputIndexes": [ 7481 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 14393214902, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7481 ], "outputIndexes": [ 7482 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7482 ], "outputIndexes": [ 4972 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4971, 4972 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_52/Mul_output_0", "outputIndexes": [ 4973 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4973 ], "outputIndexes": [ 7483 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7483 ], "outputIndexes": [ 7484 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/mlp/down_proj/Linear", "inputIndexes": [ 7484 ], "outputIndexes": [ 7485 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 14466942924, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7485 ], "outputIndexes": [ 7486 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.52/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7486 ], "outputIndexes": [ 4974 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4968, 4974 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_264_output_0", "outputIndexes": [ 4975 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4975, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_477_output_0", "outputIndexes": [ 4976 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_53/Mul_1_output_0", "inputIndexes": [ 4976 ], "outputIndexes": [ 4977 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 14540670946, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4977 ], "outputIndexes": [ 7487 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7487 ], "outputIndexes": [ 7488 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/q_proj/Linear", "inputIndexes": [ 7488 ], "outputIndexes": [ 7489 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 14540711906, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7489 ], "outputIndexes": [ 7490 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7490 ], "outputIndexes": [ 4978 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4977 ], "main_type": "NONE", "name": "/Shape_742_output_0", "outputIndexes": [ 4979 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4979 ], "main_type": "NONE", "name": "Shape20727", "outputIndexes": [ 4980 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 4979 ], "main_type": "NONE", "name": "Rank20729", "outputIndexes": [ 4981 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4981, 4981 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20731", "outputIndexes": [ 4982 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4982 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20732", "outputIndexes": [ 4983 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4982, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20734", "outputIndexes": [ 4984 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4984 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20735", "outputIndexes": [ 4985 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4980, 4983, 4985, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice20738", "outputIndexes": [ 4986 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4986 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze20739", "outputIndexes": [ 4987 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4987, 4987 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20741", "outputIndexes": [ 4988 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4979, 4988, 4982 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_798_output_0", "outputIndexes": [ 4989 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4989, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1696_output_0", "outputIndexes": [ 4990 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4982 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20749", "outputIndexes": [ 4991 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4984 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20752", "outputIndexes": [ 4992 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4980, 4991, 4992, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice20755", "outputIndexes": [ 4993 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4993 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze20756", "outputIndexes": [ 4994 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 4994 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20757", "outputIndexes": [ 4995 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4995, 4994 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20758", "outputIndexes": [ 4996 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4979, 4996, 4982 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_799_output_0", "outputIndexes": [ 4997 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4997, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1697_output_0", "outputIndexes": [ 4998 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4990, 4998, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_689_output_0", "outputIndexes": [ 4999 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4978, 4999 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_478_output_0", "outputIndexes": [ 5000 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5000 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_53/Cast_output_0", "outputIndexes": [ 5001 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_53/Mul_1_output_0", "inputIndexes": [ 5001 ], "outputIndexes": [ 5002 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 14564304888, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5002, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_424_output_0", "outputIndexes": [ 5003 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5002 ], "main_type": "NONE", "name": "/Shape_744_output_0", "outputIndexes": [ 5004 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5004 ], "main_type": "NONE", "name": "Shape20877", "outputIndexes": [ 5005 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5004 ], "main_type": "NONE", "name": "Rank20879", "outputIndexes": [ 5006 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5006, 5006 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20881", "outputIndexes": [ 5007 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5007 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20882", "outputIndexes": [ 5008 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5007, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20884", "outputIndexes": [ 5009 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5009 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20885", "outputIndexes": [ 5010 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5005, 5008, 5010, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice20888", "outputIndexes": [ 5011 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5011 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze20889", "outputIndexes": [ 5012 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5012 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20890", "outputIndexes": [ 5013 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5013, 5012 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20891", "outputIndexes": [ 5014 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5004, 5014, 5007 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_800_output_0", "outputIndexes": [ 5015 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5015, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_159_output_0", "outputIndexes": [ 5016 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5016, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1703_output_0", "outputIndexes": [ 5017 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5002, 5017, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_213_output_0", "outputIndexes": [ 5018 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5018 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_106_output_0", "outputIndexes": [ 5019 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5016, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1702_output_0", "outputIndexes": [ 5020 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5002, 20, 5020, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_212_output_0", "outputIndexes": [ 5021 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5019, 5021 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_692_output_0", "outputIndexes": [ 5022 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5022, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_425_output_0", "outputIndexes": [ 5023 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5003, 5023 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_265_output_0", "outputIndexes": [ 5024 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4977 ], "outputIndexes": [ 7491 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7491 ], "outputIndexes": [ 7492 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/k_proj/Linear", "inputIndexes": [ 7492 ], "outputIndexes": [ 7493 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 14564305912, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7493 ], "outputIndexes": [ 7494 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7494 ], "outputIndexes": [ 5025 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4989, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1698_output_0", "outputIndexes": [ 5026 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4997, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1699_output_0", "outputIndexes": [ 5027 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5026, 5027, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_690_output_0", "outputIndexes": [ 5028 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5025, 5028 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_479_output_0", "outputIndexes": [ 5029 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5029 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_53/Cast_output_0", "outputIndexes": [ 5030 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_53/Mul_1_output_0", "inputIndexes": [ 5030 ], "outputIndexes": [ 5031 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 14567255054, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5031, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_426_output_0", "outputIndexes": [ 5032 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5031 ], "main_type": "NONE", "name": "/Shape_745_output_0", "outputIndexes": [ 5033 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5033 ], "main_type": "NONE", "name": "Shape20773", "outputIndexes": [ 5034 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5033 ], "main_type": "NONE", "name": "Rank20775", "outputIndexes": [ 5035 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5035, 5035 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20777", "outputIndexes": [ 5036 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5036 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20778", "outputIndexes": [ 5037 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5036, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20780", "outputIndexes": [ 5038 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5038 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze20781", "outputIndexes": [ 5039 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5034, 5037, 5039, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice20784", "outputIndexes": [ 5040 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5040 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze20785", "outputIndexes": [ 5041 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5041 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20786", "outputIndexes": [ 5042 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5042, 5041 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp20787", "outputIndexes": [ 5043 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5033, 5043, 5036 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_801_output_0", "outputIndexes": [ 5044 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5044, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_160_output_0", "outputIndexes": [ 5045 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5045, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1705_output_0", "outputIndexes": [ 5046 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5031, 5046, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_215_output_0", "outputIndexes": [ 5047 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5047 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_107_output_0", "outputIndexes": [ 5048 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5045, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1704_output_0", "outputIndexes": [ 5049 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5031, 20, 5049, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_214_output_0", "outputIndexes": [ 5050 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5048, 5050 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_693_output_0", "outputIndexes": [ 5051 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5051, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_427_output_0", "outputIndexes": [ 5052 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5032, 5052 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_266_output_0", "outputIndexes": [ 5053 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 4977 ], "outputIndexes": [ 7495 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7495 ], "outputIndexes": [ 7496 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/v_proj/Linear", "inputIndexes": [ 7496 ], "outputIndexes": [ 7497 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 14567256078, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7497 ], "outputIndexes": [ 7498 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7498 ], "outputIndexes": [ 5054 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4989, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1700_output_0", "outputIndexes": [ 5055 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4997, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1701_output_0", "outputIndexes": [ 5056 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5055, 5056, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_691_output_0", "outputIndexes": [ 5057 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5054, 5057 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_480_output_0", "outputIndexes": [ 5058 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5024, 5053, 5058, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_485_output_0", "outputIndexes": [ 5059 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5059 ], "outputIndexes": [ 7499 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7499 ], "outputIndexes": [ 7500 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/o_proj/Linear", "inputIndexes": [ 7500 ], "outputIndexes": [ 7501 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 14570205220, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7501 ], "outputIndexes": [ 7502 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7502 ], "outputIndexes": [ 5060 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 4976, 5060 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_268_output_0", "outputIndexes": [ 5061 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_53/Mul_1_output_0", "inputIndexes": [ 5061 ], "outputIndexes": [ 5062 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 14593798202, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5062 ], "outputIndexes": [ 7503 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7503 ], "outputIndexes": [ 7504 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/mlp/gate_proj/Linear", "inputIndexes": [ 7504 ], "outputIndexes": [ 7505 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 14593839162, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7505 ], "outputIndexes": [ 7506 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7506 ], "outputIndexes": [ 5063 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5063 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_53/Mul_output_0", "outputIndexes": [ 5064 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5062 ], "outputIndexes": [ 7507 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7507 ], "outputIndexes": [ 7508 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/mlp/up_proj/Linear", "inputIndexes": [ 7508 ], "outputIndexes": [ 7509 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 14667567184, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7509 ], "outputIndexes": [ 7510 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7510 ], "outputIndexes": [ 5065 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5064, 5065 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_53/Mul_output_0", "outputIndexes": [ 5066 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5066 ], "outputIndexes": [ 7511 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7511 ], "outputIndexes": [ 7512 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/mlp/down_proj/Linear", "inputIndexes": [ 7512 ], "outputIndexes": [ 7513 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 14741295206, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7513 ], "outputIndexes": [ 7514 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.53/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7514 ], "outputIndexes": [ 5067 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5061, 5067 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_269_output_0", "outputIndexes": [ 5068 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5068, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_486_output_0", "outputIndexes": [ 5069 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_54/Mul_1_output_0", "inputIndexes": [ 5069 ], "outputIndexes": [ 5070 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 14815023228, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5070 ], "outputIndexes": [ 7515 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7515 ], "outputIndexes": [ 7516 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/q_proj/Linear", "inputIndexes": [ 7516 ], "outputIndexes": [ 7517 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 14815064188, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7517 ], "outputIndexes": [ 7518 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7518 ], "outputIndexes": [ 5071 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5070 ], "main_type": "NONE", "name": "/Shape_756_output_0", "outputIndexes": [ 5072 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5072 ], "main_type": "NONE", "name": "Shape21116", "outputIndexes": [ 5073 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5072 ], "main_type": "NONE", "name": "Rank21118", "outputIndexes": [ 5074 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5074, 5074 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21120", "outputIndexes": [ 5075 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5075 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21121", "outputIndexes": [ 5076 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5075, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21123", "outputIndexes": [ 5077 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5077 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21124", "outputIndexes": [ 5078 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5073, 5076, 5078, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice21127", "outputIndexes": [ 5079 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5079 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze21128", "outputIndexes": [ 5080 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5080, 5080 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21130", "outputIndexes": [ 5081 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5072, 5081, 5075 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_813_output_0", "outputIndexes": [ 5082 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5082, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1728_output_0", "outputIndexes": [ 5083 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5075 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21138", "outputIndexes": [ 5084 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5077 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21141", "outputIndexes": [ 5085 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5073, 5084, 5085, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice21144", "outputIndexes": [ 5086 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5086 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze21145", "outputIndexes": [ 5087 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 5087 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21146", "outputIndexes": [ 5088 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5088, 5087 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21147", "outputIndexes": [ 5089 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5072, 5089, 5075 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_814_output_0", "outputIndexes": [ 5090 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5090, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1729_output_0", "outputIndexes": [ 5091 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5083, 5091, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_702_output_0", "outputIndexes": [ 5092 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5071, 5092 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_487_output_0", "outputIndexes": [ 5093 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5093 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_54/Cast_output_0", "outputIndexes": [ 5094 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_54/Mul_1_output_0", "inputIndexes": [ 5094 ], "outputIndexes": [ 5095 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 14838657170, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5095, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_432_output_0", "outputIndexes": [ 5096 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5095 ], "main_type": "NONE", "name": "/Shape_758_output_0", "outputIndexes": [ 5097 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5097 ], "main_type": "NONE", "name": "Shape21266", "outputIndexes": [ 5098 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5097 ], "main_type": "NONE", "name": "Rank21268", "outputIndexes": [ 5099 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5099, 5099 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21270", "outputIndexes": [ 5100 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5100 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21271", "outputIndexes": [ 5101 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5100, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21273", "outputIndexes": [ 5102 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5102 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21274", "outputIndexes": [ 5103 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5098, 5101, 5103, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice21277", "outputIndexes": [ 5104 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5104 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze21278", "outputIndexes": [ 5105 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5105 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21279", "outputIndexes": [ 5106 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5106, 5105 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21280", "outputIndexes": [ 5107 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5097, 5107, 5100 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_815_output_0", "outputIndexes": [ 5108 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5108, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_162_output_0", "outputIndexes": [ 5109 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5109, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1735_output_0", "outputIndexes": [ 5110 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5095, 5110, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_217_output_0", "outputIndexes": [ 5111 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5111 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_108_output_0", "outputIndexes": [ 5112 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5109, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1734_output_0", "outputIndexes": [ 5113 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5095, 20, 5113, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_216_output_0", "outputIndexes": [ 5114 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5112, 5114 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_705_output_0", "outputIndexes": [ 5115 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5115, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_433_output_0", "outputIndexes": [ 5116 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5096, 5116 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_270_output_0", "outputIndexes": [ 5117 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5070 ], "outputIndexes": [ 7519 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7519 ], "outputIndexes": [ 7520 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/k_proj/Linear", "inputIndexes": [ 7520 ], "outputIndexes": [ 7521 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 14838658194, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7521 ], "outputIndexes": [ 7522 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7522 ], "outputIndexes": [ 5118 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5082, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1730_output_0", "outputIndexes": [ 5119 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5090, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1731_output_0", "outputIndexes": [ 5120 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5119, 5120, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_703_output_0", "outputIndexes": [ 5121 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5118, 5121 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_488_output_0", "outputIndexes": [ 5122 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5122 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_54/Cast_output_0", "outputIndexes": [ 5123 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_54/Mul_1_output_0", "inputIndexes": [ 5123 ], "outputIndexes": [ 5124 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 14841607336, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5124, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_434_output_0", "outputIndexes": [ 5125 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5124 ], "main_type": "NONE", "name": "/Shape_759_output_0", "outputIndexes": [ 5126 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5126 ], "main_type": "NONE", "name": "Shape21162", "outputIndexes": [ 5127 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5126 ], "main_type": "NONE", "name": "Rank21164", "outputIndexes": [ 5128 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5128, 5128 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21166", "outputIndexes": [ 5129 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5129 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21167", "outputIndexes": [ 5130 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5129, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21169", "outputIndexes": [ 5131 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5131 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21170", "outputIndexes": [ 5132 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5127, 5130, 5132, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice21173", "outputIndexes": [ 5133 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5133 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze21174", "outputIndexes": [ 5134 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5134 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21175", "outputIndexes": [ 5135 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5135, 5134 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21176", "outputIndexes": [ 5136 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5126, 5136, 5129 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_816_output_0", "outputIndexes": [ 5137 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5137, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_163_output_0", "outputIndexes": [ 5138 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5138, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1737_output_0", "outputIndexes": [ 5139 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5124, 5139, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_219_output_0", "outputIndexes": [ 5140 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5140 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_109_output_0", "outputIndexes": [ 5141 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5138, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1736_output_0", "outputIndexes": [ 5142 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5124, 20, 5142, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_218_output_0", "outputIndexes": [ 5143 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5141, 5143 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_706_output_0", "outputIndexes": [ 5144 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5144, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_435_output_0", "outputIndexes": [ 5145 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5125, 5145 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_271_output_0", "outputIndexes": [ 5146 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5070 ], "outputIndexes": [ 7523 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7523 ], "outputIndexes": [ 7524 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/v_proj/Linear", "inputIndexes": [ 7524 ], "outputIndexes": [ 7525 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 14841608360, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7525 ], "outputIndexes": [ 7526 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7526 ], "outputIndexes": [ 5147 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5082, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1732_output_0", "outputIndexes": [ 5148 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5090, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1733_output_0", "outputIndexes": [ 5149 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5148, 5149, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_704_output_0", "outputIndexes": [ 5150 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5147, 5150 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_489_output_0", "outputIndexes": [ 5151 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5117, 5146, 5151, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_494_output_0", "outputIndexes": [ 5152 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5152 ], "outputIndexes": [ 7527 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7527 ], "outputIndexes": [ 7528 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/o_proj/Linear", "inputIndexes": [ 7528 ], "outputIndexes": [ 7529 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 14844557502, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7529 ], "outputIndexes": [ 7530 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7530 ], "outputIndexes": [ 5153 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5069, 5153 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_273_output_0", "outputIndexes": [ 5154 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_54/Mul_1_output_0", "inputIndexes": [ 5154 ], "outputIndexes": [ 5155 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 14868150484, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5155 ], "outputIndexes": [ 7531 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7531 ], "outputIndexes": [ 7532 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/mlp/gate_proj/Linear", "inputIndexes": [ 7532 ], "outputIndexes": [ 7533 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 14868191444, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7533 ], "outputIndexes": [ 7534 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7534 ], "outputIndexes": [ 5156 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5156 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_54/Mul_output_0", "outputIndexes": [ 5157 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5155 ], "outputIndexes": [ 7535 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7535 ], "outputIndexes": [ 7536 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/mlp/up_proj/Linear", "inputIndexes": [ 7536 ], "outputIndexes": [ 7537 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 14941919466, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7537 ], "outputIndexes": [ 7538 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7538 ], "outputIndexes": [ 5158 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5157, 5158 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_54/Mul_output_0", "outputIndexes": [ 5159 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5159 ], "outputIndexes": [ 7539 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7539 ], "outputIndexes": [ 7540 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/mlp/down_proj/Linear", "inputIndexes": [ 7540 ], "outputIndexes": [ 7541 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 15015647488, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7541 ], "outputIndexes": [ 7542 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.54/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7542 ], "outputIndexes": [ 5160 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5154, 5160 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_274_output_0", "outputIndexes": [ 5161 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5161, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_495_output_0", "outputIndexes": [ 5162 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_55/Mul_1_output_0", "inputIndexes": [ 5162 ], "outputIndexes": [ 5163 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 15089375510, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5163 ], "outputIndexes": [ 7543 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7543 ], "outputIndexes": [ 7544 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/q_proj/Linear", "inputIndexes": [ 7544 ], "outputIndexes": [ 7545 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 15089416470, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7545 ], "outputIndexes": [ 7546 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7546 ], "outputIndexes": [ 5164 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5163 ], "main_type": "NONE", "name": "/Shape_770_output_0", "outputIndexes": [ 5165 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5165 ], "main_type": "NONE", "name": "Shape21505", "outputIndexes": [ 5166 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5165 ], "main_type": "NONE", "name": "Rank21507", "outputIndexes": [ 5167 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5167, 5167 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21509", "outputIndexes": [ 5168 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5168 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21510", "outputIndexes": [ 5169 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5168, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21512", "outputIndexes": [ 5170 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5170 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21513", "outputIndexes": [ 5171 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5166, 5169, 5171, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice21516", "outputIndexes": [ 5172 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5172 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze21517", "outputIndexes": [ 5173 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5173, 5173 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21519", "outputIndexes": [ 5174 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5165, 5174, 5168 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_828_output_0", "outputIndexes": [ 5175 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5175, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1760_output_0", "outputIndexes": [ 5176 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5168 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21527", "outputIndexes": [ 5177 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5170 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21530", "outputIndexes": [ 5178 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5166, 5177, 5178, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice21533", "outputIndexes": [ 5179 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5179 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze21534", "outputIndexes": [ 5180 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 5180 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21535", "outputIndexes": [ 5181 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5181, 5180 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21536", "outputIndexes": [ 5182 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5165, 5182, 5168 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_829_output_0", "outputIndexes": [ 5183 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5183, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1761_output_0", "outputIndexes": [ 5184 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5176, 5184, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_715_output_0", "outputIndexes": [ 5185 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5164, 5185 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_496_output_0", "outputIndexes": [ 5186 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5186 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_55/Cast_output_0", "outputIndexes": [ 5187 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_55/Mul_1_output_0", "inputIndexes": [ 5187 ], "outputIndexes": [ 5188 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 15113009452, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5188, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_440_output_0", "outputIndexes": [ 5189 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5188 ], "main_type": "NONE", "name": "/Shape_772_output_0", "outputIndexes": [ 5190 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5190 ], "main_type": "NONE", "name": "Shape21655", "outputIndexes": [ 5191 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5190 ], "main_type": "NONE", "name": "Rank21657", "outputIndexes": [ 5192 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5192, 5192 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21659", "outputIndexes": [ 5193 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5193 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21660", "outputIndexes": [ 5194 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5193, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21662", "outputIndexes": [ 5195 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5195 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21663", "outputIndexes": [ 5196 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5191, 5194, 5196, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice21666", "outputIndexes": [ 5197 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5197 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze21667", "outputIndexes": [ 5198 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5198 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21668", "outputIndexes": [ 5199 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5199, 5198 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21669", "outputIndexes": [ 5200 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5190, 5200, 5193 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_830_output_0", "outputIndexes": [ 5201 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5201, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_165_output_0", "outputIndexes": [ 5202 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5202, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1767_output_0", "outputIndexes": [ 5203 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5188, 5203, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_221_output_0", "outputIndexes": [ 5204 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5204 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_110_output_0", "outputIndexes": [ 5205 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5202, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1766_output_0", "outputIndexes": [ 5206 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5188, 20, 5206, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_220_output_0", "outputIndexes": [ 5207 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5205, 5207 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_718_output_0", "outputIndexes": [ 5208 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5208, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_441_output_0", "outputIndexes": [ 5209 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5189, 5209 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_275_output_0", "outputIndexes": [ 5210 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5163 ], "outputIndexes": [ 7547 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7547 ], "outputIndexes": [ 7548 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/k_proj/Linear", "inputIndexes": [ 7548 ], "outputIndexes": [ 7549 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 15113010476, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7549 ], "outputIndexes": [ 7550 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7550 ], "outputIndexes": [ 5211 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5175, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1762_output_0", "outputIndexes": [ 5212 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5183, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1763_output_0", "outputIndexes": [ 5213 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5212, 5213, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_716_output_0", "outputIndexes": [ 5214 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5211, 5214 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_497_output_0", "outputIndexes": [ 5215 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5215 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_55/Cast_output_0", "outputIndexes": [ 5216 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_55/Mul_1_output_0", "inputIndexes": [ 5216 ], "outputIndexes": [ 5217 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 15115959618, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5217, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_442_output_0", "outputIndexes": [ 5218 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5217 ], "main_type": "NONE", "name": "/Shape_773_output_0", "outputIndexes": [ 5219 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5219 ], "main_type": "NONE", "name": "Shape21551", "outputIndexes": [ 5220 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5219 ], "main_type": "NONE", "name": "Rank21553", "outputIndexes": [ 5221 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5221, 5221 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21555", "outputIndexes": [ 5222 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5222 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21556", "outputIndexes": [ 5223 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5222, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21558", "outputIndexes": [ 5224 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5224 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21559", "outputIndexes": [ 5225 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5220, 5223, 5225, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice21562", "outputIndexes": [ 5226 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5226 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze21563", "outputIndexes": [ 5227 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5227 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21564", "outputIndexes": [ 5228 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5228, 5227 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21565", "outputIndexes": [ 5229 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5219, 5229, 5222 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_831_output_0", "outputIndexes": [ 5230 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5230, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_166_output_0", "outputIndexes": [ 5231 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5231, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1769_output_0", "outputIndexes": [ 5232 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5217, 5232, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_223_output_0", "outputIndexes": [ 5233 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5233 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_111_output_0", "outputIndexes": [ 5234 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5231, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1768_output_0", "outputIndexes": [ 5235 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5217, 20, 5235, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_222_output_0", "outputIndexes": [ 5236 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5234, 5236 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_719_output_0", "outputIndexes": [ 5237 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5237, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_443_output_0", "outputIndexes": [ 5238 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5218, 5238 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_276_output_0", "outputIndexes": [ 5239 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5163 ], "outputIndexes": [ 7551 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7551 ], "outputIndexes": [ 7552 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/v_proj/Linear", "inputIndexes": [ 7552 ], "outputIndexes": [ 7553 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 15115960642, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7553 ], "outputIndexes": [ 7554 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7554 ], "outputIndexes": [ 5240 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5175, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1764_output_0", "outputIndexes": [ 5241 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5183, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1765_output_0", "outputIndexes": [ 5242 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5241, 5242, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_717_output_0", "outputIndexes": [ 5243 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5240, 5243 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_498_output_0", "outputIndexes": [ 5244 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5210, 5239, 5244, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_503_output_0", "outputIndexes": [ 5245 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5245 ], "outputIndexes": [ 7555 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7555 ], "outputIndexes": [ 7556 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/o_proj/Linear", "inputIndexes": [ 7556 ], "outputIndexes": [ 7557 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 15118909784, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7557 ], "outputIndexes": [ 7558 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7558 ], "outputIndexes": [ 5246 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5162, 5246 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_278_output_0", "outputIndexes": [ 5247 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_55/Mul_1_output_0", "inputIndexes": [ 5247 ], "outputIndexes": [ 5248 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 15142502766, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5248 ], "outputIndexes": [ 7559 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7559 ], "outputIndexes": [ 7560 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/mlp/gate_proj/Linear", "inputIndexes": [ 7560 ], "outputIndexes": [ 7561 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 15142543726, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7561 ], "outputIndexes": [ 7562 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7562 ], "outputIndexes": [ 5249 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5249 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_55/Mul_output_0", "outputIndexes": [ 5250 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5248 ], "outputIndexes": [ 7563 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7563 ], "outputIndexes": [ 7564 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/mlp/up_proj/Linear", "inputIndexes": [ 7564 ], "outputIndexes": [ 7565 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 15216271748, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7565 ], "outputIndexes": [ 7566 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7566 ], "outputIndexes": [ 5251 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5250, 5251 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_55/Mul_output_0", "outputIndexes": [ 5252 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5252 ], "outputIndexes": [ 7567 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7567 ], "outputIndexes": [ 7568 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/mlp/down_proj/Linear", "inputIndexes": [ 7568 ], "outputIndexes": [ 7569 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 15289999770, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7569 ], "outputIndexes": [ 7570 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.55/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7570 ], "outputIndexes": [ 5253 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5247, 5253 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_279_output_0", "outputIndexes": [ 5254 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5254, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_504_output_0", "outputIndexes": [ 5255 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_56/Mul_1_output_0", "inputIndexes": [ 5255 ], "outputIndexes": [ 5256 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 15363727792, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5256 ], "outputIndexes": [ 7571 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7571 ], "outputIndexes": [ 7572 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/q_proj/Linear", "inputIndexes": [ 7572 ], "outputIndexes": [ 7573 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 15363768752, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7573 ], "outputIndexes": [ 7574 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7574 ], "outputIndexes": [ 5257 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5256 ], "main_type": "NONE", "name": "/Shape_784_output_0", "outputIndexes": [ 5258 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5258 ], "main_type": "NONE", "name": "Shape21894", "outputIndexes": [ 5259 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5258 ], "main_type": "NONE", "name": "Rank21896", "outputIndexes": [ 5260 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5260, 5260 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21898", "outputIndexes": [ 5261 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5261 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21899", "outputIndexes": [ 5262 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5261, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21901", "outputIndexes": [ 5263 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5263 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21902", "outputIndexes": [ 5264 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5259, 5262, 5264, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice21905", "outputIndexes": [ 5265 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5265 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze21906", "outputIndexes": [ 5266 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5266, 5266 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21908", "outputIndexes": [ 5267 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5258, 5267, 5261 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_843_output_0", "outputIndexes": [ 5268 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5268, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1792_output_0", "outputIndexes": [ 5269 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5261 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21916", "outputIndexes": [ 5270 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5263 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21919", "outputIndexes": [ 5271 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5259, 5270, 5271, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice21922", "outputIndexes": [ 5272 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5272 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze21923", "outputIndexes": [ 5273 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 5273 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21924", "outputIndexes": [ 5274 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5274, 5273 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21925", "outputIndexes": [ 5275 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5258, 5275, 5261 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_844_output_0", "outputIndexes": [ 5276 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5276, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1793_output_0", "outputIndexes": [ 5277 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5269, 5277, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_728_output_0", "outputIndexes": [ 5278 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5257, 5278 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_505_output_0", "outputIndexes": [ 5279 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5279 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_56/Cast_output_0", "outputIndexes": [ 5280 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_56/Mul_1_output_0", "inputIndexes": [ 5280 ], "outputIndexes": [ 5281 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 15387361734, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5281, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_448_output_0", "outputIndexes": [ 5282 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5281 ], "main_type": "NONE", "name": "/Shape_786_output_0", "outputIndexes": [ 5283 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5283 ], "main_type": "NONE", "name": "Shape22044", "outputIndexes": [ 5284 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5283 ], "main_type": "NONE", "name": "Rank22046", "outputIndexes": [ 5285 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5285, 5285 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22048", "outputIndexes": [ 5286 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5286 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22049", "outputIndexes": [ 5287 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5286, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22051", "outputIndexes": [ 5288 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5288 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22052", "outputIndexes": [ 5289 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5284, 5287, 5289, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice22055", "outputIndexes": [ 5290 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5290 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze22056", "outputIndexes": [ 5291 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5291 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22057", "outputIndexes": [ 5292 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5292, 5291 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22058", "outputIndexes": [ 5293 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5283, 5293, 5286 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_845_output_0", "outputIndexes": [ 5294 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5294, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_168_output_0", "outputIndexes": [ 5295 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5295, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1799_output_0", "outputIndexes": [ 5296 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5281, 5296, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_225_output_0", "outputIndexes": [ 5297 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5297 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_112_output_0", "outputIndexes": [ 5298 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5295, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1798_output_0", "outputIndexes": [ 5299 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5281, 20, 5299, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_224_output_0", "outputIndexes": [ 5300 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5298, 5300 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_731_output_0", "outputIndexes": [ 5301 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5301, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_449_output_0", "outputIndexes": [ 5302 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5282, 5302 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_280_output_0", "outputIndexes": [ 5303 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5256 ], "outputIndexes": [ 7575 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7575 ], "outputIndexes": [ 7576 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/k_proj/Linear", "inputIndexes": [ 7576 ], "outputIndexes": [ 7577 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 15387362758, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7577 ], "outputIndexes": [ 7578 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7578 ], "outputIndexes": [ 5304 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5268, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1794_output_0", "outputIndexes": [ 5305 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5276, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1795_output_0", "outputIndexes": [ 5306 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5305, 5306, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_729_output_0", "outputIndexes": [ 5307 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5304, 5307 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_506_output_0", "outputIndexes": [ 5308 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5308 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_56/Cast_output_0", "outputIndexes": [ 5309 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_56/Mul_1_output_0", "inputIndexes": [ 5309 ], "outputIndexes": [ 5310 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 15390311900, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5310, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_450_output_0", "outputIndexes": [ 5311 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5310 ], "main_type": "NONE", "name": "/Shape_787_output_0", "outputIndexes": [ 5312 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5312 ], "main_type": "NONE", "name": "Shape21940", "outputIndexes": [ 5313 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5312 ], "main_type": "NONE", "name": "Rank21942", "outputIndexes": [ 5314 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5314, 5314 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21944", "outputIndexes": [ 5315 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5315 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21945", "outputIndexes": [ 5316 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5315, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21947", "outputIndexes": [ 5317 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5317 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze21948", "outputIndexes": [ 5318 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5313, 5316, 5318, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice21951", "outputIndexes": [ 5319 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5319 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze21952", "outputIndexes": [ 5320 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5320 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21953", "outputIndexes": [ 5321 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5321, 5320 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp21954", "outputIndexes": [ 5322 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5312, 5322, 5315 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_846_output_0", "outputIndexes": [ 5323 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5323, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_169_output_0", "outputIndexes": [ 5324 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5324, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1801_output_0", "outputIndexes": [ 5325 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5310, 5325, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_227_output_0", "outputIndexes": [ 5326 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5326 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_113_output_0", "outputIndexes": [ 5327 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5324, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1800_output_0", "outputIndexes": [ 5328 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5310, 20, 5328, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_226_output_0", "outputIndexes": [ 5329 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5327, 5329 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_732_output_0", "outputIndexes": [ 5330 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5330, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_451_output_0", "outputIndexes": [ 5331 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5311, 5331 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_281_output_0", "outputIndexes": [ 5332 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5256 ], "outputIndexes": [ 7579 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7579 ], "outputIndexes": [ 7580 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/v_proj/Linear", "inputIndexes": [ 7580 ], "outputIndexes": [ 7581 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 15390312924, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7581 ], "outputIndexes": [ 7582 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7582 ], "outputIndexes": [ 5333 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5268, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1796_output_0", "outputIndexes": [ 5334 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5276, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1797_output_0", "outputIndexes": [ 5335 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5334, 5335, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_730_output_0", "outputIndexes": [ 5336 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5333, 5336 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_507_output_0", "outputIndexes": [ 5337 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5303, 5332, 5337, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_512_output_0", "outputIndexes": [ 5338 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5338 ], "outputIndexes": [ 7583 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7583 ], "outputIndexes": [ 7584 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/o_proj/Linear", "inputIndexes": [ 7584 ], "outputIndexes": [ 7585 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 15393262066, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7585 ], "outputIndexes": [ 7586 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7586 ], "outputIndexes": [ 5339 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5255, 5339 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_283_output_0", "outputIndexes": [ 5340 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_56/Mul_1_output_0", "inputIndexes": [ 5340 ], "outputIndexes": [ 5341 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 15416855048, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5341 ], "outputIndexes": [ 7587 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7587 ], "outputIndexes": [ 7588 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/mlp/gate_proj/Linear", "inputIndexes": [ 7588 ], "outputIndexes": [ 7589 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 15416896008, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7589 ], "outputIndexes": [ 7590 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7590 ], "outputIndexes": [ 5342 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5342 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_56/Mul_output_0", "outputIndexes": [ 5343 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5341 ], "outputIndexes": [ 7591 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7591 ], "outputIndexes": [ 7592 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/mlp/up_proj/Linear", "inputIndexes": [ 7592 ], "outputIndexes": [ 7593 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 15490624030, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7593 ], "outputIndexes": [ 7594 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7594 ], "outputIndexes": [ 5344 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5343, 5344 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_56/Mul_output_0", "outputIndexes": [ 5345 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5345 ], "outputIndexes": [ 7595 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7595 ], "outputIndexes": [ 7596 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/mlp/down_proj/Linear", "inputIndexes": [ 7596 ], "outputIndexes": [ 7597 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 15564352052, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7597 ], "outputIndexes": [ 7598 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.56/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7598 ], "outputIndexes": [ 5346 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5340, 5346 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_284_output_0", "outputIndexes": [ 5347 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5347, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_513_output_0", "outputIndexes": [ 5348 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_57/Mul_1_output_0", "inputIndexes": [ 5348 ], "outputIndexes": [ 5349 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 15638080074, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5349 ], "outputIndexes": [ 7599 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7599 ], "outputIndexes": [ 7600 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/q_proj/Linear", "inputIndexes": [ 7600 ], "outputIndexes": [ 7601 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 15638121034, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7601 ], "outputIndexes": [ 7602 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7602 ], "outputIndexes": [ 5350 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5349 ], "main_type": "NONE", "name": "/Shape_798_output_0", "outputIndexes": [ 5351 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5351 ], "main_type": "NONE", "name": "Shape22283", "outputIndexes": [ 5352 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5351 ], "main_type": "NONE", "name": "Rank22285", "outputIndexes": [ 5353 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5353, 5353 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22287", "outputIndexes": [ 5354 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5354 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22288", "outputIndexes": [ 5355 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5354, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22290", "outputIndexes": [ 5356 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5356 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22291", "outputIndexes": [ 5357 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5352, 5355, 5357, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice22294", "outputIndexes": [ 5358 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5358 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze22295", "outputIndexes": [ 5359 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5359, 5359 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22297", "outputIndexes": [ 5360 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5351, 5360, 5354 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_858_output_0", "outputIndexes": [ 5361 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5361, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1824_output_0", "outputIndexes": [ 5362 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5354 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22305", "outputIndexes": [ 5363 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5356 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22308", "outputIndexes": [ 5364 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5352, 5363, 5364, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice22311", "outputIndexes": [ 5365 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5365 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze22312", "outputIndexes": [ 5366 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 5366 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22313", "outputIndexes": [ 5367 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5367, 5366 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22314", "outputIndexes": [ 5368 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5351, 5368, 5354 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_859_output_0", "outputIndexes": [ 5369 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5369, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1825_output_0", "outputIndexes": [ 5370 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5362, 5370, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_741_output_0", "outputIndexes": [ 5371 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5350, 5371 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_514_output_0", "outputIndexes": [ 5372 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5372 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_57/Cast_output_0", "outputIndexes": [ 5373 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_57/Mul_1_output_0", "inputIndexes": [ 5373 ], "outputIndexes": [ 5374 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 15661714016, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5374, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_456_output_0", "outputIndexes": [ 5375 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5374 ], "main_type": "NONE", "name": "/Shape_800_output_0", "outputIndexes": [ 5376 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5376 ], "main_type": "NONE", "name": "Shape22433", "outputIndexes": [ 5377 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5376 ], "main_type": "NONE", "name": "Rank22435", "outputIndexes": [ 5378 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5378, 5378 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22437", "outputIndexes": [ 5379 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5379 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22438", "outputIndexes": [ 5380 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5379, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22440", "outputIndexes": [ 5381 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5381 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22441", "outputIndexes": [ 5382 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5377, 5380, 5382, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice22444", "outputIndexes": [ 5383 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5383 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze22445", "outputIndexes": [ 5384 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5384 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22446", "outputIndexes": [ 5385 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5385, 5384 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22447", "outputIndexes": [ 5386 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5376, 5386, 5379 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_860_output_0", "outputIndexes": [ 5387 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5387, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_171_output_0", "outputIndexes": [ 5388 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5388, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1831_output_0", "outputIndexes": [ 5389 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5374, 5389, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_229_output_0", "outputIndexes": [ 5390 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5390 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_114_output_0", "outputIndexes": [ 5391 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5388, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1830_output_0", "outputIndexes": [ 5392 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5374, 20, 5392, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_228_output_0", "outputIndexes": [ 5393 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5391, 5393 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_744_output_0", "outputIndexes": [ 5394 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5394, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_457_output_0", "outputIndexes": [ 5395 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5375, 5395 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_285_output_0", "outputIndexes": [ 5396 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5349 ], "outputIndexes": [ 7603 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7603 ], "outputIndexes": [ 7604 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/k_proj/Linear", "inputIndexes": [ 7604 ], "outputIndexes": [ 7605 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 15661715040, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7605 ], "outputIndexes": [ 7606 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7606 ], "outputIndexes": [ 5397 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5361, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1826_output_0", "outputIndexes": [ 5398 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5369, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1827_output_0", "outputIndexes": [ 5399 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5398, 5399, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_742_output_0", "outputIndexes": [ 5400 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5397, 5400 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_515_output_0", "outputIndexes": [ 5401 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5401 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_57/Cast_output_0", "outputIndexes": [ 5402 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_57/Mul_1_output_0", "inputIndexes": [ 5402 ], "outputIndexes": [ 5403 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 15664664182, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5403, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_458_output_0", "outputIndexes": [ 5404 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5403 ], "main_type": "NONE", "name": "/Shape_801_output_0", "outputIndexes": [ 5405 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5405 ], "main_type": "NONE", "name": "Shape22329", "outputIndexes": [ 5406 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5405 ], "main_type": "NONE", "name": "Rank22331", "outputIndexes": [ 5407 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5407, 5407 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22333", "outputIndexes": [ 5408 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5408 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22334", "outputIndexes": [ 5409 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5408, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22336", "outputIndexes": [ 5410 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5410 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22337", "outputIndexes": [ 5411 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5406, 5409, 5411, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice22340", "outputIndexes": [ 5412 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5412 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze22341", "outputIndexes": [ 5413 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5413 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22342", "outputIndexes": [ 5414 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5414, 5413 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22343", "outputIndexes": [ 5415 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5405, 5415, 5408 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_861_output_0", "outputIndexes": [ 5416 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5416, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_172_output_0", "outputIndexes": [ 5417 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5417, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1833_output_0", "outputIndexes": [ 5418 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5403, 5418, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_231_output_0", "outputIndexes": [ 5419 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5419 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_115_output_0", "outputIndexes": [ 5420 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5417, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1832_output_0", "outputIndexes": [ 5421 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5403, 20, 5421, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_230_output_0", "outputIndexes": [ 5422 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5420, 5422 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_745_output_0", "outputIndexes": [ 5423 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5423, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_459_output_0", "outputIndexes": [ 5424 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5404, 5424 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_286_output_0", "outputIndexes": [ 5425 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5349 ], "outputIndexes": [ 7607 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7607 ], "outputIndexes": [ 7608 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/v_proj/Linear", "inputIndexes": [ 7608 ], "outputIndexes": [ 7609 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 15664665206, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7609 ], "outputIndexes": [ 7610 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7610 ], "outputIndexes": [ 5426 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5361, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1828_output_0", "outputIndexes": [ 5427 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5369, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1829_output_0", "outputIndexes": [ 5428 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5427, 5428, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_743_output_0", "outputIndexes": [ 5429 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5426, 5429 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_516_output_0", "outputIndexes": [ 5430 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5396, 5425, 5430, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_521_output_0", "outputIndexes": [ 5431 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5431 ], "outputIndexes": [ 7611 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7611 ], "outputIndexes": [ 7612 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/o_proj/Linear", "inputIndexes": [ 7612 ], "outputIndexes": [ 7613 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 15667614348, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7613 ], "outputIndexes": [ 7614 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7614 ], "outputIndexes": [ 5432 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5348, 5432 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_288_output_0", "outputIndexes": [ 5433 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_57/Mul_1_output_0", "inputIndexes": [ 5433 ], "outputIndexes": [ 5434 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 15691207330, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5434 ], "outputIndexes": [ 7615 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7615 ], "outputIndexes": [ 7616 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/mlp/gate_proj/Linear", "inputIndexes": [ 7616 ], "outputIndexes": [ 7617 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 15691248290, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7617 ], "outputIndexes": [ 7618 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7618 ], "outputIndexes": [ 5435 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5435 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_57/Mul_output_0", "outputIndexes": [ 5436 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5434 ], "outputIndexes": [ 7619 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7619 ], "outputIndexes": [ 7620 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/mlp/up_proj/Linear", "inputIndexes": [ 7620 ], "outputIndexes": [ 7621 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 15764976312, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7621 ], "outputIndexes": [ 7622 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7622 ], "outputIndexes": [ 5437 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5436, 5437 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_57/Mul_output_0", "outputIndexes": [ 5438 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5438 ], "outputIndexes": [ 7623 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7623 ], "outputIndexes": [ 7624 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/mlp/down_proj/Linear", "inputIndexes": [ 7624 ], "outputIndexes": [ 7625 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 15838704334, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7625 ], "outputIndexes": [ 7626 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.57/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7626 ], "outputIndexes": [ 5439 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5433, 5439 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_289_output_0", "outputIndexes": [ 5440 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5440, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_522_output_0", "outputIndexes": [ 5441 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_58/Mul_1_output_0", "inputIndexes": [ 5441 ], "outputIndexes": [ 5442 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 15912432356, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5442 ], "outputIndexes": [ 7627 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7627 ], "outputIndexes": [ 7628 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/q_proj/Linear", "inputIndexes": [ 7628 ], "outputIndexes": [ 7629 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 15912473316, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7629 ], "outputIndexes": [ 7630 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7630 ], "outputIndexes": [ 5443 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5442 ], "main_type": "NONE", "name": "/Shape_812_output_0", "outputIndexes": [ 5444 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5444 ], "main_type": "NONE", "name": "Shape22672", "outputIndexes": [ 5445 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5444 ], "main_type": "NONE", "name": "Rank22674", "outputIndexes": [ 5446 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5446, 5446 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22676", "outputIndexes": [ 5447 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5447 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22677", "outputIndexes": [ 5448 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5447, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22679", "outputIndexes": [ 5449 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5449 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22680", "outputIndexes": [ 5450 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5445, 5448, 5450, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice22683", "outputIndexes": [ 5451 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5451 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze22684", "outputIndexes": [ 5452 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5452, 5452 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22686", "outputIndexes": [ 5453 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5444, 5453, 5447 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_873_output_0", "outputIndexes": [ 5454 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5454, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1856_output_0", "outputIndexes": [ 5455 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5447 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22694", "outputIndexes": [ 5456 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5449 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22697", "outputIndexes": [ 5457 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5445, 5456, 5457, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice22700", "outputIndexes": [ 5458 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5458 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze22701", "outputIndexes": [ 5459 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 5459 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22702", "outputIndexes": [ 5460 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5460, 5459 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22703", "outputIndexes": [ 5461 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5444, 5461, 5447 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_874_output_0", "outputIndexes": [ 5462 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5462, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1857_output_0", "outputIndexes": [ 5463 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5455, 5463, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_754_output_0", "outputIndexes": [ 5464 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5443, 5464 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_523_output_0", "outputIndexes": [ 5465 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5465 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_58/Cast_output_0", "outputIndexes": [ 5466 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_58/Mul_1_output_0", "inputIndexes": [ 5466 ], "outputIndexes": [ 5467 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 15936066298, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5467, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_464_output_0", "outputIndexes": [ 5468 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5467 ], "main_type": "NONE", "name": "/Shape_814_output_0", "outputIndexes": [ 5469 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5469 ], "main_type": "NONE", "name": "Shape22822", "outputIndexes": [ 5470 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5469 ], "main_type": "NONE", "name": "Rank22824", "outputIndexes": [ 5471 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5471, 5471 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22826", "outputIndexes": [ 5472 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5472 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22827", "outputIndexes": [ 5473 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5472, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22829", "outputIndexes": [ 5474 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5474 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22830", "outputIndexes": [ 5475 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5470, 5473, 5475, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice22833", "outputIndexes": [ 5476 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5476 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze22834", "outputIndexes": [ 5477 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5477 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22835", "outputIndexes": [ 5478 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5478, 5477 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22836", "outputIndexes": [ 5479 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5469, 5479, 5472 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_875_output_0", "outputIndexes": [ 5480 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5480, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_174_output_0", "outputIndexes": [ 5481 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5481, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1863_output_0", "outputIndexes": [ 5482 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5467, 5482, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_233_output_0", "outputIndexes": [ 5483 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5483 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_116_output_0", "outputIndexes": [ 5484 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5481, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1862_output_0", "outputIndexes": [ 5485 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5467, 20, 5485, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_232_output_0", "outputIndexes": [ 5486 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5484, 5486 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_757_output_0", "outputIndexes": [ 5487 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5487, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_465_output_0", "outputIndexes": [ 5488 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5468, 5488 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_290_output_0", "outputIndexes": [ 5489 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5442 ], "outputIndexes": [ 7631 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7631 ], "outputIndexes": [ 7632 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/k_proj/Linear", "inputIndexes": [ 7632 ], "outputIndexes": [ 7633 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 15936067322, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7633 ], "outputIndexes": [ 7634 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7634 ], "outputIndexes": [ 5490 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5454, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1858_output_0", "outputIndexes": [ 5491 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5462, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1859_output_0", "outputIndexes": [ 5492 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5491, 5492, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_755_output_0", "outputIndexes": [ 5493 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5490, 5493 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_524_output_0", "outputIndexes": [ 5494 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5494 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_58/Cast_output_0", "outputIndexes": [ 5495 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_58/Mul_1_output_0", "inputIndexes": [ 5495 ], "outputIndexes": [ 5496 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 15939016464, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5496, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_466_output_0", "outputIndexes": [ 5497 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5496 ], "main_type": "NONE", "name": "/Shape_815_output_0", "outputIndexes": [ 5498 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5498 ], "main_type": "NONE", "name": "Shape22718", "outputIndexes": [ 5499 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5498 ], "main_type": "NONE", "name": "Rank22720", "outputIndexes": [ 5500 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5500, 5500 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22722", "outputIndexes": [ 5501 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5501 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22723", "outputIndexes": [ 5502 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5501, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22725", "outputIndexes": [ 5503 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5503 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze22726", "outputIndexes": [ 5504 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5499, 5502, 5504, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice22729", "outputIndexes": [ 5505 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5505 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze22730", "outputIndexes": [ 5506 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5506 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22731", "outputIndexes": [ 5507 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5507, 5506 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp22732", "outputIndexes": [ 5508 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5498, 5508, 5501 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_876_output_0", "outputIndexes": [ 5509 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5509, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_175_output_0", "outputIndexes": [ 5510 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5510, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1865_output_0", "outputIndexes": [ 5511 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5496, 5511, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_235_output_0", "outputIndexes": [ 5512 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5512 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_117_output_0", "outputIndexes": [ 5513 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5510, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1864_output_0", "outputIndexes": [ 5514 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5496, 20, 5514, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_234_output_0", "outputIndexes": [ 5515 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5513, 5515 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_758_output_0", "outputIndexes": [ 5516 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5516, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_467_output_0", "outputIndexes": [ 5517 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5497, 5517 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_291_output_0", "outputIndexes": [ 5518 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5442 ], "outputIndexes": [ 7635 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7635 ], "outputIndexes": [ 7636 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/v_proj/Linear", "inputIndexes": [ 7636 ], "outputIndexes": [ 7637 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 15939017488, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7637 ], "outputIndexes": [ 7638 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7638 ], "outputIndexes": [ 5519 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5454, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1860_output_0", "outputIndexes": [ 5520 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5462, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1861_output_0", "outputIndexes": [ 5521 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5520, 5521, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_756_output_0", "outputIndexes": [ 5522 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5519, 5522 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_525_output_0", "outputIndexes": [ 5523 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5489, 5518, 5523, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_530_output_0", "outputIndexes": [ 5524 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5524 ], "outputIndexes": [ 7639 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7639 ], "outputIndexes": [ 7640 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/o_proj/Linear", "inputIndexes": [ 7640 ], "outputIndexes": [ 7641 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 15941966630, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7641 ], "outputIndexes": [ 7642 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7642 ], "outputIndexes": [ 5525 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5441, 5525 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_293_output_0", "outputIndexes": [ 5526 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_58/Mul_1_output_0", "inputIndexes": [ 5526 ], "outputIndexes": [ 5527 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 15965559612, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5527 ], "outputIndexes": [ 7643 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7643 ], "outputIndexes": [ 7644 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/mlp/gate_proj/Linear", "inputIndexes": [ 7644 ], "outputIndexes": [ 7645 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 15965600572, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7645 ], "outputIndexes": [ 7646 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7646 ], "outputIndexes": [ 5528 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5528 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_58/Mul_output_0", "outputIndexes": [ 5529 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5527 ], "outputIndexes": [ 7647 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7647 ], "outputIndexes": [ 7648 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/mlp/up_proj/Linear", "inputIndexes": [ 7648 ], "outputIndexes": [ 7649 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 16039328594, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7649 ], "outputIndexes": [ 7650 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7650 ], "outputIndexes": [ 5530 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5529, 5530 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_58/Mul_output_0", "outputIndexes": [ 5531 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5531 ], "outputIndexes": [ 7651 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7651 ], "outputIndexes": [ 7652 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/mlp/down_proj/Linear", "inputIndexes": [ 7652 ], "outputIndexes": [ 7653 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 16113056616, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7653 ], "outputIndexes": [ 7654 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.58/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7654 ], "outputIndexes": [ 5532 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5526, 5532 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_294_output_0", "outputIndexes": [ 5533 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5533, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_531_output_0", "outputIndexes": [ 5534 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_59/Mul_1_output_0", "inputIndexes": [ 5534 ], "outputIndexes": [ 5535 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 16186784638, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5535 ], "outputIndexes": [ 7655 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7655 ], "outputIndexes": [ 7656 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/q_proj/Linear", "inputIndexes": [ 7656 ], "outputIndexes": [ 7657 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 16186825598, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7657 ], "outputIndexes": [ 7658 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7658 ], "outputIndexes": [ 5536 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5535 ], "main_type": "NONE", "name": "/Shape_826_output_0", "outputIndexes": [ 5537 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5537 ], "main_type": "NONE", "name": "Shape23061", "outputIndexes": [ 5538 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5537 ], "main_type": "NONE", "name": "Rank23063", "outputIndexes": [ 5539 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5539, 5539 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23065", "outputIndexes": [ 5540 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5540 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23066", "outputIndexes": [ 5541 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5540, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23068", "outputIndexes": [ 5542 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5542 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23069", "outputIndexes": [ 5543 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5538, 5541, 5543, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice23072", "outputIndexes": [ 5544 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5544 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze23073", "outputIndexes": [ 5545 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5545, 5545 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23075", "outputIndexes": [ 5546 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5537, 5546, 5540 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_888_output_0", "outputIndexes": [ 5547 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5547, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1888_output_0", "outputIndexes": [ 5548 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5540 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23083", "outputIndexes": [ 5549 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5542 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23086", "outputIndexes": [ 5550 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5538, 5549, 5550, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice23089", "outputIndexes": [ 5551 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5551 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze23090", "outputIndexes": [ 5552 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 5552 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23091", "outputIndexes": [ 5553 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5553, 5552 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23092", "outputIndexes": [ 5554 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5537, 5554, 5540 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_889_output_0", "outputIndexes": [ 5555 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5555, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1889_output_0", "outputIndexes": [ 5556 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5548, 5556, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_767_output_0", "outputIndexes": [ 5557 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5536, 5557 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_532_output_0", "outputIndexes": [ 5558 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5558 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_59/Cast_output_0", "outputIndexes": [ 5559 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_59/Mul_1_output_0", "inputIndexes": [ 5559 ], "outputIndexes": [ 5560 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 16210418580, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5560, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_472_output_0", "outputIndexes": [ 5561 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5560 ], "main_type": "NONE", "name": "/Shape_828_output_0", "outputIndexes": [ 5562 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5562 ], "main_type": "NONE", "name": "Shape23211", "outputIndexes": [ 5563 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5562 ], "main_type": "NONE", "name": "Rank23213", "outputIndexes": [ 5564 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5564, 5564 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23215", "outputIndexes": [ 5565 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5565 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23216", "outputIndexes": [ 5566 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5565, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23218", "outputIndexes": [ 5567 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5567 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23219", "outputIndexes": [ 5568 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5563, 5566, 5568, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice23222", "outputIndexes": [ 5569 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5569 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze23223", "outputIndexes": [ 5570 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5570 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23224", "outputIndexes": [ 5571 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5571, 5570 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23225", "outputIndexes": [ 5572 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5562, 5572, 5565 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_890_output_0", "outputIndexes": [ 5573 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5573, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_177_output_0", "outputIndexes": [ 5574 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5574, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1895_output_0", "outputIndexes": [ 5575 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5560, 5575, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_237_output_0", "outputIndexes": [ 5576 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5576 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_118_output_0", "outputIndexes": [ 5577 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5574, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1894_output_0", "outputIndexes": [ 5578 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5560, 20, 5578, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_236_output_0", "outputIndexes": [ 5579 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5577, 5579 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_770_output_0", "outputIndexes": [ 5580 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5580, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_473_output_0", "outputIndexes": [ 5581 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5561, 5581 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_295_output_0", "outputIndexes": [ 5582 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5535 ], "outputIndexes": [ 7659 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7659 ], "outputIndexes": [ 7660 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/k_proj/Linear", "inputIndexes": [ 7660 ], "outputIndexes": [ 7661 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 16210419604, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7661 ], "outputIndexes": [ 7662 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7662 ], "outputIndexes": [ 5583 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5547, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1890_output_0", "outputIndexes": [ 5584 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5555, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1891_output_0", "outputIndexes": [ 5585 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5584, 5585, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_768_output_0", "outputIndexes": [ 5586 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5583, 5586 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_533_output_0", "outputIndexes": [ 5587 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5587 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_59/Cast_output_0", "outputIndexes": [ 5588 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_59/Mul_1_output_0", "inputIndexes": [ 5588 ], "outputIndexes": [ 5589 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 16213368746, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5589, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_474_output_0", "outputIndexes": [ 5590 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5589 ], "main_type": "NONE", "name": "/Shape_829_output_0", "outputIndexes": [ 5591 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5591 ], "main_type": "NONE", "name": "Shape23107", "outputIndexes": [ 5592 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5591 ], "main_type": "NONE", "name": "Rank23109", "outputIndexes": [ 5593 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5593, 5593 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23111", "outputIndexes": [ 5594 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5594 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23112", "outputIndexes": [ 5595 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5594, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23114", "outputIndexes": [ 5596 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5596 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23115", "outputIndexes": [ 5597 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5592, 5595, 5597, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice23118", "outputIndexes": [ 5598 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5598 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze23119", "outputIndexes": [ 5599 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5599 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23120", "outputIndexes": [ 5600 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5600, 5599 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23121", "outputIndexes": [ 5601 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5591, 5601, 5594 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_891_output_0", "outputIndexes": [ 5602 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5602, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_178_output_0", "outputIndexes": [ 5603 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5603, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1897_output_0", "outputIndexes": [ 5604 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5589, 5604, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_239_output_0", "outputIndexes": [ 5605 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5605 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_119_output_0", "outputIndexes": [ 5606 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5603, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1896_output_0", "outputIndexes": [ 5607 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5589, 20, 5607, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_238_output_0", "outputIndexes": [ 5608 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5606, 5608 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_771_output_0", "outputIndexes": [ 5609 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5609, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_475_output_0", "outputIndexes": [ 5610 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5590, 5610 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_296_output_0", "outputIndexes": [ 5611 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5535 ], "outputIndexes": [ 7663 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7663 ], "outputIndexes": [ 7664 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/v_proj/Linear", "inputIndexes": [ 7664 ], "outputIndexes": [ 7665 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 16213369770, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7665 ], "outputIndexes": [ 7666 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7666 ], "outputIndexes": [ 5612 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5547, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1892_output_0", "outputIndexes": [ 5613 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5555, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1893_output_0", "outputIndexes": [ 5614 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5613, 5614, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_769_output_0", "outputIndexes": [ 5615 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5612, 5615 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_534_output_0", "outputIndexes": [ 5616 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5582, 5611, 5616, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_539_output_0", "outputIndexes": [ 5617 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5617 ], "outputIndexes": [ 7667 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7667 ], "outputIndexes": [ 7668 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/o_proj/Linear", "inputIndexes": [ 7668 ], "outputIndexes": [ 7669 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 16216318912, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7669 ], "outputIndexes": [ 7670 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7670 ], "outputIndexes": [ 5618 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5534, 5618 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_298_output_0", "outputIndexes": [ 5619 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_59/Mul_1_output_0", "inputIndexes": [ 5619 ], "outputIndexes": [ 5620 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 16239911894, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5620 ], "outputIndexes": [ 7671 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7671 ], "outputIndexes": [ 7672 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/mlp/gate_proj/Linear", "inputIndexes": [ 7672 ], "outputIndexes": [ 7673 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 16239952854, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7673 ], "outputIndexes": [ 7674 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7674 ], "outputIndexes": [ 5621 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5621 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_59/Mul_output_0", "outputIndexes": [ 5622 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5620 ], "outputIndexes": [ 7675 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7675 ], "outputIndexes": [ 7676 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/mlp/up_proj/Linear", "inputIndexes": [ 7676 ], "outputIndexes": [ 7677 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 16313680876, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7677 ], "outputIndexes": [ 7678 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7678 ], "outputIndexes": [ 5623 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5622, 5623 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_59/Mul_output_0", "outputIndexes": [ 5624 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5624 ], "outputIndexes": [ 7679 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7679 ], "outputIndexes": [ 7680 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/mlp/down_proj/Linear", "inputIndexes": [ 7680 ], "outputIndexes": [ 7681 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 16387408898, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7681 ], "outputIndexes": [ 7682 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.59/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7682 ], "outputIndexes": [ 5625 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5619, 5625 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_299_output_0", "outputIndexes": [ 5626 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5626, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_540_output_0", "outputIndexes": [ 5627 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_60/Mul_1_output_0", "inputIndexes": [ 5627 ], "outputIndexes": [ 5628 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 16461136920, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5628 ], "outputIndexes": [ 7683 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7683 ], "outputIndexes": [ 7684 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/q_proj/Linear", "inputIndexes": [ 7684 ], "outputIndexes": [ 7685 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 16461177880, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7685 ], "outputIndexes": [ 7686 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7686 ], "outputIndexes": [ 5629 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5628 ], "main_type": "NONE", "name": "/Shape_840_output_0", "outputIndexes": [ 5630 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5630 ], "main_type": "NONE", "name": "Shape23450", "outputIndexes": [ 5631 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5630 ], "main_type": "NONE", "name": "Rank23452", "outputIndexes": [ 5632 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5632, 5632 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23454", "outputIndexes": [ 5633 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5633 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23455", "outputIndexes": [ 5634 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5633, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23457", "outputIndexes": [ 5635 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5635 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23458", "outputIndexes": [ 5636 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5631, 5634, 5636, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice23461", "outputIndexes": [ 5637 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5637 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze23462", "outputIndexes": [ 5638 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5638, 5638 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23464", "outputIndexes": [ 5639 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5630, 5639, 5633 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_903_output_0", "outputIndexes": [ 5640 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5640, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1920_output_0", "outputIndexes": [ 5641 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5633 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23472", "outputIndexes": [ 5642 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5635 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23475", "outputIndexes": [ 5643 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5631, 5642, 5643, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice23478", "outputIndexes": [ 5644 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5644 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze23479", "outputIndexes": [ 5645 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 5645 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23480", "outputIndexes": [ 5646 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5646, 5645 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23481", "outputIndexes": [ 5647 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5630, 5647, 5633 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_904_output_0", "outputIndexes": [ 5648 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5648, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1921_output_0", "outputIndexes": [ 5649 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5641, 5649, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_780_output_0", "outputIndexes": [ 5650 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5629, 5650 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_541_output_0", "outputIndexes": [ 5651 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5651 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_60/Cast_output_0", "outputIndexes": [ 5652 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_60/Mul_1_output_0", "inputIndexes": [ 5652 ], "outputIndexes": [ 5653 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 16484770862, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5653, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_480_output_0", "outputIndexes": [ 5654 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5653 ], "main_type": "NONE", "name": "/Shape_842_output_0", "outputIndexes": [ 5655 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5655 ], "main_type": "NONE", "name": "Shape23600", "outputIndexes": [ 5656 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5655 ], "main_type": "NONE", "name": "Rank23602", "outputIndexes": [ 5657 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5657, 5657 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23604", "outputIndexes": [ 5658 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5658 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23605", "outputIndexes": [ 5659 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5658, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23607", "outputIndexes": [ 5660 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5660 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23608", "outputIndexes": [ 5661 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5656, 5659, 5661, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice23611", "outputIndexes": [ 5662 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5662 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze23612", "outputIndexes": [ 5663 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5663 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23613", "outputIndexes": [ 5664 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5664, 5663 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23614", "outputIndexes": [ 5665 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5655, 5665, 5658 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_905_output_0", "outputIndexes": [ 5666 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5666, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_180_output_0", "outputIndexes": [ 5667 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5667, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1927_output_0", "outputIndexes": [ 5668 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5653, 5668, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_241_output_0", "outputIndexes": [ 5669 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5669 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_120_output_0", "outputIndexes": [ 5670 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5667, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1926_output_0", "outputIndexes": [ 5671 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5653, 20, 5671, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_240_output_0", "outputIndexes": [ 5672 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5670, 5672 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_783_output_0", "outputIndexes": [ 5673 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5673, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_481_output_0", "outputIndexes": [ 5674 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5654, 5674 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_300_output_0", "outputIndexes": [ 5675 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5628 ], "outputIndexes": [ 7687 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7687 ], "outputIndexes": [ 7688 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/k_proj/Linear", "inputIndexes": [ 7688 ], "outputIndexes": [ 7689 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 16484771886, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7689 ], "outputIndexes": [ 7690 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7690 ], "outputIndexes": [ 5676 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5640, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1922_output_0", "outputIndexes": [ 5677 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5648, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1923_output_0", "outputIndexes": [ 5678 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5677, 5678, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_781_output_0", "outputIndexes": [ 5679 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5676, 5679 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_542_output_0", "outputIndexes": [ 5680 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5680 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_60/Cast_output_0", "outputIndexes": [ 5681 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_60/Mul_1_output_0", "inputIndexes": [ 5681 ], "outputIndexes": [ 5682 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 16487721028, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5682, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_482_output_0", "outputIndexes": [ 5683 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5682 ], "main_type": "NONE", "name": "/Shape_843_output_0", "outputIndexes": [ 5684 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5684 ], "main_type": "NONE", "name": "Shape23496", "outputIndexes": [ 5685 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5684 ], "main_type": "NONE", "name": "Rank23498", "outputIndexes": [ 5686 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5686, 5686 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23500", "outputIndexes": [ 5687 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5687 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23501", "outputIndexes": [ 5688 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5687, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23503", "outputIndexes": [ 5689 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5689 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23504", "outputIndexes": [ 5690 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5685, 5688, 5690, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice23507", "outputIndexes": [ 5691 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5691 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze23508", "outputIndexes": [ 5692 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5692 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23509", "outputIndexes": [ 5693 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5693, 5692 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23510", "outputIndexes": [ 5694 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5684, 5694, 5687 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_906_output_0", "outputIndexes": [ 5695 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5695, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_181_output_0", "outputIndexes": [ 5696 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5696, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1929_output_0", "outputIndexes": [ 5697 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5682, 5697, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_243_output_0", "outputIndexes": [ 5698 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5698 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_121_output_0", "outputIndexes": [ 5699 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5696, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1928_output_0", "outputIndexes": [ 5700 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5682, 20, 5700, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_242_output_0", "outputIndexes": [ 5701 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5699, 5701 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_784_output_0", "outputIndexes": [ 5702 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5702, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_483_output_0", "outputIndexes": [ 5703 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5683, 5703 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_301_output_0", "outputIndexes": [ 5704 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5628 ], "outputIndexes": [ 7691 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7691 ], "outputIndexes": [ 7692 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/v_proj/Linear", "inputIndexes": [ 7692 ], "outputIndexes": [ 7693 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 16487722052, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7693 ], "outputIndexes": [ 7694 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7694 ], "outputIndexes": [ 5705 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5640, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1924_output_0", "outputIndexes": [ 5706 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5648, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1925_output_0", "outputIndexes": [ 5707 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5706, 5707, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_782_output_0", "outputIndexes": [ 5708 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5705, 5708 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_543_output_0", "outputIndexes": [ 5709 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5675, 5704, 5709, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_548_output_0", "outputIndexes": [ 5710 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5710 ], "outputIndexes": [ 7695 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7695 ], "outputIndexes": [ 7696 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/o_proj/Linear", "inputIndexes": [ 7696 ], "outputIndexes": [ 7697 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 16490671194, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7697 ], "outputIndexes": [ 7698 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7698 ], "outputIndexes": [ 5711 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5627, 5711 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_303_output_0", "outputIndexes": [ 5712 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_60/Mul_1_output_0", "inputIndexes": [ 5712 ], "outputIndexes": [ 5713 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 16514264176, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5713 ], "outputIndexes": [ 7699 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7699 ], "outputIndexes": [ 7700 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/mlp/gate_proj/Linear", "inputIndexes": [ 7700 ], "outputIndexes": [ 7701 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 16514305136, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7701 ], "outputIndexes": [ 7702 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7702 ], "outputIndexes": [ 5714 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5714 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_60/Mul_output_0", "outputIndexes": [ 5715 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5713 ], "outputIndexes": [ 7703 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7703 ], "outputIndexes": [ 7704 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/mlp/up_proj/Linear", "inputIndexes": [ 7704 ], "outputIndexes": [ 7705 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 16588033158, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7705 ], "outputIndexes": [ 7706 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7706 ], "outputIndexes": [ 5716 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5715, 5716 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_60/Mul_output_0", "outputIndexes": [ 5717 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5717 ], "outputIndexes": [ 7707 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7707 ], "outputIndexes": [ 7708 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/mlp/down_proj/Linear", "inputIndexes": [ 7708 ], "outputIndexes": [ 7709 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 16661761180, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7709 ], "outputIndexes": [ 7710 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.60/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7710 ], "outputIndexes": [ 5718 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5712, 5718 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_304_output_0", "outputIndexes": [ 5719 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5719, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_549_output_0", "outputIndexes": [ 5720 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_61/Mul_1_output_0", "inputIndexes": [ 5720 ], "outputIndexes": [ 5721 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 16735489202, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5721 ], "outputIndexes": [ 7711 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7711 ], "outputIndexes": [ 7712 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/q_proj/Linear", "inputIndexes": [ 7712 ], "outputIndexes": [ 7713 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 16735530162, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7713 ], "outputIndexes": [ 7714 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7714 ], "outputIndexes": [ 5722 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5721 ], "main_type": "NONE", "name": "/Shape_854_output_0", "outputIndexes": [ 5723 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5723 ], "main_type": "NONE", "name": "Shape23839", "outputIndexes": [ 5724 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5723 ], "main_type": "NONE", "name": "Rank23841", "outputIndexes": [ 5725 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5725, 5725 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23843", "outputIndexes": [ 5726 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5726 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23844", "outputIndexes": [ 5727 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5726, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23846", "outputIndexes": [ 5728 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5728 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23847", "outputIndexes": [ 5729 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5724, 5727, 5729, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice23850", "outputIndexes": [ 5730 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5730 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze23851", "outputIndexes": [ 5731 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5731, 5731 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23853", "outputIndexes": [ 5732 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5723, 5732, 5726 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_918_output_0", "outputIndexes": [ 5733 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5733, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1952_output_0", "outputIndexes": [ 5734 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5726 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23861", "outputIndexes": [ 5735 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5728 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23864", "outputIndexes": [ 5736 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5724, 5735, 5736, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice23867", "outputIndexes": [ 5737 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5737 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze23868", "outputIndexes": [ 5738 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 5738 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23869", "outputIndexes": [ 5739 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5739, 5738 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23870", "outputIndexes": [ 5740 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5723, 5740, 5726 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_919_output_0", "outputIndexes": [ 5741 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5741, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1953_output_0", "outputIndexes": [ 5742 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5734, 5742, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_793_output_0", "outputIndexes": [ 5743 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5722, 5743 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_550_output_0", "outputIndexes": [ 5744 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5744 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_61/Cast_output_0", "outputIndexes": [ 5745 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_61/Mul_1_output_0", "inputIndexes": [ 5745 ], "outputIndexes": [ 5746 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 16759123144, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5746, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_488_output_0", "outputIndexes": [ 5747 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5746 ], "main_type": "NONE", "name": "/Shape_856_output_0", "outputIndexes": [ 5748 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5748 ], "main_type": "NONE", "name": "Shape23989", "outputIndexes": [ 5749 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5748 ], "main_type": "NONE", "name": "Rank23991", "outputIndexes": [ 5750 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5750, 5750 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23993", "outputIndexes": [ 5751 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5751 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23994", "outputIndexes": [ 5752 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5751, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23996", "outputIndexes": [ 5753 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5753 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23997", "outputIndexes": [ 5754 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5749, 5752, 5754, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice24000", "outputIndexes": [ 5755 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5755 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze24001", "outputIndexes": [ 5756 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5756 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24002", "outputIndexes": [ 5757 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5757, 5756 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24003", "outputIndexes": [ 5758 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5748, 5758, 5751 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_920_output_0", "outputIndexes": [ 5759 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5759, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_183_output_0", "outputIndexes": [ 5760 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5760, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1959_output_0", "outputIndexes": [ 5761 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5746, 5761, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_245_output_0", "outputIndexes": [ 5762 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5762 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_122_output_0", "outputIndexes": [ 5763 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5760, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1958_output_0", "outputIndexes": [ 5764 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5746, 20, 5764, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_244_output_0", "outputIndexes": [ 5765 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5763, 5765 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_796_output_0", "outputIndexes": [ 5766 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5766, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_489_output_0", "outputIndexes": [ 5767 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5747, 5767 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_305_output_0", "outputIndexes": [ 5768 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5721 ], "outputIndexes": [ 7715 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7715 ], "outputIndexes": [ 7716 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/k_proj/Linear", "inputIndexes": [ 7716 ], "outputIndexes": [ 7717 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 16759124168, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7717 ], "outputIndexes": [ 7718 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7718 ], "outputIndexes": [ 5769 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5733, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1954_output_0", "outputIndexes": [ 5770 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5741, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1955_output_0", "outputIndexes": [ 5771 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5770, 5771, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_794_output_0", "outputIndexes": [ 5772 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5769, 5772 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_551_output_0", "outputIndexes": [ 5773 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5773 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_61/Cast_output_0", "outputIndexes": [ 5774 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_61/Mul_1_output_0", "inputIndexes": [ 5774 ], "outputIndexes": [ 5775 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 16762073310, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5775, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_490_output_0", "outputIndexes": [ 5776 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5775 ], "main_type": "NONE", "name": "/Shape_857_output_0", "outputIndexes": [ 5777 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5777 ], "main_type": "NONE", "name": "Shape23885", "outputIndexes": [ 5778 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5777 ], "main_type": "NONE", "name": "Rank23887", "outputIndexes": [ 5779 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5779, 5779 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23889", "outputIndexes": [ 5780 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5780 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23890", "outputIndexes": [ 5781 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5780, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23892", "outputIndexes": [ 5782 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5782 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze23893", "outputIndexes": [ 5783 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5778, 5781, 5783, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice23896", "outputIndexes": [ 5784 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5784 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze23897", "outputIndexes": [ 5785 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5785 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23898", "outputIndexes": [ 5786 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5786, 5785 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp23899", "outputIndexes": [ 5787 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5777, 5787, 5780 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_921_output_0", "outputIndexes": [ 5788 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5788, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_184_output_0", "outputIndexes": [ 5789 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5789, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1961_output_0", "outputIndexes": [ 5790 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5775, 5790, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_247_output_0", "outputIndexes": [ 5791 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5791 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_123_output_0", "outputIndexes": [ 5792 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5789, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1960_output_0", "outputIndexes": [ 5793 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5775, 20, 5793, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_246_output_0", "outputIndexes": [ 5794 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5792, 5794 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_797_output_0", "outputIndexes": [ 5795 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5795, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_491_output_0", "outputIndexes": [ 5796 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5776, 5796 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_306_output_0", "outputIndexes": [ 5797 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5721 ], "outputIndexes": [ 7719 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7719 ], "outputIndexes": [ 7720 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/v_proj/Linear", "inputIndexes": [ 7720 ], "outputIndexes": [ 7721 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 16762074334, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7721 ], "outputIndexes": [ 7722 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7722 ], "outputIndexes": [ 5798 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5733, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1956_output_0", "outputIndexes": [ 5799 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5741, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1957_output_0", "outputIndexes": [ 5800 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5799, 5800, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_795_output_0", "outputIndexes": [ 5801 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5798, 5801 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_552_output_0", "outputIndexes": [ 5802 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5768, 5797, 5802, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_557_output_0", "outputIndexes": [ 5803 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5803 ], "outputIndexes": [ 7723 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7723 ], "outputIndexes": [ 7724 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/o_proj/Linear", "inputIndexes": [ 7724 ], "outputIndexes": [ 7725 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 16765023476, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7725 ], "outputIndexes": [ 7726 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7726 ], "outputIndexes": [ 5804 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5720, 5804 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_308_output_0", "outputIndexes": [ 5805 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_61/Mul_1_output_0", "inputIndexes": [ 5805 ], "outputIndexes": [ 5806 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 16788616458, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5806 ], "outputIndexes": [ 7727 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7727 ], "outputIndexes": [ 7728 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/mlp/gate_proj/Linear", "inputIndexes": [ 7728 ], "outputIndexes": [ 7729 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 16788657418, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7729 ], "outputIndexes": [ 7730 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7730 ], "outputIndexes": [ 5807 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5807 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_61/Mul_output_0", "outputIndexes": [ 5808 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5806 ], "outputIndexes": [ 7731 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7731 ], "outputIndexes": [ 7732 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/mlp/up_proj/Linear", "inputIndexes": [ 7732 ], "outputIndexes": [ 7733 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 16862385440, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7733 ], "outputIndexes": [ 7734 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7734 ], "outputIndexes": [ 5809 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5808, 5809 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_61/Mul_output_0", "outputIndexes": [ 5810 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5810 ], "outputIndexes": [ 7735 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7735 ], "outputIndexes": [ 7736 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/mlp/down_proj/Linear", "inputIndexes": [ 7736 ], "outputIndexes": [ 7737 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 16936113462, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7737 ], "outputIndexes": [ 7738 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.61/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7738 ], "outputIndexes": [ 5811 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5805, 5811 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_309_output_0", "outputIndexes": [ 5812 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5812, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_558_output_0", "outputIndexes": [ 5813 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_62/Mul_1_output_0", "inputIndexes": [ 5813 ], "outputIndexes": [ 5814 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 17009841484, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5814 ], "outputIndexes": [ 7739 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7739 ], "outputIndexes": [ 7740 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/q_proj/Linear", "inputIndexes": [ 7740 ], "outputIndexes": [ 7741 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 17009882444, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7741 ], "outputIndexes": [ 7742 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7742 ], "outputIndexes": [ 5815 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5814 ], "main_type": "NONE", "name": "/Shape_868_output_0", "outputIndexes": [ 5816 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5816 ], "main_type": "NONE", "name": "Shape24228", "outputIndexes": [ 5817 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5816 ], "main_type": "NONE", "name": "Rank24230", "outputIndexes": [ 5818 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5818, 5818 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24232", "outputIndexes": [ 5819 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5819 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze24233", "outputIndexes": [ 5820 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5819, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24235", "outputIndexes": [ 5821 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5821 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze24236", "outputIndexes": [ 5822 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5817, 5820, 5822, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice24239", "outputIndexes": [ 5823 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5823 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze24240", "outputIndexes": [ 5824 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5824, 5824 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24242", "outputIndexes": [ 5825 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5816, 5825, 5819 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_933_output_0", "outputIndexes": [ 5826 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5826, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1984_output_0", "outputIndexes": [ 5827 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5819 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze24250", "outputIndexes": [ 5828 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5821 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze24253", "outputIndexes": [ 5829 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5817, 5828, 5829, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice24256", "outputIndexes": [ 5830 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5830 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze24257", "outputIndexes": [ 5831 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 5831 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24258", "outputIndexes": [ 5832 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5832, 5831 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24259", "outputIndexes": [ 5833 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5816, 5833, 5819 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_934_output_0", "outputIndexes": [ 5834 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5834, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1985_output_0", "outputIndexes": [ 5835 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5827, 5835, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_806_output_0", "outputIndexes": [ 5836 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5815, 5836 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_559_output_0", "outputIndexes": [ 5837 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5837 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_62/Cast_output_0", "outputIndexes": [ 5838 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_62/Mul_1_output_0", "inputIndexes": [ 5838 ], "outputIndexes": [ 5839 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 17033475426, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5839, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_496_output_0", "outputIndexes": [ 5840 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5839 ], "main_type": "NONE", "name": "/Shape_870_output_0", "outputIndexes": [ 5841 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5841 ], "main_type": "NONE", "name": "Shape24378", "outputIndexes": [ 5842 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5841 ], "main_type": "NONE", "name": "Rank24380", "outputIndexes": [ 5843 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5843, 5843 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24382", "outputIndexes": [ 5844 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5844 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze24383", "outputIndexes": [ 5845 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5844, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24385", "outputIndexes": [ 5846 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5846 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze24386", "outputIndexes": [ 5847 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5842, 5845, 5847, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice24389", "outputIndexes": [ 5848 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5848 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze24390", "outputIndexes": [ 5849 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5849 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24391", "outputIndexes": [ 5850 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5850, 5849 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24392", "outputIndexes": [ 5851 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5841, 5851, 5844 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_935_output_0", "outputIndexes": [ 5852 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5852, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_186_output_0", "outputIndexes": [ 5853 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5853, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1991_output_0", "outputIndexes": [ 5854 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5839, 5854, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_249_output_0", "outputIndexes": [ 5855 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5855 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_124_output_0", "outputIndexes": [ 5856 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5853, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1990_output_0", "outputIndexes": [ 5857 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5839, 20, 5857, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_248_output_0", "outputIndexes": [ 5858 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5856, 5858 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_809_output_0", "outputIndexes": [ 5859 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5859, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_497_output_0", "outputIndexes": [ 5860 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5840, 5860 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_310_output_0", "outputIndexes": [ 5861 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5814 ], "outputIndexes": [ 7743 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7743 ], "outputIndexes": [ 7744 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/k_proj/Linear", "inputIndexes": [ 7744 ], "outputIndexes": [ 7745 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 17033476450, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7745 ], "outputIndexes": [ 7746 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7746 ], "outputIndexes": [ 5862 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5826, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1986_output_0", "outputIndexes": [ 5863 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5834, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1987_output_0", "outputIndexes": [ 5864 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5863, 5864, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_807_output_0", "outputIndexes": [ 5865 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5862, 5865 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_560_output_0", "outputIndexes": [ 5866 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5866 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_62/Cast_output_0", "outputIndexes": [ 5867 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_62/Mul_1_output_0", "inputIndexes": [ 5867 ], "outputIndexes": [ 5868 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 17036425592, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5868, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_498_output_0", "outputIndexes": [ 5869 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5868 ], "main_type": "NONE", "name": "/Shape_871_output_0", "outputIndexes": [ 5870 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5870 ], "main_type": "NONE", "name": "Shape24274", "outputIndexes": [ 5871 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5870 ], "main_type": "NONE", "name": "Rank24276", "outputIndexes": [ 5872 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5872, 5872 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24278", "outputIndexes": [ 5873 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5873 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze24279", "outputIndexes": [ 5874 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5873, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24281", "outputIndexes": [ 5875 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5875 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze24282", "outputIndexes": [ 5876 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5871, 5874, 5876, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice24285", "outputIndexes": [ 5877 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5877 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze24286", "outputIndexes": [ 5878 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5878 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24287", "outputIndexes": [ 5879 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5879, 5878 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24288", "outputIndexes": [ 5880 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5870, 5880, 5873 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_936_output_0", "outputIndexes": [ 5881 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5881, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_187_output_0", "outputIndexes": [ 5882 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5882, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1993_output_0", "outputIndexes": [ 5883 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5868, 5883, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_251_output_0", "outputIndexes": [ 5884 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5884 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_125_output_0", "outputIndexes": [ 5885 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5882, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1992_output_0", "outputIndexes": [ 5886 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5868, 20, 5886, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_250_output_0", "outputIndexes": [ 5887 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5885, 5887 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_810_output_0", "outputIndexes": [ 5888 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5888, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_499_output_0", "outputIndexes": [ 5889 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5869, 5889 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_311_output_0", "outputIndexes": [ 5890 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5814 ], "outputIndexes": [ 7747 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7747 ], "outputIndexes": [ 7748 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/v_proj/Linear", "inputIndexes": [ 7748 ], "outputIndexes": [ 7749 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 17036426616, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7749 ], "outputIndexes": [ 7750 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7750 ], "outputIndexes": [ 5891 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5826, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1988_output_0", "outputIndexes": [ 5892 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5834, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_1989_output_0", "outputIndexes": [ 5893 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5892, 5893, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_808_output_0", "outputIndexes": [ 5894 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5891, 5894 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_561_output_0", "outputIndexes": [ 5895 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5861, 5890, 5895, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_566_output_0", "outputIndexes": [ 5896 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5896 ], "outputIndexes": [ 7751 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7751 ], "outputIndexes": [ 7752 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/o_proj/Linear", "inputIndexes": [ 7752 ], "outputIndexes": [ 7753 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 17039375758, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7753 ], "outputIndexes": [ 7754 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7754 ], "outputIndexes": [ 5897 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5813, 5897 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_313_output_0", "outputIndexes": [ 5898 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_62/Mul_1_output_0", "inputIndexes": [ 5898 ], "outputIndexes": [ 5899 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 17062968740, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5899 ], "outputIndexes": [ 7755 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7755 ], "outputIndexes": [ 7756 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/mlp/gate_proj/Linear", "inputIndexes": [ 7756 ], "outputIndexes": [ 7757 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 17063009700, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7757 ], "outputIndexes": [ 7758 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7758 ], "outputIndexes": [ 5900 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5900 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_62/Mul_output_0", "outputIndexes": [ 5901 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5899 ], "outputIndexes": [ 7759 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7759 ], "outputIndexes": [ 7760 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/mlp/up_proj/Linear", "inputIndexes": [ 7760 ], "outputIndexes": [ 7761 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 17136737722, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7761 ], "outputIndexes": [ 7762 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7762 ], "outputIndexes": [ 5902 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5901, 5902 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_62/Mul_output_0", "outputIndexes": [ 5903 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5903 ], "outputIndexes": [ 7763 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7763 ], "outputIndexes": [ 7764 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/mlp/down_proj/Linear", "inputIndexes": [ 7764 ], "outputIndexes": [ 7765 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 17210465744, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7765 ], "outputIndexes": [ 7766 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.62/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7766 ], "outputIndexes": [ 5904 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5898, 5904 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_314_output_0", "outputIndexes": [ 5905 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5905, 3 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_567_output_0", "outputIndexes": [ 5906 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "name": "/input_layernorm_63/Mul_1_output_0", "inputIndexes": [ 5906 ], "outputIndexes": [ 5907 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 17284193766, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/q_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5907 ], "outputIndexes": [ 7767 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/q_proj/Linear/pre_convert", "inputIndexes": [ 7767 ], "outputIndexes": [ 7768 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/q_proj/Linear", "inputIndexes": [ 7768 ], "outputIndexes": [ 7769 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 8192, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 17284234726, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/q_proj/Linear/post_convert", "inputIndexes": [ 7769 ], "outputIndexes": [ 7770 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/q_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7770 ], "outputIndexes": [ 5908 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 8192 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5907 ], "main_type": "NONE", "name": "/Shape_882_output_0", "outputIndexes": [ 5909 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5909 ], "main_type": "NONE", "name": "Shape24617", "outputIndexes": [ 5910 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5909 ], "main_type": "NONE", "name": "Rank24619", "outputIndexes": [ 5911 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5911, 5911 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24621", "outputIndexes": [ 5912 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5912 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze24622", "outputIndexes": [ 5913 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5912, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24624", "outputIndexes": [ 5914 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5914 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze24625", "outputIndexes": [ 5915 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5910, 5913, 5915, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice24628", "outputIndexes": [ 5916 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5916 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze24629", "outputIndexes": [ 5917 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5917, 5917 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24631", "outputIndexes": [ 5918 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5909, 5918, 5912 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_948_output_0", "outputIndexes": [ 5919 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5919, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_2016_output_0", "outputIndexes": [ 5920 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5912 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze24639", "outputIndexes": [ 5921 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5914 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze24642", "outputIndexes": [ 5922 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5910, 5921, 5922, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice24645", "outputIndexes": [ 5923 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5923 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze24646", "outputIndexes": [ 5924 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 12, 5924 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24647", "outputIndexes": [ 5925 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5925, 5924 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24648", "outputIndexes": [ 5926 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5909, 5926, 5912 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_949_output_0", "outputIndexes": [ 5927 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5927, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_2017_output_0", "outputIndexes": [ 5928 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5920, 5928, 30, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_819_output_0", "outputIndexes": [ 5929 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5908, 5929 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_568_output_0", "outputIndexes": [ 5930 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5930 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/q_norm_63/Cast_output_0", "outputIndexes": [ 5931 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/q_norm_63/Mul_1_output_0", "inputIndexes": [ 5931 ], "outputIndexes": [ 5932 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 17307827708, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5932, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_504_output_0", "outputIndexes": [ 5933 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5932 ], "main_type": "NONE", "name": "/Shape_884_output_0", "outputIndexes": [ 5934 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5934 ], "main_type": "NONE", "name": "Shape24734", "outputIndexes": [ 5935 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5934 ], "main_type": "NONE", "name": "Rank24736", "outputIndexes": [ 5936 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5936, 5936 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24738", "outputIndexes": [ 5937 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5937 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze24739", "outputIndexes": [ 5938 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5937, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24741", "outputIndexes": [ 5939 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5939 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze24742", "outputIndexes": [ 5940 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5935, 5938, 5940, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice24745", "outputIndexes": [ 5941 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5941 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze24746", "outputIndexes": [ 5942 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5942 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24747", "outputIndexes": [ 5943 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5943, 5942 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24748", "outputIndexes": [ 5944 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5934, 5944, 5937 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_950_output_0", "outputIndexes": [ 5945 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5945, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_189_output_0", "outputIndexes": [ 5946 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5946, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_2023_output_0", "outputIndexes": [ 5947 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5932, 5947, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_253_output_0", "outputIndexes": [ 5948 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5948 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_126_output_0", "outputIndexes": [ 5949 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5946, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_2022_output_0", "outputIndexes": [ 5950 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5932, 20, 5950, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_252_output_0", "outputIndexes": [ 5951 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5949, 5951 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_822_output_0", "outputIndexes": [ 5952 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5952, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_505_output_0", "outputIndexes": [ 5953 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5933, 5953 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_315_output_0", "outputIndexes": [ 5954 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/k_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5907 ], "outputIndexes": [ 7771 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/k_proj/Linear/pre_convert", "inputIndexes": [ 7771 ], "outputIndexes": [ 7772 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/k_proj/Linear", "inputIndexes": [ 7772 ], "outputIndexes": [ 7773 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 17307828732, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/k_proj/Linear/post_convert", "inputIndexes": [ 7773 ], "outputIndexes": [ 7774 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/k_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7774 ], "outputIndexes": [ 5955 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5919, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_2018_output_0", "outputIndexes": [ 5956 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5927, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_2019_output_0", "outputIndexes": [ 5957 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5956, 5957, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_820_output_0", "outputIndexes": [ 5958 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5955, 5958 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_569_output_0", "outputIndexes": [ 5959 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5959 ], "main_type": "CastParam", "main": { "srcT": "DT_VARIANT", "dstT": "DT_FLOAT" }, "name": "/k_norm_63/Cast_output_0", "outputIndexes": [ 5960 ], "type": "Cast", "defaultDimentionFormat": "NHWC" }, { "name": "/k_norm_63/Mul_1_output_0", "inputIndexes": [ 5960 ], "outputIndexes": [ 5961 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 17310777874, 512, 512 ] }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5961, 60 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_506_output_0", "outputIndexes": [ 5962 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5961 ], "main_type": "NONE", "name": "/Shape_885_output_0", "outputIndexes": [ 5963 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5963 ], "main_type": "NONE", "name": "Shape24663", "outputIndexes": [ 5964 ], "type": "Shape", "defaultDimentionFormat": "NCHW" }, { "inputIndexes": [ 5963 ], "main_type": "NONE", "name": "Rank24665", "outputIndexes": [ 5965 ], "type": "Rank", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5965, 5965 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24667", "outputIndexes": [ 5966 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5966 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze24668", "outputIndexes": [ 5967 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5966, 12 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24670", "outputIndexes": [ 5968 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5968 ], "main_type": "SqueezeParam", "main": { "squeezeDims": [ 0 ] }, "name": "Unsqueeze24671", "outputIndexes": [ 5969 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5964, 5967, 5969, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_FLOAT", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 0 }, "name": "StridedSlice24674", "outputIndexes": [ 5970 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5970 ], "main_type": "SqueezeParam", "main": {}, "name": "Squeeze24675", "outputIndexes": [ 5971 ], "type": "Squeeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 63, 5971 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24676", "outputIndexes": [ 5972 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5972, 5971 ], "main_type": "BinaryOp", "main": { "opType": 19, "T": "DT_FLOAT", "activationType": 0 }, "name": "BinaryOp24677", "outputIndexes": [ 5973 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5963, 5973, 5966 ], "main_type": "GatherV2", "main": { "Taxis": "DT_INVALID", "Tindices": "DT_INVALID", "Tparams": "DT_INVALID" }, "name": "/Gather_951_output_0", "outputIndexes": [ 5974 ], "type": "GatherV2", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5974, 75 ], "main_type": "BinaryOp", "main": { "opType": 7, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Div_190_output_0", "outputIndexes": [ 5975 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5975, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_2025_output_0", "outputIndexes": [ 5976 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5961, 5976, 78, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_255_output_0", "outputIndexes": [ 5977 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5977 ], "main_type": "UnaryOp", "main": { "opType": "NEG", "T": "DT_FLOAT" }, "name": "/Neg_127_output_0", "outputIndexes": [ 5978 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5975, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_2024_output_0", "outputIndexes": [ 5979 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5961, 20, 5979, 79, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/Slice_254_output_0", "outputIndexes": [ 5980 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5978, 5980 ], "main_type": "Axis", "main": { "axis": -1 }, "name": "/Concat_823_output_0", "outputIndexes": [ 5981 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5981, 91 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Mul_507_output_0", "outputIndexes": [ 5982 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5962, 5982 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_316_output_0", "outputIndexes": [ 5983 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/v_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5907 ], "outputIndexes": [ 7775 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/v_proj/Linear/pre_convert", "inputIndexes": [ 7775 ], "outputIndexes": [ 7776 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/v_proj/Linear", "inputIndexes": [ 7776 ], "outputIndexes": [ 7777 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 1024, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 40960, "weightSize": 0 }, "external": [ 17310778898, 2621462, 327680, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/v_proj/Linear/post_convert", "inputIndexes": [ 7777 ], "outputIndexes": [ 7778 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/v_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7778 ], "outputIndexes": [ 5984 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 1024 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5919, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_2020_output_0", "outputIndexes": [ 5985 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5927, 20 ], "main_type": "SqueezeParam", "main": {}, "name": "/Unsqueeze_2021_output_0", "outputIndexes": [ 5986 ], "type": "Unsqueeze", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5985, 5986, 97, 31 ], "main_type": "Axis", "main": { "axis": 0 }, "name": "/Concat_821_output_0", "outputIndexes": [ 5987 ], "type": "Concat", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5984, 5987 ], "main_type": "Reshape", "main": { "dimType": "NCHW" }, "name": "/Reshape_570_output_0", "outputIndexes": [ 5988 ], "type": "Reshape", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5954, 5983, 5988, 129 ], "main_type": "AttentionParam", "main": { "kv_cache": true }, "name": "Attention/Reshape_575_output_0", "outputIndexes": [ 5989 ], "type": "Attention", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/o_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5989 ], "outputIndexes": [ 7779 ], "main_type": "Reshape", "main": { "dims": [ -1, 8192, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/o_proj/Linear/pre_convert", "inputIndexes": [ 7779 ], "outputIndexes": [ 7780 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/o_proj/Linear", "inputIndexes": [ 7780 ], "outputIndexes": [ 7781 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 8192, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 327680, "weightSize": 0 }, "external": [ 17313728040, 20971542, 2621440, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/o_proj/Linear/post_convert", "inputIndexes": [ 7781 ], "outputIndexes": [ 7782 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/self_attn/o_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7782 ], "outputIndexes": [ 5990 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5906, 5990 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_318_output_0", "outputIndexes": [ 5991 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/post_attention_layernorm_63/Mul_1_output_0", "inputIndexes": [ 5991 ], "outputIndexes": [ 5992 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 17337321022, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/mlp/gate_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5992 ], "outputIndexes": [ 7783 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/mlp/gate_proj/Linear/pre_convert", "inputIndexes": [ 7783 ], "outputIndexes": [ 7784 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/mlp/gate_proj/Linear", "inputIndexes": [ 7784 ], "outputIndexes": [ 7785 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 17337361982, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/mlp/gate_proj/Linear/post_convert", "inputIndexes": [ 7785 ], "outputIndexes": [ 7786 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/mlp/gate_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7786 ], "outputIndexes": [ 5993 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5993 ], "main_type": "UnaryOp", "main": { "opType": "SILU", "T": "DT_INVALID" }, "name": "/mlp/act_fn_63/Mul_output_0", "outputIndexes": [ 5994 ], "type": "UnaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/mlp/up_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5992 ], "outputIndexes": [ 7787 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/mlp/up_proj/Linear/pre_convert", "inputIndexes": [ 7787 ], "outputIndexes": [ 7788 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/mlp/up_proj/Linear", "inputIndexes": [ 7788 ], "outputIndexes": [ 7789 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 25600, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 17411090004, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/mlp/up_proj/Linear/post_convert", "inputIndexes": [ 7789 ], "outputIndexes": [ 7790 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/mlp/up_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7790 ], "outputIndexes": [ 5995 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 25600 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5994, 5995 ], "main_type": "BinaryOp", "main": { "opType": 2, "T": "DT_FLOAT", "activationType": 0 }, "name": "/mlp_63/Mul_output_0", "outputIndexes": [ 5996 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/mlp/down_proj/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 5996 ], "outputIndexes": [ 7791 ], "main_type": "Reshape", "main": { "dims": [ -1, 25600, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/mlp/down_proj/Linear/pre_convert", "inputIndexes": [ 7791 ], "outputIndexes": [ 7792 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/mlp/down_proj/Linear", "inputIndexes": [ 7792 ], "outputIndexes": [ 7793 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 5120, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 25600, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": false, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 1024000, "weightSize": 0 }, "external": [ 17484818026, 65536022, 8192000, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/mlp/down_proj/Linear/post_convert", "inputIndexes": [ 7793 ], "outputIndexes": [ 7794 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/layers.63/mlp/down_proj/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7794 ], "outputIndexes": [ 5997 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 5120 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5991, 5997 ], "main_type": "BinaryOp", "main": { "opType": 0, "T": "DT_FLOAT", "activationType": 0 }, "name": "/Add_319_output_0", "outputIndexes": [ 5998 ], "type": "BinaryOp", "defaultDimentionFormat": "NHWC" }, { "inputIndexes": [ 5998, 5999, 78, 15, 15 ], "main_type": "StridedSliceParam", "main": { "Index": "DT_INVALID", "T": "DT_INVALID", "beginMask": 0, "endMask": 0, "ellipsisMask": 0, "newAxisMask": 0, "shrinkAxisMask": 0, "fromType": 1 }, "name": "/lm/Slice_output_0", "outputIndexes": [ 6000 ], "type": "StridedSlice", "defaultDimentionFormat": "NHWC" }, { "name": "/lm/norm/Mul_1_output_0", "inputIndexes": [ 6000 ], "outputIndexes": [ 6001 ], "type": "LayerNorm", "main_type": "LayerNorm", "main": { "axis": [ -1 ], "epsilon": 1e-06, "group": 1, "useRMSNorm": true, "external": [ 17558546048, 20480, 20480 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/lm/lm_head/Linear/pre_reshape", "type": "Reshape", "inputIndexes": [ 6001 ], "outputIndexes": [ 7795 ], "main_type": "Reshape", "main": { "dims": [ -1, 5120, 1, 1 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/lm/lm_head/Linear/pre_convert", "inputIndexes": [ 7795 ], "outputIndexes": [ 7796 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NCHW", "dest": "NC4HW4" }, "defaultDimentionFormat": "NHWC" }, { "name": "/lm/lm_head/Linear", "inputIndexes": [ 7796 ], "outputIndexes": [ 7797 ], "type": "Convolution", "main_type": "Convolution2D", "main": { "common": { "dilateX": 1, "dilateY": 1, "strideX": 1, "strideY": 1, "kernelX": 1, "kernelY": 1, "padX": 0, "padY": 0, "group": 1, "outputCount": 151936, "relu": false, "padMode": "CAFFE", "relu6": false, "inputCount": 5120, "hasOutputShape": false }, "quanParameter": { "quantScale": 1.0, "scaleIn": 0.0, "scaleOut": 0.0, "useInt32": false, "has_scaleInt": false, "shapeInt32": true, "type": 1, "aMaxOrBits": 4, "aMin": 1, "readType": 6077440, "weightSize": 0 }, "external": [ 17558587008, 388956186, 48619520, 0, 0 ] }, "defaultDimentionFormat": "NHWC" }, { "name": "/lm/lm_head/Linear/post_convert", "inputIndexes": [ 7797 ], "outputIndexes": [ 7798 ], "type": "ConvertTensor", "main_type": "TensorConvertInfo", "main": { "source": "NC4HW4", "dest": "NCHW" }, "defaultDimentionFormat": "NHWC" }, { "name": "/lm/lm_head/Linear/post_reshape", "type": "Reshape", "inputIndexes": [ 7798 ], "outputIndexes": [ 6002 ], "main_type": "Reshape", "main": { "dims": [ 1, -1, 151936 ], "dimType": "NCHW" }, "defaultDimentionFormat": "NHWC" } ], "outputName": [ "logits", "presents" ], "preferForwardType": "CPU", "sourceType": "ONNX", "tensorName": [ "past_key_values", "presents", "input_ids", "/Constant_output_0", "/Reshape_output_0", "/input_layernorm/Mul_1_output_0", "/FakeLinear_output_0", "/Shape_output_0", "Shape52", "BinaryOp55", "BinaryOp56", "Unsqueeze57", "Const9", "BinaryOp59", "Unsqueeze60", "Unsqueeze62", "StridedSlice63", "BinaryOp65", "BinaryOp66", "/Gather_1_output_0", "/rotary/Constant_4_output_0", "/Unsqueeze_output_0", "Unsqueeze76", "Unsqueeze79", "StridedSlice82", "Squeeze83", "BinaryOp84", "BinaryOp85", "/Gather_2_output_0", "/Unsqueeze_1_output_0", "/Constant_3_output_0", "/Constant_4_output_0", "/Concat_output_0", "/Reshape_1_output_0", "/q_norm/Cast_output_0", "/q_norm/Mul_1_output_0", "position_ids", "/rotary/Cast_output_0", "/rotary/Constant_1_output_0", "/rotary/Reshape_output_0", "/rotary/Constant_2_output_0", "/rotary/Mul_output_0", "/rotary/Cos_output_0", "/rotary/Unsqueeze_output_0", "/rotary/Sin_output_0", "/rotary/Unsqueeze_1_output_0", "/rotary/Concat_output_0", "/rotary/Concat_1_output_0", "/rotary/Constant_7_output_0", "/rotary/Unsqueeze_2_output_0", "/rotary/Unsqueeze_3_output_0", "Shape116", "BinaryOp119", "BinaryOp120", "Unsqueeze121", "BinaryOp123", "Unsqueeze124", "StridedSlice127", "BinaryOp129", "BinaryOp130", "/Gather_3_output_0", "/Mul_output_0", "/Shape_2_output_0", "/Constant_10_output_0", "Shape258", "BinaryOp261", "BinaryOp262", "Unsqueeze263", "BinaryOp265", "Unsqueeze266", "StridedSlice269", "Squeeze270", "BinaryOp271", "BinaryOp272", "/Gather_5_output_0", "/rotary/Constant_6_output_0", "/Div_output_0", "/Unsqueeze_7_output_0", "/Constant_19_output_0", "/Constant_13_output_0", "/Slice_1_output_0", "/Neg_output_0", "/Unsqueeze_6_output_0", "/Slice_output_0", "/Concat_3_output_0", "Unsqueeze167", "Unsqueeze170", "StridedSlice173", "Squeeze174", "BinaryOp175", "BinaryOp176", "/Gather_4_output_0", "/Mul_1_output_0", "/Add_output_0", "/FakeLinear_1_output_0", "/Unsqueeze_2_output_0", "/Unsqueeze_3_output_0", "/Constant_6_output_0", "/Concat_1_output_0", "/Reshape_2_output_0", "/k_norm/Cast_output_0", "/k_norm/Mul_1_output_0", "/Mul_2_output_0", "/Shape_3_output_0", "Shape135", "BinaryOp138", "BinaryOp139", "Unsqueeze140", "BinaryOp142", "Unsqueeze143", "StridedSlice146", "Squeeze147", "BinaryOp148", "BinaryOp149", "/Gather_6_output_0", "/Div_1_output_0", "/Unsqueeze_9_output_0", "/Slice_3_output_0", "/Neg_1_output_0", "/Unsqueeze_8_output_0", "/Slice_2_output_0", "/Concat_4_output_0", "/Mul_3_output_0", "/Add_1_output_0", "/FakeLinear_2_output_0", "/Unsqueeze_4_output_0", "/Unsqueeze_5_output_0", "/Concat_2_output_0", "/Reshape_3_output_0", "attention_mask", "/Reshape_8_output_0", "/FakeLinear_3_output_0", "/Add_3_output_0", "/post_attention_layernorm/Mul_1_output_0", "/mlp/gate_proj/FakeLinear_output_0", "/mlp/act_fn/Mul_output_0", "/mlp/up_proj/FakeLinear_output_0", "/mlp/Mul_output_0", "/mlp/down_proj/FakeLinear_output_0", "/Add_4_output_0", "/Reshape_9_output_0", "/input_layernorm_1/Mul_1_output_0", "/FakeLinear_4_output_0", "/Shape_14_output_0", "Shape502", "BinaryOp505", "BinaryOp506", "Unsqueeze507", "BinaryOp509", "Unsqueeze510", "StridedSlice513", "BinaryOp515", "BinaryOp516", "/Gather_18_output_0", "/Unsqueeze_32_output_0", "Unsqueeze524", "Unsqueeze527", "StridedSlice530", "Squeeze531", "BinaryOp532", "BinaryOp533", "/Gather_19_output_0", "/Unsqueeze_33_output_0", "/Concat_13_output_0", "/Reshape_10_output_0", "/q_norm_1/Cast_output_0", "/q_norm_1/Mul_1_output_0", "/Mul_8_output_0", "/Shape_16_output_0", "Shape651", "BinaryOp654", "BinaryOp655", "Unsqueeze656", "BinaryOp658", "Unsqueeze659", "StridedSlice662", "Squeeze663", "BinaryOp664", "BinaryOp665", "/Gather_20_output_0", "/Div_3_output_0", "/Unsqueeze_39_output_0", "/Slice_5_output_0", "/Neg_2_output_0", "/Unsqueeze_38_output_0", "/Slice_4_output_0", "/Concat_16_output_0", "/Mul_9_output_0", "/Add_5_output_0", "/FakeLinear_5_output_0", "/Unsqueeze_34_output_0", "/Unsqueeze_35_output_0", "/Concat_14_output_0", "/Reshape_11_output_0", "/k_norm_1/Cast_output_0", "/k_norm_1/Mul_1_output_0", "/Mul_10_output_0", "/Shape_17_output_0", "Shape548", "BinaryOp551", "BinaryOp552", "Unsqueeze553", "BinaryOp555", "Unsqueeze556", "StridedSlice559", "Squeeze560", "BinaryOp561", "BinaryOp562", "/Gather_21_output_0", "/Div_4_output_0", "/Unsqueeze_41_output_0", "/Slice_7_output_0", "/Neg_3_output_0", "/Unsqueeze_40_output_0", "/Slice_6_output_0", "/Concat_17_output_0", "/Mul_11_output_0", "/Add_6_output_0", "/FakeLinear_6_output_0", "/Unsqueeze_36_output_0", "/Unsqueeze_37_output_0", "/Concat_15_output_0", "/Reshape_12_output_0", "/Reshape_17_output_0", "/FakeLinear_7_output_0", "/Add_8_output_0", "/post_attention_layernorm_1/Mul_1_output_0", "/mlp/gate_proj_1/FakeLinear_output_0", "/mlp/act_fn_1/Mul_output_0", "/mlp/up_proj_1/FakeLinear_output_0", "/mlp_1/Mul_output_0", "/mlp/down_proj_1/FakeLinear_output_0", "/Add_9_output_0", "/Reshape_18_output_0", "/input_layernorm_2/Mul_1_output_0", "/FakeLinear_8_output_0", "/Shape_28_output_0", "Shape890", "BinaryOp893", "BinaryOp894", "Unsqueeze895", "BinaryOp897", "Unsqueeze898", "StridedSlice901", "BinaryOp903", "BinaryOp904", "/Gather_33_output_0", "/Unsqueeze_64_output_0", "Unsqueeze912", "Unsqueeze915", "StridedSlice918", "Squeeze919", "BinaryOp920", "BinaryOp921", "/Gather_34_output_0", "/Unsqueeze_65_output_0", "/Concat_26_output_0", "/Reshape_19_output_0", "/q_norm_2/Cast_output_0", "/q_norm_2/Mul_1_output_0", "/Mul_16_output_0", "/Shape_30_output_0", "Shape1039", "BinaryOp1042", "BinaryOp1043", "Unsqueeze1044", "BinaryOp1046", "Unsqueeze1047", "StridedSlice1050", "Squeeze1051", "BinaryOp1052", "BinaryOp1053", "/Gather_35_output_0", "/Div_6_output_0", "/Unsqueeze_71_output_0", "/Slice_9_output_0", "/Neg_4_output_0", "/Unsqueeze_70_output_0", "/Slice_8_output_0", "/Concat_29_output_0", "/Mul_17_output_0", "/Add_10_output_0", "/FakeLinear_9_output_0", "/Unsqueeze_66_output_0", "/Unsqueeze_67_output_0", "/Concat_27_output_0", "/Reshape_20_output_0", "/k_norm_2/Cast_output_0", "/k_norm_2/Mul_1_output_0", "/Mul_18_output_0", "/Shape_31_output_0", "Shape936", "BinaryOp939", "BinaryOp940", "Unsqueeze941", "BinaryOp943", "Unsqueeze944", "StridedSlice947", "Squeeze948", "BinaryOp949", "BinaryOp950", "/Gather_36_output_0", "/Div_7_output_0", "/Unsqueeze_73_output_0", "/Slice_11_output_0", "/Neg_5_output_0", "/Unsqueeze_72_output_0", "/Slice_10_output_0", "/Concat_30_output_0", "/Mul_19_output_0", "/Add_11_output_0", "/FakeLinear_10_output_0", "/Unsqueeze_68_output_0", "/Unsqueeze_69_output_0", "/Concat_28_output_0", "/Reshape_21_output_0", "/Reshape_26_output_0", "/FakeLinear_11_output_0", "/Add_13_output_0", "/post_attention_layernorm_2/Mul_1_output_0", "/mlp/gate_proj_2/FakeLinear_output_0", "/mlp/act_fn_2/Mul_output_0", "/mlp/up_proj_2/FakeLinear_output_0", "/mlp_2/Mul_output_0", "/mlp/down_proj_2/FakeLinear_output_0", "/Add_14_output_0", "/Reshape_27_output_0", "/input_layernorm_3/Mul_1_output_0", "/FakeLinear_12_output_0", "/Shape_42_output_0", "Shape1278", "BinaryOp1281", "BinaryOp1282", "Unsqueeze1283", "BinaryOp1285", "Unsqueeze1286", "StridedSlice1289", "BinaryOp1291", "BinaryOp1292", "/Gather_48_output_0", "/Unsqueeze_96_output_0", "Unsqueeze1300", "Unsqueeze1303", "StridedSlice1306", "Squeeze1307", "BinaryOp1308", "BinaryOp1309", "/Gather_49_output_0", "/Unsqueeze_97_output_0", "/Concat_39_output_0", "/Reshape_28_output_0", "/q_norm_3/Cast_output_0", "/q_norm_3/Mul_1_output_0", "/Mul_24_output_0", "/Shape_44_output_0", "Shape1428", "BinaryOp1431", "BinaryOp1432", "Unsqueeze1433", "BinaryOp1435", "Unsqueeze1436", "StridedSlice1439", "Squeeze1440", "BinaryOp1441", "BinaryOp1442", "/Gather_50_output_0", "/Div_9_output_0", "/Unsqueeze_103_output_0", "/Slice_13_output_0", "/Neg_6_output_0", "/Unsqueeze_102_output_0", "/Slice_12_output_0", "/Concat_42_output_0", "/Mul_25_output_0", "/Add_15_output_0", "/FakeLinear_13_output_0", "/Unsqueeze_98_output_0", "/Unsqueeze_99_output_0", "/Concat_40_output_0", "/Reshape_29_output_0", "/k_norm_3/Cast_output_0", "/k_norm_3/Mul_1_output_0", "/Mul_26_output_0", "/Shape_45_output_0", "Shape1324", "BinaryOp1327", "BinaryOp1328", "Unsqueeze1329", "BinaryOp1331", "Unsqueeze1332", "StridedSlice1335", "Squeeze1336", "BinaryOp1337", "BinaryOp1338", "/Gather_51_output_0", "/Div_10_output_0", "/Unsqueeze_105_output_0", "/Slice_15_output_0", "/Neg_7_output_0", "/Unsqueeze_104_output_0", "/Slice_14_output_0", "/Concat_43_output_0", "/Mul_27_output_0", "/Add_16_output_0", "/FakeLinear_14_output_0", "/Unsqueeze_100_output_0", "/Unsqueeze_101_output_0", "/Concat_41_output_0", "/Reshape_30_output_0", "/Reshape_35_output_0", "/FakeLinear_15_output_0", "/Add_18_output_0", "/post_attention_layernorm_3/Mul_1_output_0", "/mlp/gate_proj_3/FakeLinear_output_0", "/mlp/act_fn_3/Mul_output_0", "/mlp/up_proj_3/FakeLinear_output_0", "/mlp_3/Mul_output_0", "/mlp/down_proj_3/FakeLinear_output_0", "/Add_19_output_0", "/Reshape_36_output_0", "/input_layernorm_4/Mul_1_output_0", "/FakeLinear_16_output_0", "/Shape_56_output_0", "Shape1667", "BinaryOp1670", "BinaryOp1671", "Unsqueeze1672", "BinaryOp1674", "Unsqueeze1675", "StridedSlice1678", "BinaryOp1680", "BinaryOp1681", "/Gather_63_output_0", "/Unsqueeze_128_output_0", "Unsqueeze1689", "Unsqueeze1692", "StridedSlice1695", "Squeeze1696", "BinaryOp1697", "BinaryOp1698", "/Gather_64_output_0", "/Unsqueeze_129_output_0", "/Concat_52_output_0", "/Reshape_37_output_0", "/q_norm_4/Cast_output_0", "/q_norm_4/Mul_1_output_0", "/Mul_32_output_0", "/Shape_58_output_0", "Shape1817", "BinaryOp1820", "BinaryOp1821", "Unsqueeze1822", "BinaryOp1824", "Unsqueeze1825", "StridedSlice1828", "Squeeze1829", "BinaryOp1830", "BinaryOp1831", "/Gather_65_output_0", "/Div_12_output_0", "/Unsqueeze_135_output_0", "/Slice_17_output_0", "/Neg_8_output_0", "/Unsqueeze_134_output_0", "/Slice_16_output_0", "/Concat_55_output_0", "/Mul_33_output_0", "/Add_20_output_0", "/FakeLinear_17_output_0", "/Unsqueeze_130_output_0", "/Unsqueeze_131_output_0", "/Concat_53_output_0", "/Reshape_38_output_0", "/k_norm_4/Cast_output_0", "/k_norm_4/Mul_1_output_0", "/Mul_34_output_0", "/Shape_59_output_0", "Shape1713", "BinaryOp1716", "BinaryOp1717", "Unsqueeze1718", "BinaryOp1720", "Unsqueeze1721", "StridedSlice1724", "Squeeze1725", "BinaryOp1726", "BinaryOp1727", "/Gather_66_output_0", "/Div_13_output_0", "/Unsqueeze_137_output_0", "/Slice_19_output_0", "/Neg_9_output_0", "/Unsqueeze_136_output_0", "/Slice_18_output_0", "/Concat_56_output_0", "/Mul_35_output_0", "/Add_21_output_0", "/FakeLinear_18_output_0", "/Unsqueeze_132_output_0", "/Unsqueeze_133_output_0", "/Concat_54_output_0", "/Reshape_39_output_0", "/Reshape_44_output_0", "/FakeLinear_19_output_0", "/Add_23_output_0", "/post_attention_layernorm_4/Mul_1_output_0", "/mlp/gate_proj_4/FakeLinear_output_0", "/mlp/act_fn_4/Mul_output_0", "/mlp/up_proj_4/FakeLinear_output_0", "/mlp_4/Mul_output_0", "/mlp/down_proj_4/FakeLinear_output_0", "/Add_24_output_0", "/Reshape_45_output_0", "/input_layernorm_5/Mul_1_output_0", "/FakeLinear_20_output_0", "/Shape_70_output_0", "Shape2056", "BinaryOp2059", "BinaryOp2060", "Unsqueeze2061", "BinaryOp2063", "Unsqueeze2064", "StridedSlice2067", "BinaryOp2069", "BinaryOp2070", "/Gather_78_output_0", "/Unsqueeze_160_output_0", "Unsqueeze2078", "Unsqueeze2081", "StridedSlice2084", "Squeeze2085", "BinaryOp2086", "BinaryOp2087", "/Gather_79_output_0", "/Unsqueeze_161_output_0", "/Concat_65_output_0", "/Reshape_46_output_0", "/q_norm_5/Cast_output_0", "/q_norm_5/Mul_1_output_0", "/Mul_40_output_0", "/Shape_72_output_0", "Shape2206", "BinaryOp2209", "BinaryOp2210", "Unsqueeze2211", "BinaryOp2213", "Unsqueeze2214", "StridedSlice2217", "Squeeze2218", "BinaryOp2219", "BinaryOp2220", "/Gather_80_output_0", "/Div_15_output_0", "/Unsqueeze_167_output_0", "/Slice_21_output_0", "/Neg_10_output_0", "/Unsqueeze_166_output_0", "/Slice_20_output_0", "/Concat_68_output_0", "/Mul_41_output_0", "/Add_25_output_0", "/FakeLinear_21_output_0", "/Unsqueeze_162_output_0", "/Unsqueeze_163_output_0", "/Concat_66_output_0", "/Reshape_47_output_0", "/k_norm_5/Cast_output_0", "/k_norm_5/Mul_1_output_0", "/Mul_42_output_0", "/Shape_73_output_0", "Shape2102", "BinaryOp2105", "BinaryOp2106", "Unsqueeze2107", "BinaryOp2109", "Unsqueeze2110", "StridedSlice2113", "Squeeze2114", "BinaryOp2115", "BinaryOp2116", "/Gather_81_output_0", "/Div_16_output_0", "/Unsqueeze_169_output_0", "/Slice_23_output_0", "/Neg_11_output_0", "/Unsqueeze_168_output_0", "/Slice_22_output_0", "/Concat_69_output_0", "/Mul_43_output_0", "/Add_26_output_0", "/FakeLinear_22_output_0", "/Unsqueeze_164_output_0", "/Unsqueeze_165_output_0", "/Concat_67_output_0", "/Reshape_48_output_0", "/Reshape_53_output_0", "/FakeLinear_23_output_0", "/Add_28_output_0", "/post_attention_layernorm_5/Mul_1_output_0", "/mlp/gate_proj_5/FakeLinear_output_0", "/mlp/act_fn_5/Mul_output_0", "/mlp/up_proj_5/FakeLinear_output_0", "/mlp_5/Mul_output_0", "/mlp/down_proj_5/FakeLinear_output_0", "/Add_29_output_0", "/Reshape_54_output_0", "/input_layernorm_6/Mul_1_output_0", "/FakeLinear_24_output_0", "/Shape_84_output_0", "Shape2445", "BinaryOp2448", "BinaryOp2449", "Unsqueeze2450", "BinaryOp2452", "Unsqueeze2453", "StridedSlice2456", "BinaryOp2458", "BinaryOp2459", "/Gather_93_output_0", "/Unsqueeze_192_output_0", "Unsqueeze2467", "Unsqueeze2470", "StridedSlice2473", "Squeeze2474", "BinaryOp2475", "BinaryOp2476", "/Gather_94_output_0", "/Unsqueeze_193_output_0", "/Concat_78_output_0", "/Reshape_55_output_0", "/q_norm_6/Cast_output_0", "/q_norm_6/Mul_1_output_0", "/Mul_48_output_0", "/Shape_86_output_0", "Shape2595", "BinaryOp2598", "BinaryOp2599", "Unsqueeze2600", "BinaryOp2602", "Unsqueeze2603", "StridedSlice2606", "Squeeze2607", "BinaryOp2608", "BinaryOp2609", "/Gather_95_output_0", "/Div_18_output_0", "/Unsqueeze_199_output_0", "/Slice_25_output_0", "/Neg_12_output_0", "/Unsqueeze_198_output_0", "/Slice_24_output_0", "/Concat_81_output_0", "/Mul_49_output_0", "/Add_30_output_0", "/FakeLinear_25_output_0", "/Unsqueeze_194_output_0", "/Unsqueeze_195_output_0", "/Concat_79_output_0", "/Reshape_56_output_0", "/k_norm_6/Cast_output_0", "/k_norm_6/Mul_1_output_0", "/Mul_50_output_0", "/Shape_87_output_0", "Shape2491", "BinaryOp2494", "BinaryOp2495", "Unsqueeze2496", "BinaryOp2498", "Unsqueeze2499", "StridedSlice2502", "Squeeze2503", "BinaryOp2504", "BinaryOp2505", "/Gather_96_output_0", "/Div_19_output_0", "/Unsqueeze_201_output_0", "/Slice_27_output_0", "/Neg_13_output_0", "/Unsqueeze_200_output_0", "/Slice_26_output_0", "/Concat_82_output_0", "/Mul_51_output_0", "/Add_31_output_0", "/FakeLinear_26_output_0", "/Unsqueeze_196_output_0", "/Unsqueeze_197_output_0", "/Concat_80_output_0", "/Reshape_57_output_0", "/Reshape_62_output_0", "/FakeLinear_27_output_0", "/Add_33_output_0", "/post_attention_layernorm_6/Mul_1_output_0", "/mlp/gate_proj_6/FakeLinear_output_0", "/mlp/act_fn_6/Mul_output_0", "/mlp/up_proj_6/FakeLinear_output_0", "/mlp_6/Mul_output_0", "/mlp/down_proj_6/FakeLinear_output_0", "/Add_34_output_0", "/Reshape_63_output_0", "/input_layernorm_7/Mul_1_output_0", "/FakeLinear_28_output_0", "/Shape_98_output_0", "Shape2834", "BinaryOp2837", "BinaryOp2838", "Unsqueeze2839", "BinaryOp2841", "Unsqueeze2842", "StridedSlice2845", "BinaryOp2847", "BinaryOp2848", "/Gather_108_output_0", "/Unsqueeze_224_output_0", "Unsqueeze2856", "Unsqueeze2859", "StridedSlice2862", "Squeeze2863", "BinaryOp2864", "BinaryOp2865", "/Gather_109_output_0", "/Unsqueeze_225_output_0", "/Concat_91_output_0", "/Reshape_64_output_0", "/q_norm_7/Cast_output_0", "/q_norm_7/Mul_1_output_0", "/Mul_56_output_0", "/Shape_100_output_0", "Shape2983", "BinaryOp2986", "BinaryOp2987", "Unsqueeze2988", "BinaryOp2990", "Unsqueeze2991", "StridedSlice2994", "Squeeze2995", "BinaryOp2996", "BinaryOp2997", "/Gather_110_output_0", "/Div_21_output_0", "/Unsqueeze_231_output_0", "/Slice_29_output_0", "/Neg_14_output_0", "/Unsqueeze_230_output_0", "/Slice_28_output_0", "/Concat_94_output_0", "/Mul_57_output_0", "/Add_35_output_0", "/FakeLinear_29_output_0", "/Unsqueeze_226_output_0", "/Unsqueeze_227_output_0", "/Concat_92_output_0", "/Reshape_65_output_0", "/k_norm_7/Cast_output_0", "/k_norm_7/Mul_1_output_0", "/Mul_58_output_0", "/Shape_101_output_0", "Shape2880", "BinaryOp2883", "BinaryOp2884", "Unsqueeze2885", "BinaryOp2887", "Unsqueeze2888", "StridedSlice2891", "Squeeze2892", "BinaryOp2893", "BinaryOp2894", "/Gather_111_output_0", "/Div_22_output_0", "/Unsqueeze_233_output_0", "/Slice_31_output_0", "/Neg_15_output_0", "/Unsqueeze_232_output_0", "/Slice_30_output_0", "/Concat_95_output_0", "/Mul_59_output_0", "/Add_36_output_0", "/FakeLinear_30_output_0", "/Unsqueeze_228_output_0", "/Unsqueeze_229_output_0", "/Concat_93_output_0", "/Reshape_66_output_0", "/Reshape_71_output_0", "/FakeLinear_31_output_0", "/Add_38_output_0", "/post_attention_layernorm_7/Mul_1_output_0", "/mlp/gate_proj_7/FakeLinear_output_0", "/mlp/act_fn_7/Mul_output_0", "/mlp/up_proj_7/FakeLinear_output_0", "/mlp_7/Mul_output_0", "/mlp/down_proj_7/FakeLinear_output_0", "/Add_39_output_0", "/Reshape_72_output_0", "/input_layernorm_8/Mul_1_output_0", "/FakeLinear_32_output_0", "/Shape_112_output_0", "Shape3222", "BinaryOp3225", "BinaryOp3226", "Unsqueeze3227", "BinaryOp3229", "Unsqueeze3230", "StridedSlice3233", "BinaryOp3235", "BinaryOp3236", "/Gather_123_output_0", "/Unsqueeze_256_output_0", "Unsqueeze3244", "Unsqueeze3247", "StridedSlice3250", "Squeeze3251", "BinaryOp3252", "BinaryOp3253", "/Gather_124_output_0", "/Unsqueeze_257_output_0", "/Concat_104_output_0", "/Reshape_73_output_0", "/q_norm_8/Cast_output_0", "/q_norm_8/Mul_1_output_0", "/Mul_64_output_0", "/Shape_114_output_0", "Shape3372", "BinaryOp3375", "BinaryOp3376", "Unsqueeze3377", "BinaryOp3379", "Unsqueeze3380", "StridedSlice3383", "Squeeze3384", "BinaryOp3385", "BinaryOp3386", "/Gather_125_output_0", "/Div_24_output_0", "/Unsqueeze_263_output_0", "/Slice_33_output_0", "/Neg_16_output_0", "/Unsqueeze_262_output_0", "/Slice_32_output_0", "/Concat_107_output_0", "/Mul_65_output_0", "/Add_40_output_0", "/FakeLinear_33_output_0", "/Unsqueeze_258_output_0", "/Unsqueeze_259_output_0", "/Concat_105_output_0", "/Reshape_74_output_0", "/k_norm_8/Cast_output_0", "/k_norm_8/Mul_1_output_0", "/Mul_66_output_0", "/Shape_115_output_0", "Shape3268", "BinaryOp3271", "BinaryOp3272", "Unsqueeze3273", "BinaryOp3275", "Unsqueeze3276", "StridedSlice3279", "Squeeze3280", "BinaryOp3281", "BinaryOp3282", "/Gather_126_output_0", "/Div_25_output_0", "/Unsqueeze_265_output_0", "/Slice_35_output_0", "/Neg_17_output_0", "/Unsqueeze_264_output_0", "/Slice_34_output_0", "/Concat_108_output_0", "/Mul_67_output_0", "/Add_41_output_0", "/FakeLinear_34_output_0", "/Unsqueeze_260_output_0", "/Unsqueeze_261_output_0", "/Concat_106_output_0", "/Reshape_75_output_0", "/Reshape_80_output_0", "/FakeLinear_35_output_0", "/Add_43_output_0", "/post_attention_layernorm_8/Mul_1_output_0", "/mlp/gate_proj_8/FakeLinear_output_0", "/mlp/act_fn_8/Mul_output_0", "/mlp/up_proj_8/FakeLinear_output_0", "/mlp_8/Mul_output_0", "/mlp/down_proj_8/FakeLinear_output_0", "/Add_44_output_0", "/Reshape_81_output_0", "/input_layernorm_9/Mul_1_output_0", "/FakeLinear_36_output_0", "/Shape_126_output_0", "Shape3611", "BinaryOp3614", "BinaryOp3615", "Unsqueeze3616", "BinaryOp3618", "Unsqueeze3619", "StridedSlice3622", "BinaryOp3624", "BinaryOp3625", "/Gather_138_output_0", "/Unsqueeze_288_output_0", "Unsqueeze3633", "Unsqueeze3636", "StridedSlice3639", "Squeeze3640", "BinaryOp3641", "BinaryOp3642", "/Gather_139_output_0", "/Unsqueeze_289_output_0", "/Concat_117_output_0", "/Reshape_82_output_0", "/q_norm_9/Cast_output_0", "/q_norm_9/Mul_1_output_0", "/Mul_72_output_0", "/Shape_128_output_0", "Shape3761", "BinaryOp3764", "BinaryOp3765", "Unsqueeze3766", "BinaryOp3768", "Unsqueeze3769", "StridedSlice3772", "Squeeze3773", "BinaryOp3774", "BinaryOp3775", "/Gather_140_output_0", "/Div_27_output_0", "/Unsqueeze_295_output_0", "/Slice_37_output_0", "/Neg_18_output_0", "/Unsqueeze_294_output_0", "/Slice_36_output_0", "/Concat_120_output_0", "/Mul_73_output_0", "/Add_45_output_0", "/FakeLinear_37_output_0", "/Unsqueeze_290_output_0", "/Unsqueeze_291_output_0", "/Concat_118_output_0", "/Reshape_83_output_0", "/k_norm_9/Cast_output_0", "/k_norm_9/Mul_1_output_0", "/Mul_74_output_0", "/Shape_129_output_0", "Shape3657", "BinaryOp3660", "BinaryOp3661", "Unsqueeze3662", "BinaryOp3664", "Unsqueeze3665", "StridedSlice3668", "Squeeze3669", "BinaryOp3670", "BinaryOp3671", "/Gather_141_output_0", "/Div_28_output_0", "/Unsqueeze_297_output_0", "/Slice_39_output_0", "/Neg_19_output_0", "/Unsqueeze_296_output_0", "/Slice_38_output_0", "/Concat_121_output_0", "/Mul_75_output_0", "/Add_46_output_0", "/FakeLinear_38_output_0", "/Unsqueeze_292_output_0", "/Unsqueeze_293_output_0", "/Concat_119_output_0", "/Reshape_84_output_0", "/Reshape_89_output_0", "/FakeLinear_39_output_0", "/Add_48_output_0", "/post_attention_layernorm_9/Mul_1_output_0", "/mlp/gate_proj_9/FakeLinear_output_0", "/mlp/act_fn_9/Mul_output_0", "/mlp/up_proj_9/FakeLinear_output_0", "/mlp_9/Mul_output_0", "/mlp/down_proj_9/FakeLinear_output_0", "/Add_49_output_0", "/Reshape_90_output_0", "/input_layernorm_10/Mul_1_output_0", "/FakeLinear_40_output_0", "/Shape_140_output_0", "Shape4000", "BinaryOp4003", "BinaryOp4004", "Unsqueeze4005", "BinaryOp4007", "Unsqueeze4008", "StridedSlice4011", "BinaryOp4013", "BinaryOp4014", "/Gather_153_output_0", "/Unsqueeze_320_output_0", "Unsqueeze4022", "Unsqueeze4025", "StridedSlice4028", "Squeeze4029", "BinaryOp4030", "BinaryOp4031", "/Gather_154_output_0", "/Unsqueeze_321_output_0", "/Concat_130_output_0", "/Reshape_91_output_0", "/q_norm_10/Cast_output_0", "/q_norm_10/Mul_1_output_0", "/Mul_80_output_0", "/Shape_142_output_0", "Shape4150", "BinaryOp4153", "BinaryOp4154", "Unsqueeze4155", "BinaryOp4157", "Unsqueeze4158", "StridedSlice4161", "Squeeze4162", "BinaryOp4163", "BinaryOp4164", "/Gather_155_output_0", "/Div_30_output_0", "/Unsqueeze_327_output_0", "/Slice_41_output_0", "/Neg_20_output_0", "/Unsqueeze_326_output_0", "/Slice_40_output_0", "/Concat_133_output_0", "/Mul_81_output_0", "/Add_50_output_0", "/FakeLinear_41_output_0", "/Unsqueeze_322_output_0", "/Unsqueeze_323_output_0", "/Concat_131_output_0", "/Reshape_92_output_0", "/k_norm_10/Cast_output_0", "/k_norm_10/Mul_1_output_0", "/Mul_82_output_0", "/Shape_143_output_0", "Shape4046", "BinaryOp4049", "BinaryOp4050", "Unsqueeze4051", "BinaryOp4053", "Unsqueeze4054", "StridedSlice4057", "Squeeze4058", "BinaryOp4059", "BinaryOp4060", "/Gather_156_output_0", "/Div_31_output_0", "/Unsqueeze_329_output_0", "/Slice_43_output_0", "/Neg_21_output_0", "/Unsqueeze_328_output_0", "/Slice_42_output_0", "/Concat_134_output_0", "/Mul_83_output_0", "/Add_51_output_0", "/FakeLinear_42_output_0", "/Unsqueeze_324_output_0", "/Unsqueeze_325_output_0", "/Concat_132_output_0", "/Reshape_93_output_0", "/Reshape_98_output_0", "/FakeLinear_43_output_0", "/Add_53_output_0", "/post_attention_layernorm_10/Mul_1_output_0", "/mlp/gate_proj_10/FakeLinear_output_0", "/mlp/act_fn_10/Mul_output_0", "/mlp/up_proj_10/FakeLinear_output_0", "/mlp_10/Mul_output_0", "/mlp/down_proj_10/FakeLinear_output_0", "/Add_54_output_0", "/Reshape_99_output_0", "/input_layernorm_11/Mul_1_output_0", "/FakeLinear_44_output_0", "/Shape_154_output_0", "Shape4389", "BinaryOp4392", "BinaryOp4393", "Unsqueeze4394", "BinaryOp4396", "Unsqueeze4397", "StridedSlice4400", "BinaryOp4402", "BinaryOp4403", "/Gather_168_output_0", "/Unsqueeze_352_output_0", "Unsqueeze4411", "Unsqueeze4414", "StridedSlice4417", "Squeeze4418", "BinaryOp4419", "BinaryOp4420", "/Gather_169_output_0", "/Unsqueeze_353_output_0", "/Concat_143_output_0", "/Reshape_100_output_0", "/q_norm_11/Cast_output_0", "/q_norm_11/Mul_1_output_0", "/Mul_88_output_0", "/Shape_156_output_0", "Shape4539", "BinaryOp4542", "BinaryOp4543", "Unsqueeze4544", "BinaryOp4546", "Unsqueeze4547", "StridedSlice4550", "Squeeze4551", "BinaryOp4552", "BinaryOp4553", "/Gather_170_output_0", "/Div_33_output_0", "/Unsqueeze_359_output_0", "/Slice_45_output_0", "/Neg_22_output_0", "/Unsqueeze_358_output_0", "/Slice_44_output_0", "/Concat_146_output_0", "/Mul_89_output_0", "/Add_55_output_0", "/FakeLinear_45_output_0", "/Unsqueeze_354_output_0", "/Unsqueeze_355_output_0", "/Concat_144_output_0", "/Reshape_101_output_0", "/k_norm_11/Cast_output_0", "/k_norm_11/Mul_1_output_0", "/Mul_90_output_0", "/Shape_157_output_0", "Shape4435", "BinaryOp4438", "BinaryOp4439", "Unsqueeze4440", "BinaryOp4442", "Unsqueeze4443", "StridedSlice4446", "Squeeze4447", "BinaryOp4448", "BinaryOp4449", "/Gather_171_output_0", "/Div_34_output_0", "/Unsqueeze_361_output_0", "/Slice_47_output_0", "/Neg_23_output_0", "/Unsqueeze_360_output_0", "/Slice_46_output_0", "/Concat_147_output_0", "/Mul_91_output_0", "/Add_56_output_0", "/FakeLinear_46_output_0", "/Unsqueeze_356_output_0", "/Unsqueeze_357_output_0", "/Concat_145_output_0", "/Reshape_102_output_0", "/Reshape_107_output_0", "/FakeLinear_47_output_0", "/Add_58_output_0", "/post_attention_layernorm_11/Mul_1_output_0", "/mlp/gate_proj_11/FakeLinear_output_0", "/mlp/act_fn_11/Mul_output_0", "/mlp/up_proj_11/FakeLinear_output_0", "/mlp_11/Mul_output_0", "/mlp/down_proj_11/FakeLinear_output_0", "/Add_59_output_0", "/Reshape_108_output_0", "/input_layernorm_12/Mul_1_output_0", "/FakeLinear_48_output_0", "/Shape_168_output_0", "Shape4778", "BinaryOp4781", "BinaryOp4782", "Unsqueeze4783", "BinaryOp4785", "Unsqueeze4786", "StridedSlice4789", "BinaryOp4791", "BinaryOp4792", "/Gather_183_output_0", "/Unsqueeze_384_output_0", "Unsqueeze4800", "Unsqueeze4803", "StridedSlice4806", "Squeeze4807", "BinaryOp4808", "BinaryOp4809", "/Gather_184_output_0", "/Unsqueeze_385_output_0", "/Concat_156_output_0", "/Reshape_109_output_0", "/q_norm_12/Cast_output_0", "/q_norm_12/Mul_1_output_0", "/Mul_96_output_0", "/Shape_170_output_0", "Shape4928", "BinaryOp4931", "BinaryOp4932", "Unsqueeze4933", "BinaryOp4935", "Unsqueeze4936", "StridedSlice4939", "Squeeze4940", "BinaryOp4941", "BinaryOp4942", "/Gather_185_output_0", "/Div_36_output_0", "/Unsqueeze_391_output_0", "/Slice_49_output_0", "/Neg_24_output_0", "/Unsqueeze_390_output_0", "/Slice_48_output_0", "/Concat_159_output_0", "/Mul_97_output_0", "/Add_60_output_0", "/FakeLinear_49_output_0", "/Unsqueeze_386_output_0", "/Unsqueeze_387_output_0", "/Concat_157_output_0", "/Reshape_110_output_0", "/k_norm_12/Cast_output_0", "/k_norm_12/Mul_1_output_0", "/Mul_98_output_0", "/Shape_171_output_0", "Shape4824", "BinaryOp4827", "BinaryOp4828", "Unsqueeze4829", "BinaryOp4831", "Unsqueeze4832", "StridedSlice4835", "Squeeze4836", "BinaryOp4837", "BinaryOp4838", "/Gather_186_output_0", "/Div_37_output_0", "/Unsqueeze_393_output_0", "/Slice_51_output_0", "/Neg_25_output_0", "/Unsqueeze_392_output_0", "/Slice_50_output_0", "/Concat_160_output_0", "/Mul_99_output_0", "/Add_61_output_0", "/FakeLinear_50_output_0", "/Unsqueeze_388_output_0", "/Unsqueeze_389_output_0", "/Concat_158_output_0", "/Reshape_111_output_0", "/Reshape_116_output_0", "/FakeLinear_51_output_0", "/Add_63_output_0", "/post_attention_layernorm_12/Mul_1_output_0", "/mlp/gate_proj_12/FakeLinear_output_0", "/mlp/act_fn_12/Mul_output_0", "/mlp/up_proj_12/FakeLinear_output_0", "/mlp_12/Mul_output_0", "/mlp/down_proj_12/FakeLinear_output_0", "/Add_64_output_0", "/Reshape_117_output_0", "/input_layernorm_13/Mul_1_output_0", "/FakeLinear_52_output_0", "/Shape_182_output_0", "Shape5167", "BinaryOp5170", "BinaryOp5171", "Unsqueeze5172", "BinaryOp5174", "Unsqueeze5175", "StridedSlice5178", "BinaryOp5180", "BinaryOp5181", "/Gather_198_output_0", "/Unsqueeze_416_output_0", "Unsqueeze5189", "Unsqueeze5192", "StridedSlice5195", "Squeeze5196", "BinaryOp5197", "BinaryOp5198", "/Gather_199_output_0", "/Unsqueeze_417_output_0", "/Concat_169_output_0", "/Reshape_118_output_0", "/q_norm_13/Cast_output_0", "/q_norm_13/Mul_1_output_0", "/Mul_104_output_0", "/Shape_184_output_0", "Shape5317", "BinaryOp5320", "BinaryOp5321", "Unsqueeze5322", "BinaryOp5324", "Unsqueeze5325", "StridedSlice5328", "Squeeze5329", "BinaryOp5330", "BinaryOp5331", "/Gather_200_output_0", "/Div_39_output_0", "/Unsqueeze_423_output_0", "/Slice_53_output_0", "/Neg_26_output_0", "/Unsqueeze_422_output_0", "/Slice_52_output_0", "/Concat_172_output_0", "/Mul_105_output_0", "/Add_65_output_0", "/FakeLinear_53_output_0", "/Unsqueeze_418_output_0", "/Unsqueeze_419_output_0", "/Concat_170_output_0", "/Reshape_119_output_0", "/k_norm_13/Cast_output_0", "/k_norm_13/Mul_1_output_0", "/Mul_106_output_0", "/Shape_185_output_0", "Shape5213", "BinaryOp5216", "BinaryOp5217", "Unsqueeze5218", "BinaryOp5220", "Unsqueeze5221", "StridedSlice5224", "Squeeze5225", "BinaryOp5226", "BinaryOp5227", "/Gather_201_output_0", "/Div_40_output_0", "/Unsqueeze_425_output_0", "/Slice_55_output_0", "/Neg_27_output_0", "/Unsqueeze_424_output_0", "/Slice_54_output_0", "/Concat_173_output_0", "/Mul_107_output_0", "/Add_66_output_0", "/FakeLinear_54_output_0", "/Unsqueeze_420_output_0", "/Unsqueeze_421_output_0", "/Concat_171_output_0", "/Reshape_120_output_0", "/Reshape_125_output_0", "/FakeLinear_55_output_0", "/Add_68_output_0", "/post_attention_layernorm_13/Mul_1_output_0", "/mlp/gate_proj_13/FakeLinear_output_0", "/mlp/act_fn_13/Mul_output_0", "/mlp/up_proj_13/FakeLinear_output_0", "/mlp_13/Mul_output_0", "/mlp/down_proj_13/FakeLinear_output_0", "/Add_69_output_0", "/Reshape_126_output_0", "/input_layernorm_14/Mul_1_output_0", "/FakeLinear_56_output_0", "/Shape_196_output_0", "Shape5556", "BinaryOp5559", "BinaryOp5560", "Unsqueeze5561", "BinaryOp5563", "Unsqueeze5564", "StridedSlice5567", "BinaryOp5569", "BinaryOp5570", "/Gather_213_output_0", "/Unsqueeze_448_output_0", "Unsqueeze5578", "Unsqueeze5581", "StridedSlice5584", "Squeeze5585", "BinaryOp5586", "BinaryOp5587", "/Gather_214_output_0", "/Unsqueeze_449_output_0", "/Concat_182_output_0", "/Reshape_127_output_0", "/q_norm_14/Cast_output_0", "/q_norm_14/Mul_1_output_0", "/Mul_112_output_0", "/Shape_198_output_0", "Shape5706", "BinaryOp5709", "BinaryOp5710", "Unsqueeze5711", "BinaryOp5713", "Unsqueeze5714", "StridedSlice5717", "Squeeze5718", "BinaryOp5719", "BinaryOp5720", "/Gather_215_output_0", "/Div_42_output_0", "/Unsqueeze_455_output_0", "/Slice_57_output_0", "/Neg_28_output_0", "/Unsqueeze_454_output_0", "/Slice_56_output_0", "/Concat_185_output_0", "/Mul_113_output_0", "/Add_70_output_0", "/FakeLinear_57_output_0", "/Unsqueeze_450_output_0", "/Unsqueeze_451_output_0", "/Concat_183_output_0", "/Reshape_128_output_0", "/k_norm_14/Cast_output_0", "/k_norm_14/Mul_1_output_0", "/Mul_114_output_0", "/Shape_199_output_0", "Shape5602", "BinaryOp5605", "BinaryOp5606", "Unsqueeze5607", "BinaryOp5609", "Unsqueeze5610", "StridedSlice5613", "Squeeze5614", "BinaryOp5615", "BinaryOp5616", "/Gather_216_output_0", "/Div_43_output_0", "/Unsqueeze_457_output_0", "/Slice_59_output_0", "/Neg_29_output_0", "/Unsqueeze_456_output_0", "/Slice_58_output_0", "/Concat_186_output_0", "/Mul_115_output_0", "/Add_71_output_0", "/FakeLinear_58_output_0", "/Unsqueeze_452_output_0", "/Unsqueeze_453_output_0", "/Concat_184_output_0", "/Reshape_129_output_0", "/Reshape_134_output_0", "/FakeLinear_59_output_0", "/Add_73_output_0", "/post_attention_layernorm_14/Mul_1_output_0", "/mlp/gate_proj_14/FakeLinear_output_0", "/mlp/act_fn_14/Mul_output_0", "/mlp/up_proj_14/FakeLinear_output_0", "/mlp_14/Mul_output_0", "/mlp/down_proj_14/FakeLinear_output_0", "/Add_74_output_0", "/Reshape_135_output_0", "/input_layernorm_15/Mul_1_output_0", "/FakeLinear_60_output_0", "/Shape_210_output_0", "Shape5945", "BinaryOp5948", "BinaryOp5949", "Unsqueeze5950", "BinaryOp5952", "Unsqueeze5953", "StridedSlice5956", "BinaryOp5958", "BinaryOp5959", "/Gather_228_output_0", "/Unsqueeze_480_output_0", "Unsqueeze5967", "Unsqueeze5970", "StridedSlice5973", "Squeeze5974", "BinaryOp5975", "BinaryOp5976", "/Gather_229_output_0", "/Unsqueeze_481_output_0", "/Concat_195_output_0", "/Reshape_136_output_0", "/q_norm_15/Cast_output_0", "/q_norm_15/Mul_1_output_0", "/Mul_120_output_0", "/Shape_212_output_0", "Shape6095", "BinaryOp6098", "BinaryOp6099", "Unsqueeze6100", "BinaryOp6102", "Unsqueeze6103", "StridedSlice6106", "Squeeze6107", "BinaryOp6108", "BinaryOp6109", "/Gather_230_output_0", "/Div_45_output_0", "/Unsqueeze_487_output_0", "/Slice_61_output_0", "/Neg_30_output_0", "/Unsqueeze_486_output_0", "/Slice_60_output_0", "/Concat_198_output_0", "/Mul_121_output_0", "/Add_75_output_0", "/FakeLinear_61_output_0", "/Unsqueeze_482_output_0", "/Unsqueeze_483_output_0", "/Concat_196_output_0", "/Reshape_137_output_0", "/k_norm_15/Cast_output_0", "/k_norm_15/Mul_1_output_0", "/Mul_122_output_0", "/Shape_213_output_0", "Shape5991", "BinaryOp5994", "BinaryOp5995", "Unsqueeze5996", "BinaryOp5998", "Unsqueeze5999", "StridedSlice6002", "Squeeze6003", "BinaryOp6004", "BinaryOp6005", "/Gather_231_output_0", "/Div_46_output_0", "/Unsqueeze_489_output_0", "/Slice_63_output_0", "/Neg_31_output_0", "/Unsqueeze_488_output_0", "/Slice_62_output_0", "/Concat_199_output_0", "/Mul_123_output_0", "/Add_76_output_0", "/FakeLinear_62_output_0", "/Unsqueeze_484_output_0", "/Unsqueeze_485_output_0", "/Concat_197_output_0", "/Reshape_138_output_0", "/Reshape_143_output_0", "/FakeLinear_63_output_0", "/Add_78_output_0", "/post_attention_layernorm_15/Mul_1_output_0", "/mlp/gate_proj_15/FakeLinear_output_0", "/mlp/act_fn_15/Mul_output_0", "/mlp/up_proj_15/FakeLinear_output_0", "/mlp_15/Mul_output_0", "/mlp/down_proj_15/FakeLinear_output_0", "/Add_79_output_0", "/Reshape_144_output_0", "/input_layernorm_16/Mul_1_output_0", "/FakeLinear_64_output_0", "/Shape_224_output_0", "Shape6334", "BinaryOp6337", "BinaryOp6338", "Unsqueeze6339", "BinaryOp6341", "Unsqueeze6342", "StridedSlice6345", "BinaryOp6347", "BinaryOp6348", "/Gather_243_output_0", "/Unsqueeze_512_output_0", "Unsqueeze6356", "Unsqueeze6359", "StridedSlice6362", "Squeeze6363", "BinaryOp6364", "BinaryOp6365", "/Gather_244_output_0", "/Unsqueeze_513_output_0", "/Concat_208_output_0", "/Reshape_145_output_0", "/q_norm_16/Cast_output_0", "/q_norm_16/Mul_1_output_0", "/Mul_128_output_0", "/Shape_226_output_0", "Shape6484", "BinaryOp6487", "BinaryOp6488", "Unsqueeze6489", "BinaryOp6491", "Unsqueeze6492", "StridedSlice6495", "Squeeze6496", "BinaryOp6497", "BinaryOp6498", "/Gather_245_output_0", "/Div_48_output_0", "/Unsqueeze_519_output_0", "/Slice_65_output_0", "/Neg_32_output_0", "/Unsqueeze_518_output_0", "/Slice_64_output_0", "/Concat_211_output_0", "/Mul_129_output_0", "/Add_80_output_0", "/FakeLinear_65_output_0", "/Unsqueeze_514_output_0", "/Unsqueeze_515_output_0", "/Concat_209_output_0", "/Reshape_146_output_0", "/k_norm_16/Cast_output_0", "/k_norm_16/Mul_1_output_0", "/Mul_130_output_0", "/Shape_227_output_0", "Shape6380", "BinaryOp6383", "BinaryOp6384", "Unsqueeze6385", "BinaryOp6387", "Unsqueeze6388", "StridedSlice6391", "Squeeze6392", "BinaryOp6393", "BinaryOp6394", "/Gather_246_output_0", "/Div_49_output_0", "/Unsqueeze_521_output_0", "/Slice_67_output_0", "/Neg_33_output_0", "/Unsqueeze_520_output_0", "/Slice_66_output_0", "/Concat_212_output_0", "/Mul_131_output_0", "/Add_81_output_0", "/FakeLinear_66_output_0", "/Unsqueeze_516_output_0", "/Unsqueeze_517_output_0", "/Concat_210_output_0", "/Reshape_147_output_0", "/Reshape_152_output_0", "/FakeLinear_67_output_0", "/Add_83_output_0", "/post_attention_layernorm_16/Mul_1_output_0", "/mlp/gate_proj_16/FakeLinear_output_0", "/mlp/act_fn_16/Mul_output_0", "/mlp/up_proj_16/FakeLinear_output_0", "/mlp_16/Mul_output_0", "/mlp/down_proj_16/FakeLinear_output_0", "/Add_84_output_0", "/Reshape_153_output_0", "/input_layernorm_17/Mul_1_output_0", "/FakeLinear_68_output_0", "/Shape_238_output_0", "Shape6723", "BinaryOp6726", "BinaryOp6727", "Unsqueeze6728", "BinaryOp6730", "Unsqueeze6731", "StridedSlice6734", "BinaryOp6736", "BinaryOp6737", "/Gather_258_output_0", "/Unsqueeze_544_output_0", "Unsqueeze6745", "Unsqueeze6748", "StridedSlice6751", "Squeeze6752", "BinaryOp6753", "BinaryOp6754", "/Gather_259_output_0", "/Unsqueeze_545_output_0", "/Concat_221_output_0", "/Reshape_154_output_0", "/q_norm_17/Cast_output_0", "/q_norm_17/Mul_1_output_0", "/Mul_136_output_0", "/Shape_240_output_0", "Shape6873", "BinaryOp6876", "BinaryOp6877", "Unsqueeze6878", "BinaryOp6880", "Unsqueeze6881", "StridedSlice6884", "Squeeze6885", "BinaryOp6886", "BinaryOp6887", "/Gather_260_output_0", "/Div_51_output_0", "/Unsqueeze_551_output_0", "/Slice_69_output_0", "/Neg_34_output_0", "/Unsqueeze_550_output_0", "/Slice_68_output_0", "/Concat_224_output_0", "/Mul_137_output_0", "/Add_85_output_0", "/FakeLinear_69_output_0", "/Unsqueeze_546_output_0", "/Unsqueeze_547_output_0", "/Concat_222_output_0", "/Reshape_155_output_0", "/k_norm_17/Cast_output_0", "/k_norm_17/Mul_1_output_0", "/Mul_138_output_0", "/Shape_241_output_0", "Shape6769", "BinaryOp6772", "BinaryOp6773", "Unsqueeze6774", "BinaryOp6776", "Unsqueeze6777", "StridedSlice6780", "Squeeze6781", "BinaryOp6782", "BinaryOp6783", "/Gather_261_output_0", "/Div_52_output_0", "/Unsqueeze_553_output_0", "/Slice_71_output_0", "/Neg_35_output_0", "/Unsqueeze_552_output_0", "/Slice_70_output_0", "/Concat_225_output_0", "/Mul_139_output_0", "/Add_86_output_0", "/FakeLinear_70_output_0", "/Unsqueeze_548_output_0", "/Unsqueeze_549_output_0", "/Concat_223_output_0", "/Reshape_156_output_0", "/Reshape_161_output_0", "/FakeLinear_71_output_0", "/Add_88_output_0", "/post_attention_layernorm_17/Mul_1_output_0", "/mlp/gate_proj_17/FakeLinear_output_0", "/mlp/act_fn_17/Mul_output_0", "/mlp/up_proj_17/FakeLinear_output_0", "/mlp_17/Mul_output_0", "/mlp/down_proj_17/FakeLinear_output_0", "/Add_89_output_0", "/Reshape_162_output_0", "/input_layernorm_18/Mul_1_output_0", "/FakeLinear_72_output_0", "/Shape_252_output_0", "Shape7112", "BinaryOp7115", "BinaryOp7116", "Unsqueeze7117", "BinaryOp7119", "Unsqueeze7120", "StridedSlice7123", "BinaryOp7125", "BinaryOp7126", "/Gather_273_output_0", "/Unsqueeze_576_output_0", "Unsqueeze7134", "Unsqueeze7137", "StridedSlice7140", "Squeeze7141", "BinaryOp7142", "BinaryOp7143", "/Gather_274_output_0", "/Unsqueeze_577_output_0", "/Concat_234_output_0", "/Reshape_163_output_0", "/q_norm_18/Cast_output_0", "/q_norm_18/Mul_1_output_0", "/Mul_144_output_0", "/Shape_254_output_0", "Shape7262", "BinaryOp7265", "BinaryOp7266", "Unsqueeze7267", "BinaryOp7269", "Unsqueeze7270", "StridedSlice7273", "Squeeze7274", "BinaryOp7275", "BinaryOp7276", "/Gather_275_output_0", "/Div_54_output_0", "/Unsqueeze_583_output_0", "/Slice_73_output_0", "/Neg_36_output_0", "/Unsqueeze_582_output_0", "/Slice_72_output_0", "/Concat_237_output_0", "/Mul_145_output_0", "/Add_90_output_0", "/FakeLinear_73_output_0", "/Unsqueeze_578_output_0", "/Unsqueeze_579_output_0", "/Concat_235_output_0", "/Reshape_164_output_0", "/k_norm_18/Cast_output_0", "/k_norm_18/Mul_1_output_0", "/Mul_146_output_0", "/Shape_255_output_0", "Shape7158", "BinaryOp7161", "BinaryOp7162", "Unsqueeze7163", "BinaryOp7165", "Unsqueeze7166", "StridedSlice7169", "Squeeze7170", "BinaryOp7171", "BinaryOp7172", "/Gather_276_output_0", "/Div_55_output_0", "/Unsqueeze_585_output_0", "/Slice_75_output_0", "/Neg_37_output_0", "/Unsqueeze_584_output_0", "/Slice_74_output_0", "/Concat_238_output_0", "/Mul_147_output_0", "/Add_91_output_0", "/FakeLinear_74_output_0", "/Unsqueeze_580_output_0", "/Unsqueeze_581_output_0", "/Concat_236_output_0", "/Reshape_165_output_0", "/Reshape_170_output_0", "/FakeLinear_75_output_0", "/Add_93_output_0", "/post_attention_layernorm_18/Mul_1_output_0", "/mlp/gate_proj_18/FakeLinear_output_0", "/mlp/act_fn_18/Mul_output_0", "/mlp/up_proj_18/FakeLinear_output_0", "/mlp_18/Mul_output_0", "/mlp/down_proj_18/FakeLinear_output_0", "/Add_94_output_0", "/Reshape_171_output_0", "/input_layernorm_19/Mul_1_output_0", "/FakeLinear_76_output_0", "/Shape_266_output_0", "Shape7501", "BinaryOp7504", "BinaryOp7505", "Unsqueeze7506", "BinaryOp7508", "Unsqueeze7509", "StridedSlice7512", "BinaryOp7514", "BinaryOp7515", "/Gather_288_output_0", "/Unsqueeze_608_output_0", "Unsqueeze7523", "Unsqueeze7526", "StridedSlice7529", "Squeeze7530", "BinaryOp7531", "BinaryOp7532", "/Gather_289_output_0", "/Unsqueeze_609_output_0", "/Concat_247_output_0", "/Reshape_172_output_0", "/q_norm_19/Cast_output_0", "/q_norm_19/Mul_1_output_0", "/Mul_152_output_0", "/Shape_268_output_0", "Shape7651", "BinaryOp7654", "BinaryOp7655", "Unsqueeze7656", "BinaryOp7658", "Unsqueeze7659", "StridedSlice7662", "Squeeze7663", "BinaryOp7664", "BinaryOp7665", "/Gather_290_output_0", "/Div_57_output_0", "/Unsqueeze_615_output_0", "/Slice_77_output_0", "/Neg_38_output_0", "/Unsqueeze_614_output_0", "/Slice_76_output_0", "/Concat_250_output_0", "/Mul_153_output_0", "/Add_95_output_0", "/FakeLinear_77_output_0", "/Unsqueeze_610_output_0", "/Unsqueeze_611_output_0", "/Concat_248_output_0", "/Reshape_173_output_0", "/k_norm_19/Cast_output_0", "/k_norm_19/Mul_1_output_0", "/Mul_154_output_0", "/Shape_269_output_0", "Shape7547", "BinaryOp7550", "BinaryOp7551", "Unsqueeze7552", "BinaryOp7554", "Unsqueeze7555", "StridedSlice7558", "Squeeze7559", "BinaryOp7560", "BinaryOp7561", "/Gather_291_output_0", "/Div_58_output_0", "/Unsqueeze_617_output_0", "/Slice_79_output_0", "/Neg_39_output_0", "/Unsqueeze_616_output_0", "/Slice_78_output_0", "/Concat_251_output_0", "/Mul_155_output_0", "/Add_96_output_0", "/FakeLinear_78_output_0", "/Unsqueeze_612_output_0", "/Unsqueeze_613_output_0", "/Concat_249_output_0", "/Reshape_174_output_0", "/Reshape_179_output_0", "/FakeLinear_79_output_0", "/Add_98_output_0", "/post_attention_layernorm_19/Mul_1_output_0", "/mlp/gate_proj_19/FakeLinear_output_0", "/mlp/act_fn_19/Mul_output_0", "/mlp/up_proj_19/FakeLinear_output_0", "/mlp_19/Mul_output_0", "/mlp/down_proj_19/FakeLinear_output_0", "/Add_99_output_0", "/Reshape_180_output_0", "/input_layernorm_20/Mul_1_output_0", "/FakeLinear_80_output_0", "/Shape_280_output_0", "Shape7890", "BinaryOp7893", "BinaryOp7894", "Unsqueeze7895", "BinaryOp7897", "Unsqueeze7898", "StridedSlice7901", "BinaryOp7903", "BinaryOp7904", "/Gather_303_output_0", "/Unsqueeze_640_output_0", "Unsqueeze7912", "Unsqueeze7915", "StridedSlice7918", "Squeeze7919", "BinaryOp7920", "BinaryOp7921", "/Gather_304_output_0", "/Unsqueeze_641_output_0", "/Concat_260_output_0", "/Reshape_181_output_0", "/q_norm_20/Cast_output_0", "/q_norm_20/Mul_1_output_0", "/Mul_160_output_0", "/Shape_282_output_0", "Shape8040", "BinaryOp8043", "BinaryOp8044", "Unsqueeze8045", "BinaryOp8047", "Unsqueeze8048", "StridedSlice8051", "Squeeze8052", "BinaryOp8053", "BinaryOp8054", "/Gather_305_output_0", "/Div_60_output_0", "/Unsqueeze_647_output_0", "/Slice_81_output_0", "/Neg_40_output_0", "/Unsqueeze_646_output_0", "/Slice_80_output_0", "/Concat_263_output_0", "/Mul_161_output_0", "/Add_100_output_0", "/FakeLinear_81_output_0", "/Unsqueeze_642_output_0", "/Unsqueeze_643_output_0", "/Concat_261_output_0", "/Reshape_182_output_0", "/k_norm_20/Cast_output_0", "/k_norm_20/Mul_1_output_0", "/Mul_162_output_0", "/Shape_283_output_0", "Shape7936", "BinaryOp7939", "BinaryOp7940", "Unsqueeze7941", "BinaryOp7943", "Unsqueeze7944", "StridedSlice7947", "Squeeze7948", "BinaryOp7949", "BinaryOp7950", "/Gather_306_output_0", "/Div_61_output_0", "/Unsqueeze_649_output_0", "/Slice_83_output_0", "/Neg_41_output_0", "/Unsqueeze_648_output_0", "/Slice_82_output_0", "/Concat_264_output_0", "/Mul_163_output_0", "/Add_101_output_0", "/FakeLinear_82_output_0", "/Unsqueeze_644_output_0", "/Unsqueeze_645_output_0", "/Concat_262_output_0", "/Reshape_183_output_0", "/Reshape_188_output_0", "/FakeLinear_83_output_0", "/Add_103_output_0", "/post_attention_layernorm_20/Mul_1_output_0", "/mlp/gate_proj_20/FakeLinear_output_0", "/mlp/act_fn_20/Mul_output_0", "/mlp/up_proj_20/FakeLinear_output_0", "/mlp_20/Mul_output_0", "/mlp/down_proj_20/FakeLinear_output_0", "/Add_104_output_0", "/Reshape_189_output_0", "/input_layernorm_21/Mul_1_output_0", "/FakeLinear_84_output_0", "/Shape_294_output_0", "Shape8279", "BinaryOp8282", "BinaryOp8283", "Unsqueeze8284", "BinaryOp8286", "Unsqueeze8287", "StridedSlice8290", "BinaryOp8292", "BinaryOp8293", "/Gather_318_output_0", "/Unsqueeze_672_output_0", "Unsqueeze8301", "Unsqueeze8304", "StridedSlice8307", "Squeeze8308", "BinaryOp8309", "BinaryOp8310", "/Gather_319_output_0", "/Unsqueeze_673_output_0", "/Concat_273_output_0", "/Reshape_190_output_0", "/q_norm_21/Cast_output_0", "/q_norm_21/Mul_1_output_0", "/Mul_168_output_0", "/Shape_296_output_0", "Shape8429", "BinaryOp8432", "BinaryOp8433", "Unsqueeze8434", "BinaryOp8436", "Unsqueeze8437", "StridedSlice8440", "Squeeze8441", "BinaryOp8442", "BinaryOp8443", "/Gather_320_output_0", "/Div_63_output_0", "/Unsqueeze_679_output_0", "/Slice_85_output_0", "/Neg_42_output_0", "/Unsqueeze_678_output_0", "/Slice_84_output_0", "/Concat_276_output_0", "/Mul_169_output_0", "/Add_105_output_0", "/FakeLinear_85_output_0", "/Unsqueeze_674_output_0", "/Unsqueeze_675_output_0", "/Concat_274_output_0", "/Reshape_191_output_0", "/k_norm_21/Cast_output_0", "/k_norm_21/Mul_1_output_0", "/Mul_170_output_0", "/Shape_297_output_0", "Shape8325", "BinaryOp8328", "BinaryOp8329", "Unsqueeze8330", "BinaryOp8332", "Unsqueeze8333", "StridedSlice8336", "Squeeze8337", "BinaryOp8338", "BinaryOp8339", "/Gather_321_output_0", "/Div_64_output_0", "/Unsqueeze_681_output_0", "/Slice_87_output_0", "/Neg_43_output_0", "/Unsqueeze_680_output_0", "/Slice_86_output_0", "/Concat_277_output_0", "/Mul_171_output_0", "/Add_106_output_0", "/FakeLinear_86_output_0", "/Unsqueeze_676_output_0", "/Unsqueeze_677_output_0", "/Concat_275_output_0", "/Reshape_192_output_0", "/Reshape_197_output_0", "/FakeLinear_87_output_0", "/Add_108_output_0", "/post_attention_layernorm_21/Mul_1_output_0", "/mlp/gate_proj_21/FakeLinear_output_0", "/mlp/act_fn_21/Mul_output_0", "/mlp/up_proj_21/FakeLinear_output_0", "/mlp_21/Mul_output_0", "/mlp/down_proj_21/FakeLinear_output_0", "/Add_109_output_0", "/Reshape_198_output_0", "/input_layernorm_22/Mul_1_output_0", "/FakeLinear_88_output_0", "/Shape_308_output_0", "Shape8668", "BinaryOp8671", "BinaryOp8672", "Unsqueeze8673", "BinaryOp8675", "Unsqueeze8676", "StridedSlice8679", "BinaryOp8681", "BinaryOp8682", "/Gather_333_output_0", "/Unsqueeze_704_output_0", "Unsqueeze8690", "Unsqueeze8693", "StridedSlice8696", "Squeeze8697", "BinaryOp8698", "BinaryOp8699", "/Gather_334_output_0", "/Unsqueeze_705_output_0", "/Concat_286_output_0", "/Reshape_199_output_0", "/q_norm_22/Cast_output_0", "/q_norm_22/Mul_1_output_0", "/Mul_176_output_0", "/Shape_310_output_0", "Shape8818", "BinaryOp8821", "BinaryOp8822", "Unsqueeze8823", "BinaryOp8825", "Unsqueeze8826", "StridedSlice8829", "Squeeze8830", "BinaryOp8831", "BinaryOp8832", "/Gather_335_output_0", "/Div_66_output_0", "/Unsqueeze_711_output_0", "/Slice_89_output_0", "/Neg_44_output_0", "/Unsqueeze_710_output_0", "/Slice_88_output_0", "/Concat_289_output_0", "/Mul_177_output_0", "/Add_110_output_0", "/FakeLinear_89_output_0", "/Unsqueeze_706_output_0", "/Unsqueeze_707_output_0", "/Concat_287_output_0", "/Reshape_200_output_0", "/k_norm_22/Cast_output_0", "/k_norm_22/Mul_1_output_0", "/Mul_178_output_0", "/Shape_311_output_0", "Shape8714", "BinaryOp8717", "BinaryOp8718", "Unsqueeze8719", "BinaryOp8721", "Unsqueeze8722", "StridedSlice8725", "Squeeze8726", "BinaryOp8727", "BinaryOp8728", "/Gather_336_output_0", "/Div_67_output_0", "/Unsqueeze_713_output_0", "/Slice_91_output_0", "/Neg_45_output_0", "/Unsqueeze_712_output_0", "/Slice_90_output_0", "/Concat_290_output_0", "/Mul_179_output_0", "/Add_111_output_0", "/FakeLinear_90_output_0", "/Unsqueeze_708_output_0", "/Unsqueeze_709_output_0", "/Concat_288_output_0", "/Reshape_201_output_0", "/Reshape_206_output_0", "/FakeLinear_91_output_0", "/Add_113_output_0", "/post_attention_layernorm_22/Mul_1_output_0", "/mlp/gate_proj_22/FakeLinear_output_0", "/mlp/act_fn_22/Mul_output_0", "/mlp/up_proj_22/FakeLinear_output_0", "/mlp_22/Mul_output_0", "/mlp/down_proj_22/FakeLinear_output_0", "/Add_114_output_0", "/Reshape_207_output_0", "/input_layernorm_23/Mul_1_output_0", "/FakeLinear_92_output_0", "/Shape_322_output_0", "Shape9057", "BinaryOp9060", "BinaryOp9061", "Unsqueeze9062", "BinaryOp9064", "Unsqueeze9065", "StridedSlice9068", "BinaryOp9070", "BinaryOp9071", "/Gather_348_output_0", "/Unsqueeze_736_output_0", "Unsqueeze9079", "Unsqueeze9082", "StridedSlice9085", "Squeeze9086", "BinaryOp9087", "BinaryOp9088", "/Gather_349_output_0", "/Unsqueeze_737_output_0", "/Concat_299_output_0", "/Reshape_208_output_0", "/q_norm_23/Cast_output_0", "/q_norm_23/Mul_1_output_0", "/Mul_184_output_0", "/Shape_324_output_0", "Shape9207", "BinaryOp9210", "BinaryOp9211", "Unsqueeze9212", "BinaryOp9214", "Unsqueeze9215", "StridedSlice9218", "Squeeze9219", "BinaryOp9220", "BinaryOp9221", "/Gather_350_output_0", "/Div_69_output_0", "/Unsqueeze_743_output_0", "/Slice_93_output_0", "/Neg_46_output_0", "/Unsqueeze_742_output_0", "/Slice_92_output_0", "/Concat_302_output_0", "/Mul_185_output_0", "/Add_115_output_0", "/FakeLinear_93_output_0", "/Unsqueeze_738_output_0", "/Unsqueeze_739_output_0", "/Concat_300_output_0", "/Reshape_209_output_0", "/k_norm_23/Cast_output_0", "/k_norm_23/Mul_1_output_0", "/Mul_186_output_0", "/Shape_325_output_0", "Shape9103", "BinaryOp9106", "BinaryOp9107", "Unsqueeze9108", "BinaryOp9110", "Unsqueeze9111", "StridedSlice9114", "Squeeze9115", "BinaryOp9116", "BinaryOp9117", "/Gather_351_output_0", "/Div_70_output_0", "/Unsqueeze_745_output_0", "/Slice_95_output_0", "/Neg_47_output_0", "/Unsqueeze_744_output_0", "/Slice_94_output_0", "/Concat_303_output_0", "/Mul_187_output_0", "/Add_116_output_0", "/FakeLinear_94_output_0", "/Unsqueeze_740_output_0", "/Unsqueeze_741_output_0", "/Concat_301_output_0", "/Reshape_210_output_0", "/Reshape_215_output_0", "/FakeLinear_95_output_0", "/Add_118_output_0", "/post_attention_layernorm_23/Mul_1_output_0", "/mlp/gate_proj_23/FakeLinear_output_0", "/mlp/act_fn_23/Mul_output_0", "/mlp/up_proj_23/FakeLinear_output_0", "/mlp_23/Mul_output_0", "/mlp/down_proj_23/FakeLinear_output_0", "/Add_119_output_0", "/Reshape_216_output_0", "/input_layernorm_24/Mul_1_output_0", "/FakeLinear_96_output_0", "/Shape_336_output_0", "Shape9446", "BinaryOp9449", "BinaryOp9450", "Unsqueeze9451", "BinaryOp9453", "Unsqueeze9454", "StridedSlice9457", "BinaryOp9459", "BinaryOp9460", "/Gather_363_output_0", "/Unsqueeze_768_output_0", "Unsqueeze9468", "Unsqueeze9471", "StridedSlice9474", "Squeeze9475", "BinaryOp9476", "BinaryOp9477", "/Gather_364_output_0", "/Unsqueeze_769_output_0", "/Concat_312_output_0", "/Reshape_217_output_0", "/q_norm_24/Cast_output_0", "/q_norm_24/Mul_1_output_0", "/Mul_192_output_0", "/Shape_338_output_0", "Shape9596", "BinaryOp9599", "BinaryOp9600", "Unsqueeze9601", "BinaryOp9603", "Unsqueeze9604", "StridedSlice9607", "Squeeze9608", "BinaryOp9609", "BinaryOp9610", "/Gather_365_output_0", "/Div_72_output_0", "/Unsqueeze_775_output_0", "/Slice_97_output_0", "/Neg_48_output_0", "/Unsqueeze_774_output_0", "/Slice_96_output_0", "/Concat_315_output_0", "/Mul_193_output_0", "/Add_120_output_0", "/FakeLinear_97_output_0", "/Unsqueeze_770_output_0", "/Unsqueeze_771_output_0", "/Concat_313_output_0", "/Reshape_218_output_0", "/k_norm_24/Cast_output_0", "/k_norm_24/Mul_1_output_0", "/Mul_194_output_0", "/Shape_339_output_0", "Shape9492", "BinaryOp9495", "BinaryOp9496", "Unsqueeze9497", "BinaryOp9499", "Unsqueeze9500", "StridedSlice9503", "Squeeze9504", "BinaryOp9505", "BinaryOp9506", "/Gather_366_output_0", "/Div_73_output_0", "/Unsqueeze_777_output_0", "/Slice_99_output_0", "/Neg_49_output_0", "/Unsqueeze_776_output_0", "/Slice_98_output_0", "/Concat_316_output_0", "/Mul_195_output_0", "/Add_121_output_0", "/FakeLinear_98_output_0", "/Unsqueeze_772_output_0", "/Unsqueeze_773_output_0", "/Concat_314_output_0", "/Reshape_219_output_0", "/Reshape_224_output_0", "/FakeLinear_99_output_0", "/Add_123_output_0", "/post_attention_layernorm_24/Mul_1_output_0", "/mlp/gate_proj_24/FakeLinear_output_0", "/mlp/act_fn_24/Mul_output_0", "/mlp/up_proj_24/FakeLinear_output_0", "/mlp_24/Mul_output_0", "/mlp/down_proj_24/FakeLinear_output_0", "/Add_124_output_0", "/Reshape_225_output_0", "/input_layernorm_25/Mul_1_output_0", "/FakeLinear_100_output_0", "/Shape_350_output_0", "Shape9835", "BinaryOp9838", "BinaryOp9839", "Unsqueeze9840", "BinaryOp9842", "Unsqueeze9843", "StridedSlice9846", "BinaryOp9848", "BinaryOp9849", "/Gather_378_output_0", "/Unsqueeze_800_output_0", "Unsqueeze9857", "Unsqueeze9860", "StridedSlice9863", "Squeeze9864", "BinaryOp9865", "BinaryOp9866", "/Gather_379_output_0", "/Unsqueeze_801_output_0", "/Concat_325_output_0", "/Reshape_226_output_0", "/q_norm_25/Cast_output_0", "/q_norm_25/Mul_1_output_0", "/Mul_200_output_0", "/Shape_352_output_0", "Shape9985", "BinaryOp9988", "BinaryOp9989", "Unsqueeze9990", "BinaryOp9992", "Unsqueeze9993", "StridedSlice9996", "Squeeze9997", "BinaryOp9998", "BinaryOp9999", "/Gather_380_output_0", "/Div_75_output_0", "/Unsqueeze_807_output_0", "/Slice_101_output_0", "/Neg_50_output_0", "/Unsqueeze_806_output_0", "/Slice_100_output_0", "/Concat_328_output_0", "/Mul_201_output_0", "/Add_125_output_0", "/FakeLinear_101_output_0", "/Unsqueeze_802_output_0", "/Unsqueeze_803_output_0", "/Concat_326_output_0", "/Reshape_227_output_0", "/k_norm_25/Cast_output_0", "/k_norm_25/Mul_1_output_0", "/Mul_202_output_0", "/Shape_353_output_0", "Shape9881", "BinaryOp9884", "BinaryOp9885", "Unsqueeze9886", "BinaryOp9888", "Unsqueeze9889", "StridedSlice9892", "Squeeze9893", "BinaryOp9894", "BinaryOp9895", "/Gather_381_output_0", "/Div_76_output_0", "/Unsqueeze_809_output_0", "/Slice_103_output_0", "/Neg_51_output_0", "/Unsqueeze_808_output_0", "/Slice_102_output_0", "/Concat_329_output_0", "/Mul_203_output_0", "/Add_126_output_0", "/FakeLinear_102_output_0", "/Unsqueeze_804_output_0", "/Unsqueeze_805_output_0", "/Concat_327_output_0", "/Reshape_228_output_0", "/Reshape_233_output_0", "/FakeLinear_103_output_0", "/Add_128_output_0", "/post_attention_layernorm_25/Mul_1_output_0", "/mlp/gate_proj_25/FakeLinear_output_0", "/mlp/act_fn_25/Mul_output_0", "/mlp/up_proj_25/FakeLinear_output_0", "/mlp_25/Mul_output_0", "/mlp/down_proj_25/FakeLinear_output_0", "/Add_129_output_0", "/Reshape_234_output_0", "/input_layernorm_26/Mul_1_output_0", "/FakeLinear_104_output_0", "/Shape_364_output_0", "Shape10224", "BinaryOp10227", "BinaryOp10228", "Unsqueeze10229", "BinaryOp10231", "Unsqueeze10232", "StridedSlice10235", "BinaryOp10237", "BinaryOp10238", "/Gather_393_output_0", "/Unsqueeze_832_output_0", "Unsqueeze10246", "Unsqueeze10249", "StridedSlice10252", "Squeeze10253", "BinaryOp10254", "BinaryOp10255", "/Gather_394_output_0", "/Unsqueeze_833_output_0", "/Concat_338_output_0", "/Reshape_235_output_0", "/q_norm_26/Cast_output_0", "/q_norm_26/Mul_1_output_0", "/Mul_208_output_0", "/Shape_366_output_0", "Shape10374", "BinaryOp10377", "BinaryOp10378", "Unsqueeze10379", "BinaryOp10381", "Unsqueeze10382", "StridedSlice10385", "Squeeze10386", "BinaryOp10387", "BinaryOp10388", "/Gather_395_output_0", "/Div_78_output_0", "/Unsqueeze_839_output_0", "/Slice_105_output_0", "/Neg_52_output_0", "/Unsqueeze_838_output_0", "/Slice_104_output_0", "/Concat_341_output_0", "/Mul_209_output_0", "/Add_130_output_0", "/FakeLinear_105_output_0", "/Unsqueeze_834_output_0", "/Unsqueeze_835_output_0", "/Concat_339_output_0", "/Reshape_236_output_0", "/k_norm_26/Cast_output_0", "/k_norm_26/Mul_1_output_0", "/Mul_210_output_0", "/Shape_367_output_0", "Shape10270", "BinaryOp10273", "BinaryOp10274", "Unsqueeze10275", "BinaryOp10277", "Unsqueeze10278", "StridedSlice10281", "Squeeze10282", "BinaryOp10283", "BinaryOp10284", "/Gather_396_output_0", "/Div_79_output_0", "/Unsqueeze_841_output_0", "/Slice_107_output_0", "/Neg_53_output_0", "/Unsqueeze_840_output_0", "/Slice_106_output_0", "/Concat_342_output_0", "/Mul_211_output_0", "/Add_131_output_0", "/FakeLinear_106_output_0", "/Unsqueeze_836_output_0", "/Unsqueeze_837_output_0", "/Concat_340_output_0", "/Reshape_237_output_0", "/Reshape_242_output_0", "/FakeLinear_107_output_0", "/Add_133_output_0", "/post_attention_layernorm_26/Mul_1_output_0", "/mlp/gate_proj_26/FakeLinear_output_0", "/mlp/act_fn_26/Mul_output_0", "/mlp/up_proj_26/FakeLinear_output_0", "/mlp_26/Mul_output_0", "/mlp/down_proj_26/FakeLinear_output_0", "/Add_134_output_0", "/Reshape_243_output_0", "/input_layernorm_27/Mul_1_output_0", "/FakeLinear_108_output_0", "/Shape_378_output_0", "Shape10613", "BinaryOp10616", "BinaryOp10617", "Unsqueeze10618", "BinaryOp10620", "Unsqueeze10621", "StridedSlice10624", "BinaryOp10626", "BinaryOp10627", "/Gather_408_output_0", "/Unsqueeze_864_output_0", "Unsqueeze10635", "Unsqueeze10638", "StridedSlice10641", "Squeeze10642", "BinaryOp10643", "BinaryOp10644", "/Gather_409_output_0", "/Unsqueeze_865_output_0", "/Concat_351_output_0", "/Reshape_244_output_0", "/q_norm_27/Cast_output_0", "/q_norm_27/Mul_1_output_0", "/Mul_216_output_0", "/Shape_380_output_0", "Shape10763", "BinaryOp10766", "BinaryOp10767", "Unsqueeze10768", "BinaryOp10770", "Unsqueeze10771", "StridedSlice10774", "Squeeze10775", "BinaryOp10776", "BinaryOp10777", "/Gather_410_output_0", "/Div_81_output_0", "/Unsqueeze_871_output_0", "/Slice_109_output_0", "/Neg_54_output_0", "/Unsqueeze_870_output_0", "/Slice_108_output_0", "/Concat_354_output_0", "/Mul_217_output_0", "/Add_135_output_0", "/FakeLinear_109_output_0", "/Unsqueeze_866_output_0", "/Unsqueeze_867_output_0", "/Concat_352_output_0", "/Reshape_245_output_0", "/k_norm_27/Cast_output_0", "/k_norm_27/Mul_1_output_0", "/Mul_218_output_0", "/Shape_381_output_0", "Shape10659", "BinaryOp10662", "BinaryOp10663", "Unsqueeze10664", "BinaryOp10666", "Unsqueeze10667", "StridedSlice10670", "Squeeze10671", "BinaryOp10672", "BinaryOp10673", "/Gather_411_output_0", "/Div_82_output_0", "/Unsqueeze_873_output_0", "/Slice_111_output_0", "/Neg_55_output_0", "/Unsqueeze_872_output_0", "/Slice_110_output_0", "/Concat_355_output_0", "/Mul_219_output_0", "/Add_136_output_0", "/FakeLinear_110_output_0", "/Unsqueeze_868_output_0", "/Unsqueeze_869_output_0", "/Concat_353_output_0", "/Reshape_246_output_0", "/Reshape_251_output_0", "/FakeLinear_111_output_0", "/Add_138_output_0", "/post_attention_layernorm_27/Mul_1_output_0", "/mlp/gate_proj_27/FakeLinear_output_0", "/mlp/act_fn_27/Mul_output_0", "/mlp/up_proj_27/FakeLinear_output_0", "/mlp_27/Mul_output_0", "/mlp/down_proj_27/FakeLinear_output_0", "/Add_139_output_0", "/Reshape_252_output_0", "/input_layernorm_28/Mul_1_output_0", "/FakeLinear_112_output_0", "/Shape_392_output_0", "Shape11002", "BinaryOp11005", "BinaryOp11006", "Unsqueeze11007", "BinaryOp11009", "Unsqueeze11010", "StridedSlice11013", "BinaryOp11015", "BinaryOp11016", "/Gather_423_output_0", "/Unsqueeze_896_output_0", "Unsqueeze11024", "Unsqueeze11027", "StridedSlice11030", "Squeeze11031", "BinaryOp11032", "BinaryOp11033", "/Gather_424_output_0", "/Unsqueeze_897_output_0", "/Concat_364_output_0", "/Reshape_253_output_0", "/q_norm_28/Cast_output_0", "/q_norm_28/Mul_1_output_0", "/Mul_224_output_0", "/Shape_394_output_0", "Shape11152", "BinaryOp11155", "BinaryOp11156", "Unsqueeze11157", "BinaryOp11159", "Unsqueeze11160", "StridedSlice11163", "Squeeze11164", "BinaryOp11165", "BinaryOp11166", "/Gather_425_output_0", "/Div_84_output_0", "/Unsqueeze_903_output_0", "/Slice_113_output_0", "/Neg_56_output_0", "/Unsqueeze_902_output_0", "/Slice_112_output_0", "/Concat_367_output_0", "/Mul_225_output_0", "/Add_140_output_0", "/FakeLinear_113_output_0", "/Unsqueeze_898_output_0", "/Unsqueeze_899_output_0", "/Concat_365_output_0", "/Reshape_254_output_0", "/k_norm_28/Cast_output_0", "/k_norm_28/Mul_1_output_0", "/Mul_226_output_0", "/Shape_395_output_0", "Shape11048", "BinaryOp11051", "BinaryOp11052", "Unsqueeze11053", "BinaryOp11055", "Unsqueeze11056", "StridedSlice11059", "Squeeze11060", "BinaryOp11061", "BinaryOp11062", "/Gather_426_output_0", "/Div_85_output_0", "/Unsqueeze_905_output_0", "/Slice_115_output_0", "/Neg_57_output_0", "/Unsqueeze_904_output_0", "/Slice_114_output_0", "/Concat_368_output_0", "/Mul_227_output_0", "/Add_141_output_0", "/FakeLinear_114_output_0", "/Unsqueeze_900_output_0", "/Unsqueeze_901_output_0", "/Concat_366_output_0", "/Reshape_255_output_0", "/Reshape_260_output_0", "/FakeLinear_115_output_0", "/Add_143_output_0", "/post_attention_layernorm_28/Mul_1_output_0", "/mlp/gate_proj_28/FakeLinear_output_0", "/mlp/act_fn_28/Mul_output_0", "/mlp/up_proj_28/FakeLinear_output_0", "/mlp_28/Mul_output_0", "/mlp/down_proj_28/FakeLinear_output_0", "/Add_144_output_0", "/Reshape_261_output_0", "/input_layernorm_29/Mul_1_output_0", "/FakeLinear_116_output_0", "/Shape_406_output_0", "Shape11391", "BinaryOp11394", "BinaryOp11395", "Unsqueeze11396", "BinaryOp11398", "Unsqueeze11399", "StridedSlice11402", "BinaryOp11404", "BinaryOp11405", "/Gather_438_output_0", "/Unsqueeze_928_output_0", "Unsqueeze11413", "Unsqueeze11416", "StridedSlice11419", "Squeeze11420", "BinaryOp11421", "BinaryOp11422", "/Gather_439_output_0", "/Unsqueeze_929_output_0", "/Concat_377_output_0", "/Reshape_262_output_0", "/q_norm_29/Cast_output_0", "/q_norm_29/Mul_1_output_0", "/Mul_232_output_0", "/Shape_408_output_0", "Shape11541", "BinaryOp11544", "BinaryOp11545", "Unsqueeze11546", "BinaryOp11548", "Unsqueeze11549", "StridedSlice11552", "Squeeze11553", "BinaryOp11554", "BinaryOp11555", "/Gather_440_output_0", "/Div_87_output_0", "/Unsqueeze_935_output_0", "/Slice_117_output_0", "/Neg_58_output_0", "/Unsqueeze_934_output_0", "/Slice_116_output_0", "/Concat_380_output_0", "/Mul_233_output_0", "/Add_145_output_0", "/FakeLinear_117_output_0", "/Unsqueeze_930_output_0", "/Unsqueeze_931_output_0", "/Concat_378_output_0", "/Reshape_263_output_0", "/k_norm_29/Cast_output_0", "/k_norm_29/Mul_1_output_0", "/Mul_234_output_0", "/Shape_409_output_0", "Shape11437", "BinaryOp11440", "BinaryOp11441", "Unsqueeze11442", "BinaryOp11444", "Unsqueeze11445", "StridedSlice11448", "Squeeze11449", "BinaryOp11450", "BinaryOp11451", "/Gather_441_output_0", "/Div_88_output_0", "/Unsqueeze_937_output_0", "/Slice_119_output_0", "/Neg_59_output_0", "/Unsqueeze_936_output_0", "/Slice_118_output_0", "/Concat_381_output_0", "/Mul_235_output_0", "/Add_146_output_0", "/FakeLinear_118_output_0", "/Unsqueeze_932_output_0", "/Unsqueeze_933_output_0", "/Concat_379_output_0", "/Reshape_264_output_0", "/Reshape_269_output_0", "/FakeLinear_119_output_0", "/Add_148_output_0", "/post_attention_layernorm_29/Mul_1_output_0", "/mlp/gate_proj_29/FakeLinear_output_0", "/mlp/act_fn_29/Mul_output_0", "/mlp/up_proj_29/FakeLinear_output_0", "/mlp_29/Mul_output_0", "/mlp/down_proj_29/FakeLinear_output_0", "/Add_149_output_0", "/Reshape_270_output_0", "/input_layernorm_30/Mul_1_output_0", "/FakeLinear_120_output_0", "/Shape_420_output_0", "Shape11780", "BinaryOp11783", "BinaryOp11784", "Unsqueeze11785", "BinaryOp11787", "Unsqueeze11788", "StridedSlice11791", "BinaryOp11793", "BinaryOp11794", "/Gather_453_output_0", "/Unsqueeze_960_output_0", "Unsqueeze11802", "Unsqueeze11805", "StridedSlice11808", "Squeeze11809", "BinaryOp11810", "BinaryOp11811", "/Gather_454_output_0", "/Unsqueeze_961_output_0", "/Concat_390_output_0", "/Reshape_271_output_0", "/q_norm_30/Cast_output_0", "/q_norm_30/Mul_1_output_0", "/Mul_240_output_0", "/Shape_422_output_0", "Shape11930", "BinaryOp11933", "BinaryOp11934", "Unsqueeze11935", "BinaryOp11937", "Unsqueeze11938", "StridedSlice11941", "Squeeze11942", "BinaryOp11943", "BinaryOp11944", "/Gather_455_output_0", "/Div_90_output_0", "/Unsqueeze_967_output_0", "/Slice_121_output_0", "/Neg_60_output_0", "/Unsqueeze_966_output_0", "/Slice_120_output_0", "/Concat_393_output_0", "/Mul_241_output_0", "/Add_150_output_0", "/FakeLinear_121_output_0", "/Unsqueeze_962_output_0", "/Unsqueeze_963_output_0", "/Concat_391_output_0", "/Reshape_272_output_0", "/k_norm_30/Cast_output_0", "/k_norm_30/Mul_1_output_0", "/Mul_242_output_0", "/Shape_423_output_0", "Shape11826", "BinaryOp11829", "BinaryOp11830", "Unsqueeze11831", "BinaryOp11833", "Unsqueeze11834", "StridedSlice11837", "Squeeze11838", "BinaryOp11839", "BinaryOp11840", "/Gather_456_output_0", "/Div_91_output_0", "/Unsqueeze_969_output_0", "/Slice_123_output_0", "/Neg_61_output_0", "/Unsqueeze_968_output_0", "/Slice_122_output_0", "/Concat_394_output_0", "/Mul_243_output_0", "/Add_151_output_0", "/FakeLinear_122_output_0", "/Unsqueeze_964_output_0", "/Unsqueeze_965_output_0", "/Concat_392_output_0", "/Reshape_273_output_0", "/Reshape_278_output_0", "/FakeLinear_123_output_0", "/Add_153_output_0", "/post_attention_layernorm_30/Mul_1_output_0", "/mlp/gate_proj_30/FakeLinear_output_0", "/mlp/act_fn_30/Mul_output_0", "/mlp/up_proj_30/FakeLinear_output_0", "/mlp_30/Mul_output_0", "/mlp/down_proj_30/FakeLinear_output_0", "/Add_154_output_0", "/Reshape_279_output_0", "/input_layernorm_31/Mul_1_output_0", "/FakeLinear_124_output_0", "/Shape_434_output_0", "Shape12169", "BinaryOp12172", "BinaryOp12173", "Unsqueeze12174", "BinaryOp12176", "Unsqueeze12177", "StridedSlice12180", "BinaryOp12182", "BinaryOp12183", "/Gather_468_output_0", "/Unsqueeze_992_output_0", "Unsqueeze12191", "Unsqueeze12194", "StridedSlice12197", "Squeeze12198", "BinaryOp12199", "BinaryOp12200", "/Gather_469_output_0", "/Unsqueeze_993_output_0", "/Concat_403_output_0", "/Reshape_280_output_0", "/q_norm_31/Cast_output_0", "/q_norm_31/Mul_1_output_0", "/Mul_248_output_0", "/Shape_436_output_0", "Shape12319", "BinaryOp12322", "BinaryOp12323", "Unsqueeze12324", "BinaryOp12326", "Unsqueeze12327", "StridedSlice12330", "Squeeze12331", "BinaryOp12332", "BinaryOp12333", "/Gather_470_output_0", "/Div_93_output_0", "/Unsqueeze_999_output_0", "/Slice_125_output_0", "/Neg_62_output_0", "/Unsqueeze_998_output_0", "/Slice_124_output_0", "/Concat_406_output_0", "/Mul_249_output_0", "/Add_155_output_0", "/FakeLinear_125_output_0", "/Unsqueeze_994_output_0", "/Unsqueeze_995_output_0", "/Concat_404_output_0", "/Reshape_281_output_0", "/k_norm_31/Cast_output_0", "/k_norm_31/Mul_1_output_0", "/Mul_250_output_0", "/Shape_437_output_0", "Shape12215", "BinaryOp12218", "BinaryOp12219", "Unsqueeze12220", "BinaryOp12222", "Unsqueeze12223", "StridedSlice12226", "Squeeze12227", "BinaryOp12228", "BinaryOp12229", "/Gather_471_output_0", "/Div_94_output_0", "/Unsqueeze_1001_output_0", "/Slice_127_output_0", "/Neg_63_output_0", "/Unsqueeze_1000_output_0", "/Slice_126_output_0", "/Concat_407_output_0", "/Mul_251_output_0", "/Add_156_output_0", "/FakeLinear_126_output_0", "/Unsqueeze_996_output_0", "/Unsqueeze_997_output_0", "/Concat_405_output_0", "/Reshape_282_output_0", "/Reshape_287_output_0", "/FakeLinear_127_output_0", "/Add_158_output_0", "/post_attention_layernorm_31/Mul_1_output_0", "/mlp/gate_proj_31/FakeLinear_output_0", "/mlp/act_fn_31/Mul_output_0", "/mlp/up_proj_31/FakeLinear_output_0", "/mlp_31/Mul_output_0", "/mlp/down_proj_31/FakeLinear_output_0", "/Add_159_output_0", "/Reshape_288_output_0", "/input_layernorm_32/Mul_1_output_0", "/FakeLinear_128_output_0", "/Shape_448_output_0", "Shape12558", "BinaryOp12561", "BinaryOp12562", "Unsqueeze12563", "BinaryOp12565", "Unsqueeze12566", "StridedSlice12569", "BinaryOp12571", "BinaryOp12572", "/Gather_483_output_0", "/Unsqueeze_1024_output_0", "Unsqueeze12580", "Unsqueeze12583", "StridedSlice12586", "Squeeze12587", "BinaryOp12588", "BinaryOp12589", "/Gather_484_output_0", "/Unsqueeze_1025_output_0", "/Concat_416_output_0", "/Reshape_289_output_0", "/q_norm_32/Cast_output_0", "/q_norm_32/Mul_1_output_0", "/Mul_256_output_0", "/Shape_450_output_0", "Shape12708", "BinaryOp12711", "BinaryOp12712", "Unsqueeze12713", "BinaryOp12715", "Unsqueeze12716", "StridedSlice12719", "Squeeze12720", "BinaryOp12721", "BinaryOp12722", "/Gather_485_output_0", "/Div_96_output_0", "/Unsqueeze_1031_output_0", "/Slice_129_output_0", "/Neg_64_output_0", "/Unsqueeze_1030_output_0", "/Slice_128_output_0", "/Concat_419_output_0", "/Mul_257_output_0", "/Add_160_output_0", "/FakeLinear_129_output_0", "/Unsqueeze_1026_output_0", "/Unsqueeze_1027_output_0", "/Concat_417_output_0", "/Reshape_290_output_0", "/k_norm_32/Cast_output_0", "/k_norm_32/Mul_1_output_0", "/Mul_258_output_0", "/Shape_451_output_0", "Shape12604", "BinaryOp12607", "BinaryOp12608", "Unsqueeze12609", "BinaryOp12611", "Unsqueeze12612", "StridedSlice12615", "Squeeze12616", "BinaryOp12617", "BinaryOp12618", "/Gather_486_output_0", "/Div_97_output_0", "/Unsqueeze_1033_output_0", "/Slice_131_output_0", "/Neg_65_output_0", "/Unsqueeze_1032_output_0", "/Slice_130_output_0", "/Concat_420_output_0", "/Mul_259_output_0", "/Add_161_output_0", "/FakeLinear_130_output_0", "/Unsqueeze_1028_output_0", "/Unsqueeze_1029_output_0", "/Concat_418_output_0", "/Reshape_291_output_0", "/Reshape_296_output_0", "/FakeLinear_131_output_0", "/Add_163_output_0", "/post_attention_layernorm_32/Mul_1_output_0", "/mlp/gate_proj_32/FakeLinear_output_0", "/mlp/act_fn_32/Mul_output_0", "/mlp/up_proj_32/FakeLinear_output_0", "/mlp_32/Mul_output_0", "/mlp/down_proj_32/FakeLinear_output_0", "/Add_164_output_0", "/Reshape_297_output_0", "/input_layernorm_33/Mul_1_output_0", "/FakeLinear_132_output_0", "/Shape_462_output_0", "Shape12947", "BinaryOp12950", "BinaryOp12951", "Unsqueeze12952", "BinaryOp12954", "Unsqueeze12955", "StridedSlice12958", "BinaryOp12960", "BinaryOp12961", "/Gather_498_output_0", "/Unsqueeze_1056_output_0", "Unsqueeze12969", "Unsqueeze12972", "StridedSlice12975", "Squeeze12976", "BinaryOp12977", "BinaryOp12978", "/Gather_499_output_0", "/Unsqueeze_1057_output_0", "/Concat_429_output_0", "/Reshape_298_output_0", "/q_norm_33/Cast_output_0", "/q_norm_33/Mul_1_output_0", "/Mul_264_output_0", "/Shape_464_output_0", "Shape13097", "BinaryOp13100", "BinaryOp13101", "Unsqueeze13102", "BinaryOp13104", "Unsqueeze13105", "StridedSlice13108", "Squeeze13109", "BinaryOp13110", "BinaryOp13111", "/Gather_500_output_0", "/Div_99_output_0", "/Unsqueeze_1063_output_0", "/Slice_133_output_0", "/Neg_66_output_0", "/Unsqueeze_1062_output_0", "/Slice_132_output_0", "/Concat_432_output_0", "/Mul_265_output_0", "/Add_165_output_0", "/FakeLinear_133_output_0", "/Unsqueeze_1058_output_0", "/Unsqueeze_1059_output_0", "/Concat_430_output_0", "/Reshape_299_output_0", "/k_norm_33/Cast_output_0", "/k_norm_33/Mul_1_output_0", "/Mul_266_output_0", "/Shape_465_output_0", "Shape12993", "BinaryOp12996", "BinaryOp12997", "Unsqueeze12998", "BinaryOp13000", "Unsqueeze13001", "StridedSlice13004", "Squeeze13005", "BinaryOp13006", "BinaryOp13007", "/Gather_501_output_0", "/Div_100_output_0", "/Unsqueeze_1065_output_0", "/Slice_135_output_0", "/Neg_67_output_0", "/Unsqueeze_1064_output_0", "/Slice_134_output_0", "/Concat_433_output_0", "/Mul_267_output_0", "/Add_166_output_0", "/FakeLinear_134_output_0", "/Unsqueeze_1060_output_0", "/Unsqueeze_1061_output_0", "/Concat_431_output_0", "/Reshape_300_output_0", "/Reshape_305_output_0", "/FakeLinear_135_output_0", "/Add_168_output_0", "/post_attention_layernorm_33/Mul_1_output_0", "/mlp/gate_proj_33/FakeLinear_output_0", "/mlp/act_fn_33/Mul_output_0", "/mlp/up_proj_33/FakeLinear_output_0", "/mlp_33/Mul_output_0", "/mlp/down_proj_33/FakeLinear_output_0", "/Add_169_output_0", "/Reshape_306_output_0", "/input_layernorm_34/Mul_1_output_0", "/FakeLinear_136_output_0", "/Shape_476_output_0", "Shape13336", "BinaryOp13339", "BinaryOp13340", "Unsqueeze13341", "BinaryOp13343", "Unsqueeze13344", "StridedSlice13347", "BinaryOp13349", "BinaryOp13350", "/Gather_513_output_0", "/Unsqueeze_1088_output_0", "Unsqueeze13358", "Unsqueeze13361", "StridedSlice13364", "Squeeze13365", "BinaryOp13366", "BinaryOp13367", "/Gather_514_output_0", "/Unsqueeze_1089_output_0", "/Concat_442_output_0", "/Reshape_307_output_0", "/q_norm_34/Cast_output_0", "/q_norm_34/Mul_1_output_0", "/Mul_272_output_0", "/Shape_478_output_0", "Shape13486", "BinaryOp13489", "BinaryOp13490", "Unsqueeze13491", "BinaryOp13493", "Unsqueeze13494", "StridedSlice13497", "Squeeze13498", "BinaryOp13499", "BinaryOp13500", "/Gather_515_output_0", "/Div_102_output_0", "/Unsqueeze_1095_output_0", "/Slice_137_output_0", "/Neg_68_output_0", "/Unsqueeze_1094_output_0", "/Slice_136_output_0", "/Concat_445_output_0", "/Mul_273_output_0", "/Add_170_output_0", "/FakeLinear_137_output_0", "/Unsqueeze_1090_output_0", "/Unsqueeze_1091_output_0", "/Concat_443_output_0", "/Reshape_308_output_0", "/k_norm_34/Cast_output_0", "/k_norm_34/Mul_1_output_0", "/Mul_274_output_0", "/Shape_479_output_0", "Shape13382", "BinaryOp13385", "BinaryOp13386", "Unsqueeze13387", "BinaryOp13389", "Unsqueeze13390", "StridedSlice13393", "Squeeze13394", "BinaryOp13395", "BinaryOp13396", "/Gather_516_output_0", "/Div_103_output_0", "/Unsqueeze_1097_output_0", "/Slice_139_output_0", "/Neg_69_output_0", "/Unsqueeze_1096_output_0", "/Slice_138_output_0", "/Concat_446_output_0", "/Mul_275_output_0", "/Add_171_output_0", "/FakeLinear_138_output_0", "/Unsqueeze_1092_output_0", "/Unsqueeze_1093_output_0", "/Concat_444_output_0", "/Reshape_309_output_0", "/Reshape_314_output_0", "/FakeLinear_139_output_0", "/Add_173_output_0", "/post_attention_layernorm_34/Mul_1_output_0", "/mlp/gate_proj_34/FakeLinear_output_0", "/mlp/act_fn_34/Mul_output_0", "/mlp/up_proj_34/FakeLinear_output_0", "/mlp_34/Mul_output_0", "/mlp/down_proj_34/FakeLinear_output_0", "/Add_174_output_0", "/Reshape_315_output_0", "/input_layernorm_35/Mul_1_output_0", "/FakeLinear_140_output_0", "/Shape_490_output_0", "Shape13725", "BinaryOp13728", "BinaryOp13729", "Unsqueeze13730", "BinaryOp13732", "Unsqueeze13733", "StridedSlice13736", "BinaryOp13738", "BinaryOp13739", "/Gather_528_output_0", "/Unsqueeze_1120_output_0", "Unsqueeze13747", "Unsqueeze13750", "StridedSlice13753", "Squeeze13754", "BinaryOp13755", "BinaryOp13756", "/Gather_529_output_0", "/Unsqueeze_1121_output_0", "/Concat_455_output_0", "/Reshape_316_output_0", "/q_norm_35/Cast_output_0", "/q_norm_35/Mul_1_output_0", "/Mul_280_output_0", "/Shape_492_output_0", "Shape13875", "BinaryOp13878", "BinaryOp13879", "Unsqueeze13880", "BinaryOp13882", "Unsqueeze13883", "StridedSlice13886", "Squeeze13887", "BinaryOp13888", "BinaryOp13889", "/Gather_530_output_0", "/Div_105_output_0", "/Unsqueeze_1127_output_0", "/Slice_141_output_0", "/Neg_70_output_0", "/Unsqueeze_1126_output_0", "/Slice_140_output_0", "/Concat_458_output_0", "/Mul_281_output_0", "/Add_175_output_0", "/FakeLinear_141_output_0", "/Unsqueeze_1122_output_0", "/Unsqueeze_1123_output_0", "/Concat_456_output_0", "/Reshape_317_output_0", "/k_norm_35/Cast_output_0", "/k_norm_35/Mul_1_output_0", "/Mul_282_output_0", "/Shape_493_output_0", "Shape13771", "BinaryOp13774", "BinaryOp13775", "Unsqueeze13776", "BinaryOp13778", "Unsqueeze13779", "StridedSlice13782", "Squeeze13783", "BinaryOp13784", "BinaryOp13785", "/Gather_531_output_0", "/Div_106_output_0", "/Unsqueeze_1129_output_0", "/Slice_143_output_0", "/Neg_71_output_0", "/Unsqueeze_1128_output_0", "/Slice_142_output_0", "/Concat_459_output_0", "/Mul_283_output_0", "/Add_176_output_0", "/FakeLinear_142_output_0", "/Unsqueeze_1124_output_0", "/Unsqueeze_1125_output_0", "/Concat_457_output_0", "/Reshape_318_output_0", "/Reshape_323_output_0", "/FakeLinear_143_output_0", "/Add_178_output_0", "/post_attention_layernorm_35/Mul_1_output_0", "/mlp/gate_proj_35/FakeLinear_output_0", "/mlp/act_fn_35/Mul_output_0", "/mlp/up_proj_35/FakeLinear_output_0", "/mlp_35/Mul_output_0", "/mlp/down_proj_35/FakeLinear_output_0", "/Add_179_output_0", "/Reshape_324_output_0", "/input_layernorm_36/Mul_1_output_0", "/FakeLinear_144_output_0", "/Shape_504_output_0", "Shape14114", "BinaryOp14117", "BinaryOp14118", "Unsqueeze14119", "BinaryOp14121", "Unsqueeze14122", "StridedSlice14125", "BinaryOp14127", "BinaryOp14128", "/Gather_543_output_0", "/Unsqueeze_1152_output_0", "Unsqueeze14136", "Unsqueeze14139", "StridedSlice14142", "Squeeze14143", "BinaryOp14144", "BinaryOp14145", "/Gather_544_output_0", "/Unsqueeze_1153_output_0", "/Concat_468_output_0", "/Reshape_325_output_0", "/q_norm_36/Cast_output_0", "/q_norm_36/Mul_1_output_0", "/Mul_288_output_0", "/Shape_506_output_0", "Shape14264", "BinaryOp14267", "BinaryOp14268", "Unsqueeze14269", "BinaryOp14271", "Unsqueeze14272", "StridedSlice14275", "Squeeze14276", "BinaryOp14277", "BinaryOp14278", "/Gather_545_output_0", "/Div_108_output_0", "/Unsqueeze_1159_output_0", "/Slice_145_output_0", "/Neg_72_output_0", "/Unsqueeze_1158_output_0", "/Slice_144_output_0", "/Concat_471_output_0", "/Mul_289_output_0", "/Add_180_output_0", "/FakeLinear_145_output_0", "/Unsqueeze_1154_output_0", "/Unsqueeze_1155_output_0", "/Concat_469_output_0", "/Reshape_326_output_0", "/k_norm_36/Cast_output_0", "/k_norm_36/Mul_1_output_0", "/Mul_290_output_0", "/Shape_507_output_0", "Shape14160", "BinaryOp14163", "BinaryOp14164", "Unsqueeze14165", "BinaryOp14167", "Unsqueeze14168", "StridedSlice14171", "Squeeze14172", "BinaryOp14173", "BinaryOp14174", "/Gather_546_output_0", "/Div_109_output_0", "/Unsqueeze_1161_output_0", "/Slice_147_output_0", "/Neg_73_output_0", "/Unsqueeze_1160_output_0", "/Slice_146_output_0", "/Concat_472_output_0", "/Mul_291_output_0", "/Add_181_output_0", "/FakeLinear_146_output_0", "/Unsqueeze_1156_output_0", "/Unsqueeze_1157_output_0", "/Concat_470_output_0", "/Reshape_327_output_0", "/Reshape_332_output_0", "/FakeLinear_147_output_0", "/Add_183_output_0", "/post_attention_layernorm_36/Mul_1_output_0", "/mlp/gate_proj_36/FakeLinear_output_0", "/mlp/act_fn_36/Mul_output_0", "/mlp/up_proj_36/FakeLinear_output_0", "/mlp_36/Mul_output_0", "/mlp/down_proj_36/FakeLinear_output_0", "/Add_184_output_0", "/Reshape_333_output_0", "/input_layernorm_37/Mul_1_output_0", "/FakeLinear_148_output_0", "/Shape_518_output_0", "Shape14503", "BinaryOp14506", "BinaryOp14507", "Unsqueeze14508", "BinaryOp14510", "Unsqueeze14511", "StridedSlice14514", "BinaryOp14516", "BinaryOp14517", "/Gather_558_output_0", "/Unsqueeze_1184_output_0", "Unsqueeze14525", "Unsqueeze14528", "StridedSlice14531", "Squeeze14532", "BinaryOp14533", "BinaryOp14534", "/Gather_559_output_0", "/Unsqueeze_1185_output_0", "/Concat_481_output_0", "/Reshape_334_output_0", "/q_norm_37/Cast_output_0", "/q_norm_37/Mul_1_output_0", "/Mul_296_output_0", "/Shape_520_output_0", "Shape14653", "BinaryOp14656", "BinaryOp14657", "Unsqueeze14658", "BinaryOp14660", "Unsqueeze14661", "StridedSlice14664", "Squeeze14665", "BinaryOp14666", "BinaryOp14667", "/Gather_560_output_0", "/Div_111_output_0", "/Unsqueeze_1191_output_0", "/Slice_149_output_0", "/Neg_74_output_0", "/Unsqueeze_1190_output_0", "/Slice_148_output_0", "/Concat_484_output_0", "/Mul_297_output_0", "/Add_185_output_0", "/FakeLinear_149_output_0", "/Unsqueeze_1186_output_0", "/Unsqueeze_1187_output_0", "/Concat_482_output_0", "/Reshape_335_output_0", "/k_norm_37/Cast_output_0", "/k_norm_37/Mul_1_output_0", "/Mul_298_output_0", "/Shape_521_output_0", "Shape14549", "BinaryOp14552", "BinaryOp14553", "Unsqueeze14554", "BinaryOp14556", "Unsqueeze14557", "StridedSlice14560", "Squeeze14561", "BinaryOp14562", "BinaryOp14563", "/Gather_561_output_0", "/Div_112_output_0", "/Unsqueeze_1193_output_0", "/Slice_151_output_0", "/Neg_75_output_0", "/Unsqueeze_1192_output_0", "/Slice_150_output_0", "/Concat_485_output_0", "/Mul_299_output_0", "/Add_186_output_0", "/FakeLinear_150_output_0", "/Unsqueeze_1188_output_0", "/Unsqueeze_1189_output_0", "/Concat_483_output_0", "/Reshape_336_output_0", "/Reshape_341_output_0", "/FakeLinear_151_output_0", "/Add_188_output_0", "/post_attention_layernorm_37/Mul_1_output_0", "/mlp/gate_proj_37/FakeLinear_output_0", "/mlp/act_fn_37/Mul_output_0", "/mlp/up_proj_37/FakeLinear_output_0", "/mlp_37/Mul_output_0", "/mlp/down_proj_37/FakeLinear_output_0", "/Add_189_output_0", "/Reshape_342_output_0", "/input_layernorm_38/Mul_1_output_0", "/FakeLinear_152_output_0", "/Shape_532_output_0", "Shape14892", "BinaryOp14895", "BinaryOp14896", "Unsqueeze14897", "BinaryOp14899", "Unsqueeze14900", "StridedSlice14903", "BinaryOp14905", "BinaryOp14906", "/Gather_573_output_0", "/Unsqueeze_1216_output_0", "Unsqueeze14914", "Unsqueeze14917", "StridedSlice14920", "Squeeze14921", "BinaryOp14922", "BinaryOp14923", "/Gather_574_output_0", "/Unsqueeze_1217_output_0", "/Concat_494_output_0", "/Reshape_343_output_0", "/q_norm_38/Cast_output_0", "/q_norm_38/Mul_1_output_0", "/Mul_304_output_0", "/Shape_534_output_0", "Shape15042", "BinaryOp15045", "BinaryOp15046", "Unsqueeze15047", "BinaryOp15049", "Unsqueeze15050", "StridedSlice15053", "Squeeze15054", "BinaryOp15055", "BinaryOp15056", "/Gather_575_output_0", "/Div_114_output_0", "/Unsqueeze_1223_output_0", "/Slice_153_output_0", "/Neg_76_output_0", "/Unsqueeze_1222_output_0", "/Slice_152_output_0", "/Concat_497_output_0", "/Mul_305_output_0", "/Add_190_output_0", "/FakeLinear_153_output_0", "/Unsqueeze_1218_output_0", "/Unsqueeze_1219_output_0", "/Concat_495_output_0", "/Reshape_344_output_0", "/k_norm_38/Cast_output_0", "/k_norm_38/Mul_1_output_0", "/Mul_306_output_0", "/Shape_535_output_0", "Shape14938", "BinaryOp14941", "BinaryOp14942", "Unsqueeze14943", "BinaryOp14945", "Unsqueeze14946", "StridedSlice14949", "Squeeze14950", "BinaryOp14951", "BinaryOp14952", "/Gather_576_output_0", "/Div_115_output_0", "/Unsqueeze_1225_output_0", "/Slice_155_output_0", "/Neg_77_output_0", "/Unsqueeze_1224_output_0", "/Slice_154_output_0", "/Concat_498_output_0", "/Mul_307_output_0", "/Add_191_output_0", "/FakeLinear_154_output_0", "/Unsqueeze_1220_output_0", "/Unsqueeze_1221_output_0", "/Concat_496_output_0", "/Reshape_345_output_0", "/Reshape_350_output_0", "/FakeLinear_155_output_0", "/Add_193_output_0", "/post_attention_layernorm_38/Mul_1_output_0", "/mlp/gate_proj_38/FakeLinear_output_0", "/mlp/act_fn_38/Mul_output_0", "/mlp/up_proj_38/FakeLinear_output_0", "/mlp_38/Mul_output_0", "/mlp/down_proj_38/FakeLinear_output_0", "/Add_194_output_0", "/Reshape_351_output_0", "/input_layernorm_39/Mul_1_output_0", "/FakeLinear_156_output_0", "/Shape_546_output_0", "Shape15281", "BinaryOp15284", "BinaryOp15285", "Unsqueeze15286", "BinaryOp15288", "Unsqueeze15289", "StridedSlice15292", "BinaryOp15294", "BinaryOp15295", "/Gather_588_output_0", "/Unsqueeze_1248_output_0", "Unsqueeze15303", "Unsqueeze15306", "StridedSlice15309", "Squeeze15310", "BinaryOp15311", "BinaryOp15312", "/Gather_589_output_0", "/Unsqueeze_1249_output_0", "/Concat_507_output_0", "/Reshape_352_output_0", "/q_norm_39/Cast_output_0", "/q_norm_39/Mul_1_output_0", "/Mul_312_output_0", "/Shape_548_output_0", "Shape15431", "BinaryOp15434", "BinaryOp15435", "Unsqueeze15436", "BinaryOp15438", "Unsqueeze15439", "StridedSlice15442", "Squeeze15443", "BinaryOp15444", "BinaryOp15445", "/Gather_590_output_0", "/Div_117_output_0", "/Unsqueeze_1255_output_0", "/Slice_157_output_0", "/Neg_78_output_0", "/Unsqueeze_1254_output_0", "/Slice_156_output_0", "/Concat_510_output_0", "/Mul_313_output_0", "/Add_195_output_0", "/FakeLinear_157_output_0", "/Unsqueeze_1250_output_0", "/Unsqueeze_1251_output_0", "/Concat_508_output_0", "/Reshape_353_output_0", "/k_norm_39/Cast_output_0", "/k_norm_39/Mul_1_output_0", "/Mul_314_output_0", "/Shape_549_output_0", "Shape15327", "BinaryOp15330", "BinaryOp15331", "Unsqueeze15332", "BinaryOp15334", "Unsqueeze15335", "StridedSlice15338", "Squeeze15339", "BinaryOp15340", "BinaryOp15341", "/Gather_591_output_0", "/Div_118_output_0", "/Unsqueeze_1257_output_0", "/Slice_159_output_0", "/Neg_79_output_0", "/Unsqueeze_1256_output_0", "/Slice_158_output_0", "/Concat_511_output_0", "/Mul_315_output_0", "/Add_196_output_0", "/FakeLinear_158_output_0", "/Unsqueeze_1252_output_0", "/Unsqueeze_1253_output_0", "/Concat_509_output_0", "/Reshape_354_output_0", "/Reshape_359_output_0", "/FakeLinear_159_output_0", "/Add_198_output_0", "/post_attention_layernorm_39/Mul_1_output_0", "/mlp/gate_proj_39/FakeLinear_output_0", "/mlp/act_fn_39/Mul_output_0", "/mlp/up_proj_39/FakeLinear_output_0", "/mlp_39/Mul_output_0", "/mlp/down_proj_39/FakeLinear_output_0", "/Add_199_output_0", "/Reshape_360_output_0", "/input_layernorm_40/Mul_1_output_0", "/FakeLinear_160_output_0", "/Shape_560_output_0", "Shape15670", "BinaryOp15673", "BinaryOp15674", "Unsqueeze15675", "BinaryOp15677", "Unsqueeze15678", "StridedSlice15681", "BinaryOp15683", "BinaryOp15684", "/Gather_603_output_0", "/Unsqueeze_1280_output_0", "Unsqueeze15692", "Unsqueeze15695", "StridedSlice15698", "Squeeze15699", "BinaryOp15700", "BinaryOp15701", "/Gather_604_output_0", "/Unsqueeze_1281_output_0", "/Concat_520_output_0", "/Reshape_361_output_0", "/q_norm_40/Cast_output_0", "/q_norm_40/Mul_1_output_0", "/Mul_320_output_0", "/Shape_562_output_0", "Shape15820", "BinaryOp15823", "BinaryOp15824", "Unsqueeze15825", "BinaryOp15827", "Unsqueeze15828", "StridedSlice15831", "Squeeze15832", "BinaryOp15833", "BinaryOp15834", "/Gather_605_output_0", "/Div_120_output_0", "/Unsqueeze_1287_output_0", "/Slice_161_output_0", "/Neg_80_output_0", "/Unsqueeze_1286_output_0", "/Slice_160_output_0", "/Concat_523_output_0", "/Mul_321_output_0", "/Add_200_output_0", "/FakeLinear_161_output_0", "/Unsqueeze_1282_output_0", "/Unsqueeze_1283_output_0", "/Concat_521_output_0", "/Reshape_362_output_0", "/k_norm_40/Cast_output_0", "/k_norm_40/Mul_1_output_0", "/Mul_322_output_0", "/Shape_563_output_0", "Shape15716", "BinaryOp15719", "BinaryOp15720", "Unsqueeze15721", "BinaryOp15723", "Unsqueeze15724", "StridedSlice15727", "Squeeze15728", "BinaryOp15729", "BinaryOp15730", "/Gather_606_output_0", "/Div_121_output_0", "/Unsqueeze_1289_output_0", "/Slice_163_output_0", "/Neg_81_output_0", "/Unsqueeze_1288_output_0", "/Slice_162_output_0", "/Concat_524_output_0", "/Mul_323_output_0", "/Add_201_output_0", "/FakeLinear_162_output_0", "/Unsqueeze_1284_output_0", "/Unsqueeze_1285_output_0", "/Concat_522_output_0", "/Reshape_363_output_0", "/Reshape_368_output_0", "/FakeLinear_163_output_0", "/Add_203_output_0", "/post_attention_layernorm_40/Mul_1_output_0", "/mlp/gate_proj_40/FakeLinear_output_0", "/mlp/act_fn_40/Mul_output_0", "/mlp/up_proj_40/FakeLinear_output_0", "/mlp_40/Mul_output_0", "/mlp/down_proj_40/FakeLinear_output_0", "/Add_204_output_0", "/Reshape_369_output_0", "/input_layernorm_41/Mul_1_output_0", "/FakeLinear_164_output_0", "/Shape_574_output_0", "Shape16059", "BinaryOp16062", "BinaryOp16063", "Unsqueeze16064", "BinaryOp16066", "Unsqueeze16067", "StridedSlice16070", "BinaryOp16072", "BinaryOp16073", "/Gather_618_output_0", "/Unsqueeze_1312_output_0", "Unsqueeze16081", "Unsqueeze16084", "StridedSlice16087", "Squeeze16088", "BinaryOp16089", "BinaryOp16090", "/Gather_619_output_0", "/Unsqueeze_1313_output_0", "/Concat_533_output_0", "/Reshape_370_output_0", "/q_norm_41/Cast_output_0", "/q_norm_41/Mul_1_output_0", "/Mul_328_output_0", "/Shape_576_output_0", "Shape16209", "BinaryOp16212", "BinaryOp16213", "Unsqueeze16214", "BinaryOp16216", "Unsqueeze16217", "StridedSlice16220", "Squeeze16221", "BinaryOp16222", "BinaryOp16223", "/Gather_620_output_0", "/Div_123_output_0", "/Unsqueeze_1319_output_0", "/Slice_165_output_0", "/Neg_82_output_0", "/Unsqueeze_1318_output_0", "/Slice_164_output_0", "/Concat_536_output_0", "/Mul_329_output_0", "/Add_205_output_0", "/FakeLinear_165_output_0", "/Unsqueeze_1314_output_0", "/Unsqueeze_1315_output_0", "/Concat_534_output_0", "/Reshape_371_output_0", "/k_norm_41/Cast_output_0", "/k_norm_41/Mul_1_output_0", "/Mul_330_output_0", "/Shape_577_output_0", "Shape16105", "BinaryOp16108", "BinaryOp16109", "Unsqueeze16110", "BinaryOp16112", "Unsqueeze16113", "StridedSlice16116", "Squeeze16117", "BinaryOp16118", "BinaryOp16119", "/Gather_621_output_0", "/Div_124_output_0", "/Unsqueeze_1321_output_0", "/Slice_167_output_0", "/Neg_83_output_0", "/Unsqueeze_1320_output_0", "/Slice_166_output_0", "/Concat_537_output_0", "/Mul_331_output_0", "/Add_206_output_0", "/FakeLinear_166_output_0", "/Unsqueeze_1316_output_0", "/Unsqueeze_1317_output_0", "/Concat_535_output_0", "/Reshape_372_output_0", "/Reshape_377_output_0", "/FakeLinear_167_output_0", "/Add_208_output_0", "/post_attention_layernorm_41/Mul_1_output_0", "/mlp/gate_proj_41/FakeLinear_output_0", "/mlp/act_fn_41/Mul_output_0", "/mlp/up_proj_41/FakeLinear_output_0", "/mlp_41/Mul_output_0", "/mlp/down_proj_41/FakeLinear_output_0", "/Add_209_output_0", "/Reshape_378_output_0", "/input_layernorm_42/Mul_1_output_0", "/FakeLinear_168_output_0", "/Shape_588_output_0", "Shape16448", "BinaryOp16451", "BinaryOp16452", "Unsqueeze16453", "BinaryOp16455", "Unsqueeze16456", "StridedSlice16459", "BinaryOp16461", "BinaryOp16462", "/Gather_633_output_0", "/Unsqueeze_1344_output_0", "Unsqueeze16470", "Unsqueeze16473", "StridedSlice16476", "Squeeze16477", "BinaryOp16478", "BinaryOp16479", "/Gather_634_output_0", "/Unsqueeze_1345_output_0", "/Concat_546_output_0", "/Reshape_379_output_0", "/q_norm_42/Cast_output_0", "/q_norm_42/Mul_1_output_0", "/Mul_336_output_0", "/Shape_590_output_0", "Shape16598", "BinaryOp16601", "BinaryOp16602", "Unsqueeze16603", "BinaryOp16605", "Unsqueeze16606", "StridedSlice16609", "Squeeze16610", "BinaryOp16611", "BinaryOp16612", "/Gather_635_output_0", "/Div_126_output_0", "/Unsqueeze_1351_output_0", "/Slice_169_output_0", "/Neg_84_output_0", "/Unsqueeze_1350_output_0", "/Slice_168_output_0", "/Concat_549_output_0", "/Mul_337_output_0", "/Add_210_output_0", "/FakeLinear_169_output_0", "/Unsqueeze_1346_output_0", "/Unsqueeze_1347_output_0", "/Concat_547_output_0", "/Reshape_380_output_0", "/k_norm_42/Cast_output_0", "/k_norm_42/Mul_1_output_0", "/Mul_338_output_0", "/Shape_591_output_0", "Shape16494", "BinaryOp16497", "BinaryOp16498", "Unsqueeze16499", "BinaryOp16501", "Unsqueeze16502", "StridedSlice16505", "Squeeze16506", "BinaryOp16507", "BinaryOp16508", "/Gather_636_output_0", "/Div_127_output_0", "/Unsqueeze_1353_output_0", "/Slice_171_output_0", "/Neg_85_output_0", "/Unsqueeze_1352_output_0", "/Slice_170_output_0", "/Concat_550_output_0", "/Mul_339_output_0", "/Add_211_output_0", "/FakeLinear_170_output_0", "/Unsqueeze_1348_output_0", "/Unsqueeze_1349_output_0", "/Concat_548_output_0", "/Reshape_381_output_0", "/Reshape_386_output_0", "/FakeLinear_171_output_0", "/Add_213_output_0", "/post_attention_layernorm_42/Mul_1_output_0", "/mlp/gate_proj_42/FakeLinear_output_0", "/mlp/act_fn_42/Mul_output_0", "/mlp/up_proj_42/FakeLinear_output_0", "/mlp_42/Mul_output_0", "/mlp/down_proj_42/FakeLinear_output_0", "/Add_214_output_0", "/Reshape_387_output_0", "/input_layernorm_43/Mul_1_output_0", "/FakeLinear_172_output_0", "/Shape_602_output_0", "Shape16837", "BinaryOp16840", "BinaryOp16841", "Unsqueeze16842", "BinaryOp16844", "Unsqueeze16845", "StridedSlice16848", "BinaryOp16850", "BinaryOp16851", "/Gather_648_output_0", "/Unsqueeze_1376_output_0", "Unsqueeze16859", "Unsqueeze16862", "StridedSlice16865", "Squeeze16866", "BinaryOp16867", "BinaryOp16868", "/Gather_649_output_0", "/Unsqueeze_1377_output_0", "/Concat_559_output_0", "/Reshape_388_output_0", "/q_norm_43/Cast_output_0", "/q_norm_43/Mul_1_output_0", "/Mul_344_output_0", "/Shape_604_output_0", "Shape16987", "BinaryOp16990", "BinaryOp16991", "Unsqueeze16992", "BinaryOp16994", "Unsqueeze16995", "StridedSlice16998", "Squeeze16999", "BinaryOp17000", "BinaryOp17001", "/Gather_650_output_0", "/Div_129_output_0", "/Unsqueeze_1383_output_0", "/Slice_173_output_0", "/Neg_86_output_0", "/Unsqueeze_1382_output_0", "/Slice_172_output_0", "/Concat_562_output_0", "/Mul_345_output_0", "/Add_215_output_0", "/FakeLinear_173_output_0", "/Unsqueeze_1378_output_0", "/Unsqueeze_1379_output_0", "/Concat_560_output_0", "/Reshape_389_output_0", "/k_norm_43/Cast_output_0", "/k_norm_43/Mul_1_output_0", "/Mul_346_output_0", "/Shape_605_output_0", "Shape16883", "BinaryOp16886", "BinaryOp16887", "Unsqueeze16888", "BinaryOp16890", "Unsqueeze16891", "StridedSlice16894", "Squeeze16895", "BinaryOp16896", "BinaryOp16897", "/Gather_651_output_0", "/Div_130_output_0", "/Unsqueeze_1385_output_0", "/Slice_175_output_0", "/Neg_87_output_0", "/Unsqueeze_1384_output_0", "/Slice_174_output_0", "/Concat_563_output_0", "/Mul_347_output_0", "/Add_216_output_0", "/FakeLinear_174_output_0", "/Unsqueeze_1380_output_0", "/Unsqueeze_1381_output_0", "/Concat_561_output_0", "/Reshape_390_output_0", "/Reshape_395_output_0", "/FakeLinear_175_output_0", "/Add_218_output_0", "/post_attention_layernorm_43/Mul_1_output_0", "/mlp/gate_proj_43/FakeLinear_output_0", "/mlp/act_fn_43/Mul_output_0", "/mlp/up_proj_43/FakeLinear_output_0", "/mlp_43/Mul_output_0", "/mlp/down_proj_43/FakeLinear_output_0", "/Add_219_output_0", "/Reshape_396_output_0", "/input_layernorm_44/Mul_1_output_0", "/FakeLinear_176_output_0", "/Shape_616_output_0", "Shape17226", "BinaryOp17229", "BinaryOp17230", "Unsqueeze17231", "BinaryOp17233", "Unsqueeze17234", "StridedSlice17237", "BinaryOp17239", "BinaryOp17240", "/Gather_663_output_0", "/Unsqueeze_1408_output_0", "Unsqueeze17248", "Unsqueeze17251", "StridedSlice17254", "Squeeze17255", "BinaryOp17256", "BinaryOp17257", "/Gather_664_output_0", "/Unsqueeze_1409_output_0", "/Concat_572_output_0", "/Reshape_397_output_0", "/q_norm_44/Cast_output_0", "/q_norm_44/Mul_1_output_0", "/Mul_352_output_0", "/Shape_618_output_0", "Shape17376", "BinaryOp17379", "BinaryOp17380", "Unsqueeze17381", "BinaryOp17383", "Unsqueeze17384", "StridedSlice17387", "Squeeze17388", "BinaryOp17389", "BinaryOp17390", "/Gather_665_output_0", "/Div_132_output_0", "/Unsqueeze_1415_output_0", "/Slice_177_output_0", "/Neg_88_output_0", "/Unsqueeze_1414_output_0", "/Slice_176_output_0", "/Concat_575_output_0", "/Mul_353_output_0", "/Add_220_output_0", "/FakeLinear_177_output_0", "/Unsqueeze_1410_output_0", "/Unsqueeze_1411_output_0", "/Concat_573_output_0", "/Reshape_398_output_0", "/k_norm_44/Cast_output_0", "/k_norm_44/Mul_1_output_0", "/Mul_354_output_0", "/Shape_619_output_0", "Shape17272", "BinaryOp17275", "BinaryOp17276", "Unsqueeze17277", "BinaryOp17279", "Unsqueeze17280", "StridedSlice17283", "Squeeze17284", "BinaryOp17285", "BinaryOp17286", "/Gather_666_output_0", "/Div_133_output_0", "/Unsqueeze_1417_output_0", "/Slice_179_output_0", "/Neg_89_output_0", "/Unsqueeze_1416_output_0", "/Slice_178_output_0", "/Concat_576_output_0", "/Mul_355_output_0", "/Add_221_output_0", "/FakeLinear_178_output_0", "/Unsqueeze_1412_output_0", "/Unsqueeze_1413_output_0", "/Concat_574_output_0", "/Reshape_399_output_0", "/Reshape_404_output_0", "/FakeLinear_179_output_0", "/Add_223_output_0", "/post_attention_layernorm_44/Mul_1_output_0", "/mlp/gate_proj_44/FakeLinear_output_0", "/mlp/act_fn_44/Mul_output_0", "/mlp/up_proj_44/FakeLinear_output_0", "/mlp_44/Mul_output_0", "/mlp/down_proj_44/FakeLinear_output_0", "/Add_224_output_0", "/Reshape_405_output_0", "/input_layernorm_45/Mul_1_output_0", "/FakeLinear_180_output_0", "/Shape_630_output_0", "Shape17615", "BinaryOp17618", "BinaryOp17619", "Unsqueeze17620", "BinaryOp17622", "Unsqueeze17623", "StridedSlice17626", "BinaryOp17628", "BinaryOp17629", "/Gather_678_output_0", "/Unsqueeze_1440_output_0", "Unsqueeze17637", "Unsqueeze17640", "StridedSlice17643", "Squeeze17644", "BinaryOp17645", "BinaryOp17646", "/Gather_679_output_0", "/Unsqueeze_1441_output_0", "/Concat_585_output_0", "/Reshape_406_output_0", "/q_norm_45/Cast_output_0", "/q_norm_45/Mul_1_output_0", "/Mul_360_output_0", "/Shape_632_output_0", "Shape17765", "BinaryOp17768", "BinaryOp17769", "Unsqueeze17770", "BinaryOp17772", "Unsqueeze17773", "StridedSlice17776", "Squeeze17777", "BinaryOp17778", "BinaryOp17779", "/Gather_680_output_0", "/Div_135_output_0", "/Unsqueeze_1447_output_0", "/Slice_181_output_0", "/Neg_90_output_0", "/Unsqueeze_1446_output_0", "/Slice_180_output_0", "/Concat_588_output_0", "/Mul_361_output_0", "/Add_225_output_0", "/FakeLinear_181_output_0", "/Unsqueeze_1442_output_0", "/Unsqueeze_1443_output_0", "/Concat_586_output_0", "/Reshape_407_output_0", "/k_norm_45/Cast_output_0", "/k_norm_45/Mul_1_output_0", "/Mul_362_output_0", "/Shape_633_output_0", "Shape17661", "BinaryOp17664", "BinaryOp17665", "Unsqueeze17666", "BinaryOp17668", "Unsqueeze17669", "StridedSlice17672", "Squeeze17673", "BinaryOp17674", "BinaryOp17675", "/Gather_681_output_0", "/Div_136_output_0", "/Unsqueeze_1449_output_0", "/Slice_183_output_0", "/Neg_91_output_0", "/Unsqueeze_1448_output_0", "/Slice_182_output_0", "/Concat_589_output_0", "/Mul_363_output_0", "/Add_226_output_0", "/FakeLinear_182_output_0", "/Unsqueeze_1444_output_0", "/Unsqueeze_1445_output_0", "/Concat_587_output_0", "/Reshape_408_output_0", "/Reshape_413_output_0", "/FakeLinear_183_output_0", "/Add_228_output_0", "/post_attention_layernorm_45/Mul_1_output_0", "/mlp/gate_proj_45/FakeLinear_output_0", "/mlp/act_fn_45/Mul_output_0", "/mlp/up_proj_45/FakeLinear_output_0", "/mlp_45/Mul_output_0", "/mlp/down_proj_45/FakeLinear_output_0", "/Add_229_output_0", "/Reshape_414_output_0", "/input_layernorm_46/Mul_1_output_0", "/FakeLinear_184_output_0", "/Shape_644_output_0", "Shape18004", "BinaryOp18007", "BinaryOp18008", "Unsqueeze18009", "BinaryOp18011", "Unsqueeze18012", "StridedSlice18015", "BinaryOp18017", "BinaryOp18018", "/Gather_693_output_0", "/Unsqueeze_1472_output_0", "Unsqueeze18026", "Unsqueeze18029", "StridedSlice18032", "Squeeze18033", "BinaryOp18034", "BinaryOp18035", "/Gather_694_output_0", "/Unsqueeze_1473_output_0", "/Concat_598_output_0", "/Reshape_415_output_0", "/q_norm_46/Cast_output_0", "/q_norm_46/Mul_1_output_0", "/Mul_368_output_0", "/Shape_646_output_0", "Shape18154", "BinaryOp18157", "BinaryOp18158", "Unsqueeze18159", "BinaryOp18161", "Unsqueeze18162", "StridedSlice18165", "Squeeze18166", "BinaryOp18167", "BinaryOp18168", "/Gather_695_output_0", "/Div_138_output_0", "/Unsqueeze_1479_output_0", "/Slice_185_output_0", "/Neg_92_output_0", "/Unsqueeze_1478_output_0", "/Slice_184_output_0", "/Concat_601_output_0", "/Mul_369_output_0", "/Add_230_output_0", "/FakeLinear_185_output_0", "/Unsqueeze_1474_output_0", "/Unsqueeze_1475_output_0", "/Concat_599_output_0", "/Reshape_416_output_0", "/k_norm_46/Cast_output_0", "/k_norm_46/Mul_1_output_0", "/Mul_370_output_0", "/Shape_647_output_0", "Shape18050", "BinaryOp18053", "BinaryOp18054", "Unsqueeze18055", "BinaryOp18057", "Unsqueeze18058", "StridedSlice18061", "Squeeze18062", "BinaryOp18063", "BinaryOp18064", "/Gather_696_output_0", "/Div_139_output_0", "/Unsqueeze_1481_output_0", "/Slice_187_output_0", "/Neg_93_output_0", "/Unsqueeze_1480_output_0", "/Slice_186_output_0", "/Concat_602_output_0", "/Mul_371_output_0", "/Add_231_output_0", "/FakeLinear_186_output_0", "/Unsqueeze_1476_output_0", "/Unsqueeze_1477_output_0", "/Concat_600_output_0", "/Reshape_417_output_0", "/Reshape_422_output_0", "/FakeLinear_187_output_0", "/Add_233_output_0", "/post_attention_layernorm_46/Mul_1_output_0", "/mlp/gate_proj_46/FakeLinear_output_0", "/mlp/act_fn_46/Mul_output_0", "/mlp/up_proj_46/FakeLinear_output_0", "/mlp_46/Mul_output_0", "/mlp/down_proj_46/FakeLinear_output_0", "/Add_234_output_0", "/Reshape_423_output_0", "/input_layernorm_47/Mul_1_output_0", "/FakeLinear_188_output_0", "/Shape_658_output_0", "Shape18393", "BinaryOp18396", "BinaryOp18397", "Unsqueeze18398", "BinaryOp18400", "Unsqueeze18401", "StridedSlice18404", "BinaryOp18406", "BinaryOp18407", "/Gather_708_output_0", "/Unsqueeze_1504_output_0", "Unsqueeze18415", "Unsqueeze18418", "StridedSlice18421", "Squeeze18422", "BinaryOp18423", "BinaryOp18424", "/Gather_709_output_0", "/Unsqueeze_1505_output_0", "/Concat_611_output_0", "/Reshape_424_output_0", "/q_norm_47/Cast_output_0", "/q_norm_47/Mul_1_output_0", "/Mul_376_output_0", "/Shape_660_output_0", "Shape18543", "BinaryOp18546", "BinaryOp18547", "Unsqueeze18548", "BinaryOp18550", "Unsqueeze18551", "StridedSlice18554", "Squeeze18555", "BinaryOp18556", "BinaryOp18557", "/Gather_710_output_0", "/Div_141_output_0", "/Unsqueeze_1511_output_0", "/Slice_189_output_0", "/Neg_94_output_0", "/Unsqueeze_1510_output_0", "/Slice_188_output_0", "/Concat_614_output_0", "/Mul_377_output_0", "/Add_235_output_0", "/FakeLinear_189_output_0", "/Unsqueeze_1506_output_0", "/Unsqueeze_1507_output_0", "/Concat_612_output_0", "/Reshape_425_output_0", "/k_norm_47/Cast_output_0", "/k_norm_47/Mul_1_output_0", "/Mul_378_output_0", "/Shape_661_output_0", "Shape18439", "BinaryOp18442", "BinaryOp18443", "Unsqueeze18444", "BinaryOp18446", "Unsqueeze18447", "StridedSlice18450", "Squeeze18451", "BinaryOp18452", "BinaryOp18453", "/Gather_711_output_0", "/Div_142_output_0", "/Unsqueeze_1513_output_0", "/Slice_191_output_0", "/Neg_95_output_0", "/Unsqueeze_1512_output_0", "/Slice_190_output_0", "/Concat_615_output_0", "/Mul_379_output_0", "/Add_236_output_0", "/FakeLinear_190_output_0", "/Unsqueeze_1508_output_0", "/Unsqueeze_1509_output_0", "/Concat_613_output_0", "/Reshape_426_output_0", "/Reshape_431_output_0", "/FakeLinear_191_output_0", "/Add_238_output_0", "/post_attention_layernorm_47/Mul_1_output_0", "/mlp/gate_proj_47/FakeLinear_output_0", "/mlp/act_fn_47/Mul_output_0", "/mlp/up_proj_47/FakeLinear_output_0", "/mlp_47/Mul_output_0", "/mlp/down_proj_47/FakeLinear_output_0", "/Add_239_output_0", "/Reshape_432_output_0", "/input_layernorm_48/Mul_1_output_0", "/FakeLinear_192_output_0", "/Shape_672_output_0", "Shape18782", "BinaryOp18785", "BinaryOp18786", "Unsqueeze18787", "BinaryOp18789", "Unsqueeze18790", "StridedSlice18793", "BinaryOp18795", "BinaryOp18796", "/Gather_723_output_0", "/Unsqueeze_1536_output_0", "Unsqueeze18804", "Unsqueeze18807", "StridedSlice18810", "Squeeze18811", "BinaryOp18812", "BinaryOp18813", "/Gather_724_output_0", "/Unsqueeze_1537_output_0", "/Concat_624_output_0", "/Reshape_433_output_0", "/q_norm_48/Cast_output_0", "/q_norm_48/Mul_1_output_0", "/Mul_384_output_0", "/Shape_674_output_0", "Shape18932", "BinaryOp18935", "BinaryOp18936", "Unsqueeze18937", "BinaryOp18939", "Unsqueeze18940", "StridedSlice18943", "Squeeze18944", "BinaryOp18945", "BinaryOp18946", "/Gather_725_output_0", "/Div_144_output_0", "/Unsqueeze_1543_output_0", "/Slice_193_output_0", "/Neg_96_output_0", "/Unsqueeze_1542_output_0", "/Slice_192_output_0", "/Concat_627_output_0", "/Mul_385_output_0", "/Add_240_output_0", "/FakeLinear_193_output_0", "/Unsqueeze_1538_output_0", "/Unsqueeze_1539_output_0", "/Concat_625_output_0", "/Reshape_434_output_0", "/k_norm_48/Cast_output_0", "/k_norm_48/Mul_1_output_0", "/Mul_386_output_0", "/Shape_675_output_0", "Shape18828", "BinaryOp18831", "BinaryOp18832", "Unsqueeze18833", "BinaryOp18835", "Unsqueeze18836", "StridedSlice18839", "Squeeze18840", "BinaryOp18841", "BinaryOp18842", "/Gather_726_output_0", "/Div_145_output_0", "/Unsqueeze_1545_output_0", "/Slice_195_output_0", "/Neg_97_output_0", "/Unsqueeze_1544_output_0", "/Slice_194_output_0", "/Concat_628_output_0", "/Mul_387_output_0", "/Add_241_output_0", "/FakeLinear_194_output_0", "/Unsqueeze_1540_output_0", "/Unsqueeze_1541_output_0", "/Concat_626_output_0", "/Reshape_435_output_0", "/Reshape_440_output_0", "/FakeLinear_195_output_0", "/Add_243_output_0", "/post_attention_layernorm_48/Mul_1_output_0", "/mlp/gate_proj_48/FakeLinear_output_0", "/mlp/act_fn_48/Mul_output_0", "/mlp/up_proj_48/FakeLinear_output_0", "/mlp_48/Mul_output_0", "/mlp/down_proj_48/FakeLinear_output_0", "/Add_244_output_0", "/Reshape_441_output_0", "/input_layernorm_49/Mul_1_output_0", "/FakeLinear_196_output_0", "/Shape_686_output_0", "Shape19171", "BinaryOp19174", "BinaryOp19175", "Unsqueeze19176", "BinaryOp19178", "Unsqueeze19179", "StridedSlice19182", "BinaryOp19184", "BinaryOp19185", "/Gather_738_output_0", "/Unsqueeze_1568_output_0", "Unsqueeze19193", "Unsqueeze19196", "StridedSlice19199", "Squeeze19200", "BinaryOp19201", "BinaryOp19202", "/Gather_739_output_0", "/Unsqueeze_1569_output_0", "/Concat_637_output_0", "/Reshape_442_output_0", "/q_norm_49/Cast_output_0", "/q_norm_49/Mul_1_output_0", "/Mul_392_output_0", "/Shape_688_output_0", "Shape19321", "BinaryOp19324", "BinaryOp19325", "Unsqueeze19326", "BinaryOp19328", "Unsqueeze19329", "StridedSlice19332", "Squeeze19333", "BinaryOp19334", "BinaryOp19335", "/Gather_740_output_0", "/Div_147_output_0", "/Unsqueeze_1575_output_0", "/Slice_197_output_0", "/Neg_98_output_0", "/Unsqueeze_1574_output_0", "/Slice_196_output_0", "/Concat_640_output_0", "/Mul_393_output_0", "/Add_245_output_0", "/FakeLinear_197_output_0", "/Unsqueeze_1570_output_0", "/Unsqueeze_1571_output_0", "/Concat_638_output_0", "/Reshape_443_output_0", "/k_norm_49/Cast_output_0", "/k_norm_49/Mul_1_output_0", "/Mul_394_output_0", "/Shape_689_output_0", "Shape19217", "BinaryOp19220", "BinaryOp19221", "Unsqueeze19222", "BinaryOp19224", "Unsqueeze19225", "StridedSlice19228", "Squeeze19229", "BinaryOp19230", "BinaryOp19231", "/Gather_741_output_0", "/Div_148_output_0", "/Unsqueeze_1577_output_0", "/Slice_199_output_0", "/Neg_99_output_0", "/Unsqueeze_1576_output_0", "/Slice_198_output_0", "/Concat_641_output_0", "/Mul_395_output_0", "/Add_246_output_0", "/FakeLinear_198_output_0", "/Unsqueeze_1572_output_0", "/Unsqueeze_1573_output_0", "/Concat_639_output_0", "/Reshape_444_output_0", "/Reshape_449_output_0", "/FakeLinear_199_output_0", "/Add_248_output_0", "/post_attention_layernorm_49/Mul_1_output_0", "/mlp/gate_proj_49/FakeLinear_output_0", "/mlp/act_fn_49/Mul_output_0", "/mlp/up_proj_49/FakeLinear_output_0", "/mlp_49/Mul_output_0", "/mlp/down_proj_49/FakeLinear_output_0", "/Add_249_output_0", "/Reshape_450_output_0", "/input_layernorm_50/Mul_1_output_0", "/FakeLinear_200_output_0", "/Shape_700_output_0", "Shape19560", "BinaryOp19563", "BinaryOp19564", "Unsqueeze19565", "BinaryOp19567", "Unsqueeze19568", "StridedSlice19571", "BinaryOp19573", "BinaryOp19574", "/Gather_753_output_0", "/Unsqueeze_1600_output_0", "Unsqueeze19582", "Unsqueeze19585", "StridedSlice19588", "Squeeze19589", "BinaryOp19590", "BinaryOp19591", "/Gather_754_output_0", "/Unsqueeze_1601_output_0", "/Concat_650_output_0", "/Reshape_451_output_0", "/q_norm_50/Cast_output_0", "/q_norm_50/Mul_1_output_0", "/Mul_400_output_0", "/Shape_702_output_0", "Shape19710", "BinaryOp19713", "BinaryOp19714", "Unsqueeze19715", "BinaryOp19717", "Unsqueeze19718", "StridedSlice19721", "Squeeze19722", "BinaryOp19723", "BinaryOp19724", "/Gather_755_output_0", "/Div_150_output_0", "/Unsqueeze_1607_output_0", "/Slice_201_output_0", "/Neg_100_output_0", "/Unsqueeze_1606_output_0", "/Slice_200_output_0", "/Concat_653_output_0", "/Mul_401_output_0", "/Add_250_output_0", "/FakeLinear_201_output_0", "/Unsqueeze_1602_output_0", "/Unsqueeze_1603_output_0", "/Concat_651_output_0", "/Reshape_452_output_0", "/k_norm_50/Cast_output_0", "/k_norm_50/Mul_1_output_0", "/Mul_402_output_0", "/Shape_703_output_0", "Shape19606", "BinaryOp19609", "BinaryOp19610", "Unsqueeze19611", "BinaryOp19613", "Unsqueeze19614", "StridedSlice19617", "Squeeze19618", "BinaryOp19619", "BinaryOp19620", "/Gather_756_output_0", "/Div_151_output_0", "/Unsqueeze_1609_output_0", "/Slice_203_output_0", "/Neg_101_output_0", "/Unsqueeze_1608_output_0", "/Slice_202_output_0", "/Concat_654_output_0", "/Mul_403_output_0", "/Add_251_output_0", "/FakeLinear_202_output_0", "/Unsqueeze_1604_output_0", "/Unsqueeze_1605_output_0", "/Concat_652_output_0", "/Reshape_453_output_0", "/Reshape_458_output_0", "/FakeLinear_203_output_0", "/Add_253_output_0", "/post_attention_layernorm_50/Mul_1_output_0", "/mlp/gate_proj_50/FakeLinear_output_0", "/mlp/act_fn_50/Mul_output_0", "/mlp/up_proj_50/FakeLinear_output_0", "/mlp_50/Mul_output_0", "/mlp/down_proj_50/FakeLinear_output_0", "/Add_254_output_0", "/Reshape_459_output_0", "/input_layernorm_51/Mul_1_output_0", "/FakeLinear_204_output_0", "/Shape_714_output_0", "Shape19949", "BinaryOp19952", "BinaryOp19953", "Unsqueeze19954", "BinaryOp19956", "Unsqueeze19957", "StridedSlice19960", "BinaryOp19962", "BinaryOp19963", "/Gather_768_output_0", "/Unsqueeze_1632_output_0", "Unsqueeze19971", "Unsqueeze19974", "StridedSlice19977", "Squeeze19978", "BinaryOp19979", "BinaryOp19980", "/Gather_769_output_0", "/Unsqueeze_1633_output_0", "/Concat_663_output_0", "/Reshape_460_output_0", "/q_norm_51/Cast_output_0", "/q_norm_51/Mul_1_output_0", "/Mul_408_output_0", "/Shape_716_output_0", "Shape20099", "BinaryOp20102", "BinaryOp20103", "Unsqueeze20104", "BinaryOp20106", "Unsqueeze20107", "StridedSlice20110", "Squeeze20111", "BinaryOp20112", "BinaryOp20113", "/Gather_770_output_0", "/Div_153_output_0", "/Unsqueeze_1639_output_0", "/Slice_205_output_0", "/Neg_102_output_0", "/Unsqueeze_1638_output_0", "/Slice_204_output_0", "/Concat_666_output_0", "/Mul_409_output_0", "/Add_255_output_0", "/FakeLinear_205_output_0", "/Unsqueeze_1634_output_0", "/Unsqueeze_1635_output_0", "/Concat_664_output_0", "/Reshape_461_output_0", "/k_norm_51/Cast_output_0", "/k_norm_51/Mul_1_output_0", "/Mul_410_output_0", "/Shape_717_output_0", "Shape19995", "BinaryOp19998", "BinaryOp19999", "Unsqueeze20000", "BinaryOp20002", "Unsqueeze20003", "StridedSlice20006", "Squeeze20007", "BinaryOp20008", "BinaryOp20009", "/Gather_771_output_0", "/Div_154_output_0", "/Unsqueeze_1641_output_0", "/Slice_207_output_0", "/Neg_103_output_0", "/Unsqueeze_1640_output_0", "/Slice_206_output_0", "/Concat_667_output_0", "/Mul_411_output_0", "/Add_256_output_0", "/FakeLinear_206_output_0", "/Unsqueeze_1636_output_0", "/Unsqueeze_1637_output_0", "/Concat_665_output_0", "/Reshape_462_output_0", "/Reshape_467_output_0", "/FakeLinear_207_output_0", "/Add_258_output_0", "/post_attention_layernorm_51/Mul_1_output_0", "/mlp/gate_proj_51/FakeLinear_output_0", "/mlp/act_fn_51/Mul_output_0", "/mlp/up_proj_51/FakeLinear_output_0", "/mlp_51/Mul_output_0", "/mlp/down_proj_51/FakeLinear_output_0", "/Add_259_output_0", "/Reshape_468_output_0", "/input_layernorm_52/Mul_1_output_0", "/FakeLinear_208_output_0", "/Shape_728_output_0", "Shape20338", "BinaryOp20341", "BinaryOp20342", "Unsqueeze20343", "BinaryOp20345", "Unsqueeze20346", "StridedSlice20349", "BinaryOp20351", "BinaryOp20352", "/Gather_783_output_0", "/Unsqueeze_1664_output_0", "Unsqueeze20360", "Unsqueeze20363", "StridedSlice20366", "Squeeze20367", "BinaryOp20368", "BinaryOp20369", "/Gather_784_output_0", "/Unsqueeze_1665_output_0", "/Concat_676_output_0", "/Reshape_469_output_0", "/q_norm_52/Cast_output_0", "/q_norm_52/Mul_1_output_0", "/Mul_416_output_0", "/Shape_730_output_0", "Shape20488", "BinaryOp20491", "BinaryOp20492", "Unsqueeze20493", "BinaryOp20495", "Unsqueeze20496", "StridedSlice20499", "Squeeze20500", "BinaryOp20501", "BinaryOp20502", "/Gather_785_output_0", "/Div_156_output_0", "/Unsqueeze_1671_output_0", "/Slice_209_output_0", "/Neg_104_output_0", "/Unsqueeze_1670_output_0", "/Slice_208_output_0", "/Concat_679_output_0", "/Mul_417_output_0", "/Add_260_output_0", "/FakeLinear_209_output_0", "/Unsqueeze_1666_output_0", "/Unsqueeze_1667_output_0", "/Concat_677_output_0", "/Reshape_470_output_0", "/k_norm_52/Cast_output_0", "/k_norm_52/Mul_1_output_0", "/Mul_418_output_0", "/Shape_731_output_0", "Shape20384", "BinaryOp20387", "BinaryOp20388", "Unsqueeze20389", "BinaryOp20391", "Unsqueeze20392", "StridedSlice20395", "Squeeze20396", "BinaryOp20397", "BinaryOp20398", "/Gather_786_output_0", "/Div_157_output_0", "/Unsqueeze_1673_output_0", "/Slice_211_output_0", "/Neg_105_output_0", "/Unsqueeze_1672_output_0", "/Slice_210_output_0", "/Concat_680_output_0", "/Mul_419_output_0", "/Add_261_output_0", "/FakeLinear_210_output_0", "/Unsqueeze_1668_output_0", "/Unsqueeze_1669_output_0", "/Concat_678_output_0", "/Reshape_471_output_0", "/Reshape_476_output_0", "/FakeLinear_211_output_0", "/Add_263_output_0", "/post_attention_layernorm_52/Mul_1_output_0", "/mlp/gate_proj_52/FakeLinear_output_0", "/mlp/act_fn_52/Mul_output_0", "/mlp/up_proj_52/FakeLinear_output_0", "/mlp_52/Mul_output_0", "/mlp/down_proj_52/FakeLinear_output_0", "/Add_264_output_0", "/Reshape_477_output_0", "/input_layernorm_53/Mul_1_output_0", "/FakeLinear_212_output_0", "/Shape_742_output_0", "Shape20727", "BinaryOp20730", "BinaryOp20731", "Unsqueeze20732", "BinaryOp20734", "Unsqueeze20735", "StridedSlice20738", "BinaryOp20740", "BinaryOp20741", "/Gather_798_output_0", "/Unsqueeze_1696_output_0", "Unsqueeze20749", "Unsqueeze20752", "StridedSlice20755", "Squeeze20756", "BinaryOp20757", "BinaryOp20758", "/Gather_799_output_0", "/Unsqueeze_1697_output_0", "/Concat_689_output_0", "/Reshape_478_output_0", "/q_norm_53/Cast_output_0", "/q_norm_53/Mul_1_output_0", "/Mul_424_output_0", "/Shape_744_output_0", "Shape20877", "BinaryOp20880", "BinaryOp20881", "Unsqueeze20882", "BinaryOp20884", "Unsqueeze20885", "StridedSlice20888", "Squeeze20889", "BinaryOp20890", "BinaryOp20891", "/Gather_800_output_0", "/Div_159_output_0", "/Unsqueeze_1703_output_0", "/Slice_213_output_0", "/Neg_106_output_0", "/Unsqueeze_1702_output_0", "/Slice_212_output_0", "/Concat_692_output_0", "/Mul_425_output_0", "/Add_265_output_0", "/FakeLinear_213_output_0", "/Unsqueeze_1698_output_0", "/Unsqueeze_1699_output_0", "/Concat_690_output_0", "/Reshape_479_output_0", "/k_norm_53/Cast_output_0", "/k_norm_53/Mul_1_output_0", "/Mul_426_output_0", "/Shape_745_output_0", "Shape20773", "BinaryOp20776", "BinaryOp20777", "Unsqueeze20778", "BinaryOp20780", "Unsqueeze20781", "StridedSlice20784", "Squeeze20785", "BinaryOp20786", "BinaryOp20787", "/Gather_801_output_0", "/Div_160_output_0", "/Unsqueeze_1705_output_0", "/Slice_215_output_0", "/Neg_107_output_0", "/Unsqueeze_1704_output_0", "/Slice_214_output_0", "/Concat_693_output_0", "/Mul_427_output_0", "/Add_266_output_0", "/FakeLinear_214_output_0", "/Unsqueeze_1700_output_0", "/Unsqueeze_1701_output_0", "/Concat_691_output_0", "/Reshape_480_output_0", "/Reshape_485_output_0", "/FakeLinear_215_output_0", "/Add_268_output_0", "/post_attention_layernorm_53/Mul_1_output_0", "/mlp/gate_proj_53/FakeLinear_output_0", "/mlp/act_fn_53/Mul_output_0", "/mlp/up_proj_53/FakeLinear_output_0", "/mlp_53/Mul_output_0", "/mlp/down_proj_53/FakeLinear_output_0", "/Add_269_output_0", "/Reshape_486_output_0", "/input_layernorm_54/Mul_1_output_0", "/FakeLinear_216_output_0", "/Shape_756_output_0", "Shape21116", "BinaryOp21119", "BinaryOp21120", "Unsqueeze21121", "BinaryOp21123", "Unsqueeze21124", "StridedSlice21127", "BinaryOp21129", "BinaryOp21130", "/Gather_813_output_0", "/Unsqueeze_1728_output_0", "Unsqueeze21138", "Unsqueeze21141", "StridedSlice21144", "Squeeze21145", "BinaryOp21146", "BinaryOp21147", "/Gather_814_output_0", "/Unsqueeze_1729_output_0", "/Concat_702_output_0", "/Reshape_487_output_0", "/q_norm_54/Cast_output_0", "/q_norm_54/Mul_1_output_0", "/Mul_432_output_0", "/Shape_758_output_0", "Shape21266", "BinaryOp21269", "BinaryOp21270", "Unsqueeze21271", "BinaryOp21273", "Unsqueeze21274", "StridedSlice21277", "Squeeze21278", "BinaryOp21279", "BinaryOp21280", "/Gather_815_output_0", "/Div_162_output_0", "/Unsqueeze_1735_output_0", "/Slice_217_output_0", "/Neg_108_output_0", "/Unsqueeze_1734_output_0", "/Slice_216_output_0", "/Concat_705_output_0", "/Mul_433_output_0", "/Add_270_output_0", "/FakeLinear_217_output_0", "/Unsqueeze_1730_output_0", "/Unsqueeze_1731_output_0", "/Concat_703_output_0", "/Reshape_488_output_0", "/k_norm_54/Cast_output_0", "/k_norm_54/Mul_1_output_0", "/Mul_434_output_0", "/Shape_759_output_0", "Shape21162", "BinaryOp21165", "BinaryOp21166", "Unsqueeze21167", "BinaryOp21169", "Unsqueeze21170", "StridedSlice21173", "Squeeze21174", "BinaryOp21175", "BinaryOp21176", "/Gather_816_output_0", "/Div_163_output_0", "/Unsqueeze_1737_output_0", "/Slice_219_output_0", "/Neg_109_output_0", "/Unsqueeze_1736_output_0", "/Slice_218_output_0", "/Concat_706_output_0", "/Mul_435_output_0", "/Add_271_output_0", "/FakeLinear_218_output_0", "/Unsqueeze_1732_output_0", "/Unsqueeze_1733_output_0", "/Concat_704_output_0", "/Reshape_489_output_0", "/Reshape_494_output_0", "/FakeLinear_219_output_0", "/Add_273_output_0", "/post_attention_layernorm_54/Mul_1_output_0", "/mlp/gate_proj_54/FakeLinear_output_0", "/mlp/act_fn_54/Mul_output_0", "/mlp/up_proj_54/FakeLinear_output_0", "/mlp_54/Mul_output_0", "/mlp/down_proj_54/FakeLinear_output_0", "/Add_274_output_0", "/Reshape_495_output_0", "/input_layernorm_55/Mul_1_output_0", "/FakeLinear_220_output_0", "/Shape_770_output_0", "Shape21505", "BinaryOp21508", "BinaryOp21509", "Unsqueeze21510", "BinaryOp21512", "Unsqueeze21513", "StridedSlice21516", "BinaryOp21518", "BinaryOp21519", "/Gather_828_output_0", "/Unsqueeze_1760_output_0", "Unsqueeze21527", "Unsqueeze21530", "StridedSlice21533", "Squeeze21534", "BinaryOp21535", "BinaryOp21536", "/Gather_829_output_0", "/Unsqueeze_1761_output_0", "/Concat_715_output_0", "/Reshape_496_output_0", "/q_norm_55/Cast_output_0", "/q_norm_55/Mul_1_output_0", "/Mul_440_output_0", "/Shape_772_output_0", "Shape21655", "BinaryOp21658", "BinaryOp21659", "Unsqueeze21660", "BinaryOp21662", "Unsqueeze21663", "StridedSlice21666", "Squeeze21667", "BinaryOp21668", "BinaryOp21669", "/Gather_830_output_0", "/Div_165_output_0", "/Unsqueeze_1767_output_0", "/Slice_221_output_0", "/Neg_110_output_0", "/Unsqueeze_1766_output_0", "/Slice_220_output_0", "/Concat_718_output_0", "/Mul_441_output_0", "/Add_275_output_0", "/FakeLinear_221_output_0", "/Unsqueeze_1762_output_0", "/Unsqueeze_1763_output_0", "/Concat_716_output_0", "/Reshape_497_output_0", "/k_norm_55/Cast_output_0", "/k_norm_55/Mul_1_output_0", "/Mul_442_output_0", "/Shape_773_output_0", "Shape21551", "BinaryOp21554", "BinaryOp21555", "Unsqueeze21556", "BinaryOp21558", "Unsqueeze21559", "StridedSlice21562", "Squeeze21563", "BinaryOp21564", "BinaryOp21565", "/Gather_831_output_0", "/Div_166_output_0", "/Unsqueeze_1769_output_0", "/Slice_223_output_0", "/Neg_111_output_0", "/Unsqueeze_1768_output_0", "/Slice_222_output_0", "/Concat_719_output_0", "/Mul_443_output_0", "/Add_276_output_0", "/FakeLinear_222_output_0", "/Unsqueeze_1764_output_0", "/Unsqueeze_1765_output_0", "/Concat_717_output_0", "/Reshape_498_output_0", "/Reshape_503_output_0", "/FakeLinear_223_output_0", "/Add_278_output_0", "/post_attention_layernorm_55/Mul_1_output_0", "/mlp/gate_proj_55/FakeLinear_output_0", "/mlp/act_fn_55/Mul_output_0", "/mlp/up_proj_55/FakeLinear_output_0", "/mlp_55/Mul_output_0", "/mlp/down_proj_55/FakeLinear_output_0", "/Add_279_output_0", "/Reshape_504_output_0", "/input_layernorm_56/Mul_1_output_0", "/FakeLinear_224_output_0", "/Shape_784_output_0", "Shape21894", "BinaryOp21897", "BinaryOp21898", "Unsqueeze21899", "BinaryOp21901", "Unsqueeze21902", "StridedSlice21905", "BinaryOp21907", "BinaryOp21908", "/Gather_843_output_0", "/Unsqueeze_1792_output_0", "Unsqueeze21916", "Unsqueeze21919", "StridedSlice21922", "Squeeze21923", "BinaryOp21924", "BinaryOp21925", "/Gather_844_output_0", "/Unsqueeze_1793_output_0", "/Concat_728_output_0", "/Reshape_505_output_0", "/q_norm_56/Cast_output_0", "/q_norm_56/Mul_1_output_0", "/Mul_448_output_0", "/Shape_786_output_0", "Shape22044", "BinaryOp22047", "BinaryOp22048", "Unsqueeze22049", "BinaryOp22051", "Unsqueeze22052", "StridedSlice22055", "Squeeze22056", "BinaryOp22057", "BinaryOp22058", "/Gather_845_output_0", "/Div_168_output_0", "/Unsqueeze_1799_output_0", "/Slice_225_output_0", "/Neg_112_output_0", "/Unsqueeze_1798_output_0", "/Slice_224_output_0", "/Concat_731_output_0", "/Mul_449_output_0", "/Add_280_output_0", "/FakeLinear_225_output_0", "/Unsqueeze_1794_output_0", "/Unsqueeze_1795_output_0", "/Concat_729_output_0", "/Reshape_506_output_0", "/k_norm_56/Cast_output_0", "/k_norm_56/Mul_1_output_0", "/Mul_450_output_0", "/Shape_787_output_0", "Shape21940", "BinaryOp21943", "BinaryOp21944", "Unsqueeze21945", "BinaryOp21947", "Unsqueeze21948", "StridedSlice21951", "Squeeze21952", "BinaryOp21953", "BinaryOp21954", "/Gather_846_output_0", "/Div_169_output_0", "/Unsqueeze_1801_output_0", "/Slice_227_output_0", "/Neg_113_output_0", "/Unsqueeze_1800_output_0", "/Slice_226_output_0", "/Concat_732_output_0", "/Mul_451_output_0", "/Add_281_output_0", "/FakeLinear_226_output_0", "/Unsqueeze_1796_output_0", "/Unsqueeze_1797_output_0", "/Concat_730_output_0", "/Reshape_507_output_0", "/Reshape_512_output_0", "/FakeLinear_227_output_0", "/Add_283_output_0", "/post_attention_layernorm_56/Mul_1_output_0", "/mlp/gate_proj_56/FakeLinear_output_0", "/mlp/act_fn_56/Mul_output_0", "/mlp/up_proj_56/FakeLinear_output_0", "/mlp_56/Mul_output_0", "/mlp/down_proj_56/FakeLinear_output_0", "/Add_284_output_0", "/Reshape_513_output_0", "/input_layernorm_57/Mul_1_output_0", "/FakeLinear_228_output_0", "/Shape_798_output_0", "Shape22283", "BinaryOp22286", "BinaryOp22287", "Unsqueeze22288", "BinaryOp22290", "Unsqueeze22291", "StridedSlice22294", "BinaryOp22296", "BinaryOp22297", "/Gather_858_output_0", "/Unsqueeze_1824_output_0", "Unsqueeze22305", "Unsqueeze22308", "StridedSlice22311", "Squeeze22312", "BinaryOp22313", "BinaryOp22314", "/Gather_859_output_0", "/Unsqueeze_1825_output_0", "/Concat_741_output_0", "/Reshape_514_output_0", "/q_norm_57/Cast_output_0", "/q_norm_57/Mul_1_output_0", "/Mul_456_output_0", "/Shape_800_output_0", "Shape22433", "BinaryOp22436", "BinaryOp22437", "Unsqueeze22438", "BinaryOp22440", "Unsqueeze22441", "StridedSlice22444", "Squeeze22445", "BinaryOp22446", "BinaryOp22447", "/Gather_860_output_0", "/Div_171_output_0", "/Unsqueeze_1831_output_0", "/Slice_229_output_0", "/Neg_114_output_0", "/Unsqueeze_1830_output_0", "/Slice_228_output_0", "/Concat_744_output_0", "/Mul_457_output_0", "/Add_285_output_0", "/FakeLinear_229_output_0", "/Unsqueeze_1826_output_0", "/Unsqueeze_1827_output_0", "/Concat_742_output_0", "/Reshape_515_output_0", "/k_norm_57/Cast_output_0", "/k_norm_57/Mul_1_output_0", "/Mul_458_output_0", "/Shape_801_output_0", "Shape22329", "BinaryOp22332", "BinaryOp22333", "Unsqueeze22334", "BinaryOp22336", "Unsqueeze22337", "StridedSlice22340", "Squeeze22341", "BinaryOp22342", "BinaryOp22343", "/Gather_861_output_0", "/Div_172_output_0", "/Unsqueeze_1833_output_0", "/Slice_231_output_0", "/Neg_115_output_0", "/Unsqueeze_1832_output_0", "/Slice_230_output_0", "/Concat_745_output_0", "/Mul_459_output_0", "/Add_286_output_0", "/FakeLinear_230_output_0", "/Unsqueeze_1828_output_0", "/Unsqueeze_1829_output_0", "/Concat_743_output_0", "/Reshape_516_output_0", "/Reshape_521_output_0", "/FakeLinear_231_output_0", "/Add_288_output_0", "/post_attention_layernorm_57/Mul_1_output_0", "/mlp/gate_proj_57/FakeLinear_output_0", "/mlp/act_fn_57/Mul_output_0", "/mlp/up_proj_57/FakeLinear_output_0", "/mlp_57/Mul_output_0", "/mlp/down_proj_57/FakeLinear_output_0", "/Add_289_output_0", "/Reshape_522_output_0", "/input_layernorm_58/Mul_1_output_0", "/FakeLinear_232_output_0", "/Shape_812_output_0", "Shape22672", "BinaryOp22675", "BinaryOp22676", "Unsqueeze22677", "BinaryOp22679", "Unsqueeze22680", "StridedSlice22683", "BinaryOp22685", "BinaryOp22686", "/Gather_873_output_0", "/Unsqueeze_1856_output_0", "Unsqueeze22694", "Unsqueeze22697", "StridedSlice22700", "Squeeze22701", "BinaryOp22702", "BinaryOp22703", "/Gather_874_output_0", "/Unsqueeze_1857_output_0", "/Concat_754_output_0", "/Reshape_523_output_0", "/q_norm_58/Cast_output_0", "/q_norm_58/Mul_1_output_0", "/Mul_464_output_0", "/Shape_814_output_0", "Shape22822", "BinaryOp22825", "BinaryOp22826", "Unsqueeze22827", "BinaryOp22829", "Unsqueeze22830", "StridedSlice22833", "Squeeze22834", "BinaryOp22835", "BinaryOp22836", "/Gather_875_output_0", "/Div_174_output_0", "/Unsqueeze_1863_output_0", "/Slice_233_output_0", "/Neg_116_output_0", "/Unsqueeze_1862_output_0", "/Slice_232_output_0", "/Concat_757_output_0", "/Mul_465_output_0", "/Add_290_output_0", "/FakeLinear_233_output_0", "/Unsqueeze_1858_output_0", "/Unsqueeze_1859_output_0", "/Concat_755_output_0", "/Reshape_524_output_0", "/k_norm_58/Cast_output_0", "/k_norm_58/Mul_1_output_0", "/Mul_466_output_0", "/Shape_815_output_0", "Shape22718", "BinaryOp22721", "BinaryOp22722", "Unsqueeze22723", "BinaryOp22725", "Unsqueeze22726", "StridedSlice22729", "Squeeze22730", "BinaryOp22731", "BinaryOp22732", "/Gather_876_output_0", "/Div_175_output_0", "/Unsqueeze_1865_output_0", "/Slice_235_output_0", "/Neg_117_output_0", "/Unsqueeze_1864_output_0", "/Slice_234_output_0", "/Concat_758_output_0", "/Mul_467_output_0", "/Add_291_output_0", "/FakeLinear_234_output_0", "/Unsqueeze_1860_output_0", "/Unsqueeze_1861_output_0", "/Concat_756_output_0", "/Reshape_525_output_0", "/Reshape_530_output_0", "/FakeLinear_235_output_0", "/Add_293_output_0", "/post_attention_layernorm_58/Mul_1_output_0", "/mlp/gate_proj_58/FakeLinear_output_0", "/mlp/act_fn_58/Mul_output_0", "/mlp/up_proj_58/FakeLinear_output_0", "/mlp_58/Mul_output_0", "/mlp/down_proj_58/FakeLinear_output_0", "/Add_294_output_0", "/Reshape_531_output_0", "/input_layernorm_59/Mul_1_output_0", "/FakeLinear_236_output_0", "/Shape_826_output_0", "Shape23061", "BinaryOp23064", "BinaryOp23065", "Unsqueeze23066", "BinaryOp23068", "Unsqueeze23069", "StridedSlice23072", "BinaryOp23074", "BinaryOp23075", "/Gather_888_output_0", "/Unsqueeze_1888_output_0", "Unsqueeze23083", "Unsqueeze23086", "StridedSlice23089", "Squeeze23090", "BinaryOp23091", "BinaryOp23092", "/Gather_889_output_0", "/Unsqueeze_1889_output_0", "/Concat_767_output_0", "/Reshape_532_output_0", "/q_norm_59/Cast_output_0", "/q_norm_59/Mul_1_output_0", "/Mul_472_output_0", "/Shape_828_output_0", "Shape23211", "BinaryOp23214", "BinaryOp23215", "Unsqueeze23216", "BinaryOp23218", "Unsqueeze23219", "StridedSlice23222", "Squeeze23223", "BinaryOp23224", "BinaryOp23225", "/Gather_890_output_0", "/Div_177_output_0", "/Unsqueeze_1895_output_0", "/Slice_237_output_0", "/Neg_118_output_0", "/Unsqueeze_1894_output_0", "/Slice_236_output_0", "/Concat_770_output_0", "/Mul_473_output_0", "/Add_295_output_0", "/FakeLinear_237_output_0", "/Unsqueeze_1890_output_0", "/Unsqueeze_1891_output_0", "/Concat_768_output_0", "/Reshape_533_output_0", "/k_norm_59/Cast_output_0", "/k_norm_59/Mul_1_output_0", "/Mul_474_output_0", "/Shape_829_output_0", "Shape23107", "BinaryOp23110", "BinaryOp23111", "Unsqueeze23112", "BinaryOp23114", "Unsqueeze23115", "StridedSlice23118", "Squeeze23119", "BinaryOp23120", "BinaryOp23121", "/Gather_891_output_0", "/Div_178_output_0", "/Unsqueeze_1897_output_0", "/Slice_239_output_0", "/Neg_119_output_0", "/Unsqueeze_1896_output_0", "/Slice_238_output_0", "/Concat_771_output_0", "/Mul_475_output_0", "/Add_296_output_0", "/FakeLinear_238_output_0", "/Unsqueeze_1892_output_0", "/Unsqueeze_1893_output_0", "/Concat_769_output_0", "/Reshape_534_output_0", "/Reshape_539_output_0", "/FakeLinear_239_output_0", "/Add_298_output_0", "/post_attention_layernorm_59/Mul_1_output_0", "/mlp/gate_proj_59/FakeLinear_output_0", "/mlp/act_fn_59/Mul_output_0", "/mlp/up_proj_59/FakeLinear_output_0", "/mlp_59/Mul_output_0", "/mlp/down_proj_59/FakeLinear_output_0", "/Add_299_output_0", "/Reshape_540_output_0", "/input_layernorm_60/Mul_1_output_0", "/FakeLinear_240_output_0", "/Shape_840_output_0", "Shape23450", "BinaryOp23453", "BinaryOp23454", "Unsqueeze23455", "BinaryOp23457", "Unsqueeze23458", "StridedSlice23461", "BinaryOp23463", "BinaryOp23464", "/Gather_903_output_0", "/Unsqueeze_1920_output_0", "Unsqueeze23472", "Unsqueeze23475", "StridedSlice23478", "Squeeze23479", "BinaryOp23480", "BinaryOp23481", "/Gather_904_output_0", "/Unsqueeze_1921_output_0", "/Concat_780_output_0", "/Reshape_541_output_0", "/q_norm_60/Cast_output_0", "/q_norm_60/Mul_1_output_0", "/Mul_480_output_0", "/Shape_842_output_0", "Shape23600", "BinaryOp23603", "BinaryOp23604", "Unsqueeze23605", "BinaryOp23607", "Unsqueeze23608", "StridedSlice23611", "Squeeze23612", "BinaryOp23613", "BinaryOp23614", "/Gather_905_output_0", "/Div_180_output_0", "/Unsqueeze_1927_output_0", "/Slice_241_output_0", "/Neg_120_output_0", "/Unsqueeze_1926_output_0", "/Slice_240_output_0", "/Concat_783_output_0", "/Mul_481_output_0", "/Add_300_output_0", "/FakeLinear_241_output_0", "/Unsqueeze_1922_output_0", "/Unsqueeze_1923_output_0", "/Concat_781_output_0", "/Reshape_542_output_0", "/k_norm_60/Cast_output_0", "/k_norm_60/Mul_1_output_0", "/Mul_482_output_0", "/Shape_843_output_0", "Shape23496", "BinaryOp23499", "BinaryOp23500", "Unsqueeze23501", "BinaryOp23503", "Unsqueeze23504", "StridedSlice23507", "Squeeze23508", "BinaryOp23509", "BinaryOp23510", "/Gather_906_output_0", "/Div_181_output_0", "/Unsqueeze_1929_output_0", "/Slice_243_output_0", "/Neg_121_output_0", "/Unsqueeze_1928_output_0", "/Slice_242_output_0", "/Concat_784_output_0", "/Mul_483_output_0", "/Add_301_output_0", "/FakeLinear_242_output_0", "/Unsqueeze_1924_output_0", "/Unsqueeze_1925_output_0", "/Concat_782_output_0", "/Reshape_543_output_0", "/Reshape_548_output_0", "/FakeLinear_243_output_0", "/Add_303_output_0", "/post_attention_layernorm_60/Mul_1_output_0", "/mlp/gate_proj_60/FakeLinear_output_0", "/mlp/act_fn_60/Mul_output_0", "/mlp/up_proj_60/FakeLinear_output_0", "/mlp_60/Mul_output_0", "/mlp/down_proj_60/FakeLinear_output_0", "/Add_304_output_0", "/Reshape_549_output_0", "/input_layernorm_61/Mul_1_output_0", "/FakeLinear_244_output_0", "/Shape_854_output_0", "Shape23839", "BinaryOp23842", "BinaryOp23843", "Unsqueeze23844", "BinaryOp23846", "Unsqueeze23847", "StridedSlice23850", "BinaryOp23852", "BinaryOp23853", "/Gather_918_output_0", "/Unsqueeze_1952_output_0", "Unsqueeze23861", "Unsqueeze23864", "StridedSlice23867", "Squeeze23868", "BinaryOp23869", "BinaryOp23870", "/Gather_919_output_0", "/Unsqueeze_1953_output_0", "/Concat_793_output_0", "/Reshape_550_output_0", "/q_norm_61/Cast_output_0", "/q_norm_61/Mul_1_output_0", "/Mul_488_output_0", "/Shape_856_output_0", "Shape23989", "BinaryOp23992", "BinaryOp23993", "Unsqueeze23994", "BinaryOp23996", "Unsqueeze23997", "StridedSlice24000", "Squeeze24001", "BinaryOp24002", "BinaryOp24003", "/Gather_920_output_0", "/Div_183_output_0", "/Unsqueeze_1959_output_0", "/Slice_245_output_0", "/Neg_122_output_0", "/Unsqueeze_1958_output_0", "/Slice_244_output_0", "/Concat_796_output_0", "/Mul_489_output_0", "/Add_305_output_0", "/FakeLinear_245_output_0", "/Unsqueeze_1954_output_0", "/Unsqueeze_1955_output_0", "/Concat_794_output_0", "/Reshape_551_output_0", "/k_norm_61/Cast_output_0", "/k_norm_61/Mul_1_output_0", "/Mul_490_output_0", "/Shape_857_output_0", "Shape23885", "BinaryOp23888", "BinaryOp23889", "Unsqueeze23890", "BinaryOp23892", "Unsqueeze23893", "StridedSlice23896", "Squeeze23897", "BinaryOp23898", "BinaryOp23899", "/Gather_921_output_0", "/Div_184_output_0", "/Unsqueeze_1961_output_0", "/Slice_247_output_0", "/Neg_123_output_0", "/Unsqueeze_1960_output_0", "/Slice_246_output_0", "/Concat_797_output_0", "/Mul_491_output_0", "/Add_306_output_0", "/FakeLinear_246_output_0", "/Unsqueeze_1956_output_0", "/Unsqueeze_1957_output_0", "/Concat_795_output_0", "/Reshape_552_output_0", "/Reshape_557_output_0", "/FakeLinear_247_output_0", "/Add_308_output_0", "/post_attention_layernorm_61/Mul_1_output_0", "/mlp/gate_proj_61/FakeLinear_output_0", "/mlp/act_fn_61/Mul_output_0", "/mlp/up_proj_61/FakeLinear_output_0", "/mlp_61/Mul_output_0", "/mlp/down_proj_61/FakeLinear_output_0", "/Add_309_output_0", "/Reshape_558_output_0", "/input_layernorm_62/Mul_1_output_0", "/FakeLinear_248_output_0", "/Shape_868_output_0", "Shape24228", "BinaryOp24231", "BinaryOp24232", "Unsqueeze24233", "BinaryOp24235", "Unsqueeze24236", "StridedSlice24239", "BinaryOp24241", "BinaryOp24242", "/Gather_933_output_0", "/Unsqueeze_1984_output_0", "Unsqueeze24250", "Unsqueeze24253", "StridedSlice24256", "Squeeze24257", "BinaryOp24258", "BinaryOp24259", "/Gather_934_output_0", "/Unsqueeze_1985_output_0", "/Concat_806_output_0", "/Reshape_559_output_0", "/q_norm_62/Cast_output_0", "/q_norm_62/Mul_1_output_0", "/Mul_496_output_0", "/Shape_870_output_0", "Shape24378", "BinaryOp24381", "BinaryOp24382", "Unsqueeze24383", "BinaryOp24385", "Unsqueeze24386", "StridedSlice24389", "Squeeze24390", "BinaryOp24391", "BinaryOp24392", "/Gather_935_output_0", "/Div_186_output_0", "/Unsqueeze_1991_output_0", "/Slice_249_output_0", "/Neg_124_output_0", "/Unsqueeze_1990_output_0", "/Slice_248_output_0", "/Concat_809_output_0", "/Mul_497_output_0", "/Add_310_output_0", "/FakeLinear_249_output_0", "/Unsqueeze_1986_output_0", "/Unsqueeze_1987_output_0", "/Concat_807_output_0", "/Reshape_560_output_0", "/k_norm_62/Cast_output_0", "/k_norm_62/Mul_1_output_0", "/Mul_498_output_0", "/Shape_871_output_0", "Shape24274", "BinaryOp24277", "BinaryOp24278", "Unsqueeze24279", "BinaryOp24281", "Unsqueeze24282", "StridedSlice24285", "Squeeze24286", "BinaryOp24287", "BinaryOp24288", "/Gather_936_output_0", "/Div_187_output_0", "/Unsqueeze_1993_output_0", "/Slice_251_output_0", "/Neg_125_output_0", "/Unsqueeze_1992_output_0", "/Slice_250_output_0", "/Concat_810_output_0", "/Mul_499_output_0", "/Add_311_output_0", "/FakeLinear_250_output_0", "/Unsqueeze_1988_output_0", "/Unsqueeze_1989_output_0", "/Concat_808_output_0", "/Reshape_561_output_0", "/Reshape_566_output_0", "/FakeLinear_251_output_0", "/Add_313_output_0", "/post_attention_layernorm_62/Mul_1_output_0", "/mlp/gate_proj_62/FakeLinear_output_0", "/mlp/act_fn_62/Mul_output_0", "/mlp/up_proj_62/FakeLinear_output_0", "/mlp_62/Mul_output_0", "/mlp/down_proj_62/FakeLinear_output_0", "/Add_314_output_0", "/Reshape_567_output_0", "/input_layernorm_63/Mul_1_output_0", "/FakeLinear_252_output_0", "/Shape_882_output_0", "Shape24617", "BinaryOp24620", "BinaryOp24621", "Unsqueeze24622", "BinaryOp24624", "Unsqueeze24625", "StridedSlice24628", "BinaryOp24630", "BinaryOp24631", "/Gather_948_output_0", "/Unsqueeze_2016_output_0", "Unsqueeze24639", "Unsqueeze24642", "StridedSlice24645", "Squeeze24646", "BinaryOp24647", "BinaryOp24648", "/Gather_949_output_0", "/Unsqueeze_2017_output_0", "/Concat_819_output_0", "/Reshape_568_output_0", "/q_norm_63/Cast_output_0", "/q_norm_63/Mul_1_output_0", "/Mul_504_output_0", "/Shape_884_output_0", "Shape24734", "BinaryOp24737", "BinaryOp24738", "Unsqueeze24739", "BinaryOp24741", "Unsqueeze24742", "StridedSlice24745", "Squeeze24746", "BinaryOp24747", "BinaryOp24748", "/Gather_950_output_0", "/Div_189_output_0", "/Unsqueeze_2023_output_0", "/Slice_253_output_0", "/Neg_126_output_0", "/Unsqueeze_2022_output_0", "/Slice_252_output_0", "/Concat_822_output_0", "/Mul_505_output_0", "/Add_315_output_0", "/FakeLinear_253_output_0", "/Unsqueeze_2018_output_0", "/Unsqueeze_2019_output_0", "/Concat_820_output_0", "/Reshape_569_output_0", "/k_norm_63/Cast_output_0", "/k_norm_63/Mul_1_output_0", "/Mul_506_output_0", "/Shape_885_output_0", "Shape24663", "BinaryOp24666", "BinaryOp24667", "Unsqueeze24668", "BinaryOp24670", "Unsqueeze24671", "StridedSlice24674", "Squeeze24675", "BinaryOp24676", "BinaryOp24677", "/Gather_951_output_0", "/Div_190_output_0", "/Unsqueeze_2025_output_0", "/Slice_255_output_0", "/Neg_127_output_0", "/Unsqueeze_2024_output_0", "/Slice_254_output_0", "/Concat_823_output_0", "/Mul_507_output_0", "/Add_316_output_0", "/FakeLinear_254_output_0", "/Unsqueeze_2020_output_0", "/Unsqueeze_2021_output_0", "/Concat_821_output_0", "/Reshape_570_output_0", "/Reshape_575_output_0", "/FakeLinear_255_output_0", "/Add_318_output_0", "/post_attention_layernorm_63/Mul_1_output_0", "/mlp/gate_proj_63/FakeLinear_output_0", "/mlp/act_fn_63/Mul_output_0", "/mlp/up_proj_63/FakeLinear_output_0", "/mlp_63/Mul_output_0", "/mlp/down_proj_63/FakeLinear_output_0", "/Add_319_output_0", "logits_index", "/lm/Slice_output_0", "/lm/norm/Mul_1_output_0", "logits", "/layers.0/self_attn/q_proj/Linear/pre_reshape", "/layers.0/self_attn/q_proj/Linear/pre_convert", "/layers.0/self_attn/q_proj/Linear", "/layers.0/self_attn/q_proj/Linear/post_convert", "/layers.0/self_attn/k_proj/Linear/pre_reshape", "/layers.0/self_attn/k_proj/Linear/pre_convert", "/layers.0/self_attn/k_proj/Linear", "/layers.0/self_attn/k_proj/Linear/post_convert", "/layers.0/self_attn/v_proj/Linear/pre_reshape", "/layers.0/self_attn/v_proj/Linear/pre_convert", "/layers.0/self_attn/v_proj/Linear", "/layers.0/self_attn/v_proj/Linear/post_convert", "/layers.0/self_attn/o_proj/Linear/pre_reshape", "/layers.0/self_attn/o_proj/Linear/pre_convert", "/layers.0/self_attn/o_proj/Linear", "/layers.0/self_attn/o_proj/Linear/post_convert", "/layers.0/mlp/gate_proj/Linear/pre_reshape", "/layers.0/mlp/gate_proj/Linear/pre_convert", "/layers.0/mlp/gate_proj/Linear", "/layers.0/mlp/gate_proj/Linear/post_convert", "/layers.0/mlp/up_proj/Linear/pre_reshape", "/layers.0/mlp/up_proj/Linear/pre_convert", "/layers.0/mlp/up_proj/Linear", "/layers.0/mlp/up_proj/Linear/post_convert", "/layers.0/mlp/down_proj/Linear/pre_reshape", "/layers.0/mlp/down_proj/Linear/pre_convert", "/layers.0/mlp/down_proj/Linear", "/layers.0/mlp/down_proj/Linear/post_convert", "/layers.1/self_attn/q_proj/Linear/pre_reshape", "/layers.1/self_attn/q_proj/Linear/pre_convert", "/layers.1/self_attn/q_proj/Linear", "/layers.1/self_attn/q_proj/Linear/post_convert", "/layers.1/self_attn/k_proj/Linear/pre_reshape", "/layers.1/self_attn/k_proj/Linear/pre_convert", "/layers.1/self_attn/k_proj/Linear", "/layers.1/self_attn/k_proj/Linear/post_convert", "/layers.1/self_attn/v_proj/Linear/pre_reshape", "/layers.1/self_attn/v_proj/Linear/pre_convert", "/layers.1/self_attn/v_proj/Linear", "/layers.1/self_attn/v_proj/Linear/post_convert", "/layers.1/self_attn/o_proj/Linear/pre_reshape", "/layers.1/self_attn/o_proj/Linear/pre_convert", "/layers.1/self_attn/o_proj/Linear", "/layers.1/self_attn/o_proj/Linear/post_convert", "/layers.1/mlp/gate_proj/Linear/pre_reshape", "/layers.1/mlp/gate_proj/Linear/pre_convert", "/layers.1/mlp/gate_proj/Linear", "/layers.1/mlp/gate_proj/Linear/post_convert", "/layers.1/mlp/up_proj/Linear/pre_reshape", "/layers.1/mlp/up_proj/Linear/pre_convert", "/layers.1/mlp/up_proj/Linear", "/layers.1/mlp/up_proj/Linear/post_convert", "/layers.1/mlp/down_proj/Linear/pre_reshape", "/layers.1/mlp/down_proj/Linear/pre_convert", "/layers.1/mlp/down_proj/Linear", "/layers.1/mlp/down_proj/Linear/post_convert", "/layers.2/self_attn/q_proj/Linear/pre_reshape", "/layers.2/self_attn/q_proj/Linear/pre_convert", "/layers.2/self_attn/q_proj/Linear", "/layers.2/self_attn/q_proj/Linear/post_convert", "/layers.2/self_attn/k_proj/Linear/pre_reshape", "/layers.2/self_attn/k_proj/Linear/pre_convert", "/layers.2/self_attn/k_proj/Linear", "/layers.2/self_attn/k_proj/Linear/post_convert", "/layers.2/self_attn/v_proj/Linear/pre_reshape", "/layers.2/self_attn/v_proj/Linear/pre_convert", "/layers.2/self_attn/v_proj/Linear", "/layers.2/self_attn/v_proj/Linear/post_convert", "/layers.2/self_attn/o_proj/Linear/pre_reshape", "/layers.2/self_attn/o_proj/Linear/pre_convert", "/layers.2/self_attn/o_proj/Linear", "/layers.2/self_attn/o_proj/Linear/post_convert", "/layers.2/mlp/gate_proj/Linear/pre_reshape", "/layers.2/mlp/gate_proj/Linear/pre_convert", "/layers.2/mlp/gate_proj/Linear", "/layers.2/mlp/gate_proj/Linear/post_convert", "/layers.2/mlp/up_proj/Linear/pre_reshape", "/layers.2/mlp/up_proj/Linear/pre_convert", "/layers.2/mlp/up_proj/Linear", "/layers.2/mlp/up_proj/Linear/post_convert", "/layers.2/mlp/down_proj/Linear/pre_reshape", "/layers.2/mlp/down_proj/Linear/pre_convert", "/layers.2/mlp/down_proj/Linear", "/layers.2/mlp/down_proj/Linear/post_convert", "/layers.3/self_attn/q_proj/Linear/pre_reshape", "/layers.3/self_attn/q_proj/Linear/pre_convert", "/layers.3/self_attn/q_proj/Linear", "/layers.3/self_attn/q_proj/Linear/post_convert", "/layers.3/self_attn/k_proj/Linear/pre_reshape", "/layers.3/self_attn/k_proj/Linear/pre_convert", "/layers.3/self_attn/k_proj/Linear", "/layers.3/self_attn/k_proj/Linear/post_convert", "/layers.3/self_attn/v_proj/Linear/pre_reshape", "/layers.3/self_attn/v_proj/Linear/pre_convert", "/layers.3/self_attn/v_proj/Linear", "/layers.3/self_attn/v_proj/Linear/post_convert", "/layers.3/self_attn/o_proj/Linear/pre_reshape", "/layers.3/self_attn/o_proj/Linear/pre_convert", "/layers.3/self_attn/o_proj/Linear", "/layers.3/self_attn/o_proj/Linear/post_convert", "/layers.3/mlp/gate_proj/Linear/pre_reshape", "/layers.3/mlp/gate_proj/Linear/pre_convert", "/layers.3/mlp/gate_proj/Linear", "/layers.3/mlp/gate_proj/Linear/post_convert", "/layers.3/mlp/up_proj/Linear/pre_reshape", "/layers.3/mlp/up_proj/Linear/pre_convert", "/layers.3/mlp/up_proj/Linear", "/layers.3/mlp/up_proj/Linear/post_convert", "/layers.3/mlp/down_proj/Linear/pre_reshape", "/layers.3/mlp/down_proj/Linear/pre_convert", "/layers.3/mlp/down_proj/Linear", "/layers.3/mlp/down_proj/Linear/post_convert", "/layers.4/self_attn/q_proj/Linear/pre_reshape", "/layers.4/self_attn/q_proj/Linear/pre_convert", "/layers.4/self_attn/q_proj/Linear", "/layers.4/self_attn/q_proj/Linear/post_convert", "/layers.4/self_attn/k_proj/Linear/pre_reshape", "/layers.4/self_attn/k_proj/Linear/pre_convert", "/layers.4/self_attn/k_proj/Linear", "/layers.4/self_attn/k_proj/Linear/post_convert", "/layers.4/self_attn/v_proj/Linear/pre_reshape", "/layers.4/self_attn/v_proj/Linear/pre_convert", "/layers.4/self_attn/v_proj/Linear", "/layers.4/self_attn/v_proj/Linear/post_convert", "/layers.4/self_attn/o_proj/Linear/pre_reshape", "/layers.4/self_attn/o_proj/Linear/pre_convert", "/layers.4/self_attn/o_proj/Linear", "/layers.4/self_attn/o_proj/Linear/post_convert", "/layers.4/mlp/gate_proj/Linear/pre_reshape", "/layers.4/mlp/gate_proj/Linear/pre_convert", "/layers.4/mlp/gate_proj/Linear", "/layers.4/mlp/gate_proj/Linear/post_convert", "/layers.4/mlp/up_proj/Linear/pre_reshape", "/layers.4/mlp/up_proj/Linear/pre_convert", "/layers.4/mlp/up_proj/Linear", "/layers.4/mlp/up_proj/Linear/post_convert", "/layers.4/mlp/down_proj/Linear/pre_reshape", "/layers.4/mlp/down_proj/Linear/pre_convert", "/layers.4/mlp/down_proj/Linear", "/layers.4/mlp/down_proj/Linear/post_convert", "/layers.5/self_attn/q_proj/Linear/pre_reshape", "/layers.5/self_attn/q_proj/Linear/pre_convert", "/layers.5/self_attn/q_proj/Linear", "/layers.5/self_attn/q_proj/Linear/post_convert", "/layers.5/self_attn/k_proj/Linear/pre_reshape", "/layers.5/self_attn/k_proj/Linear/pre_convert", "/layers.5/self_attn/k_proj/Linear", "/layers.5/self_attn/k_proj/Linear/post_convert", "/layers.5/self_attn/v_proj/Linear/pre_reshape", "/layers.5/self_attn/v_proj/Linear/pre_convert", "/layers.5/self_attn/v_proj/Linear", "/layers.5/self_attn/v_proj/Linear/post_convert", "/layers.5/self_attn/o_proj/Linear/pre_reshape", "/layers.5/self_attn/o_proj/Linear/pre_convert", "/layers.5/self_attn/o_proj/Linear", "/layers.5/self_attn/o_proj/Linear/post_convert", "/layers.5/mlp/gate_proj/Linear/pre_reshape", "/layers.5/mlp/gate_proj/Linear/pre_convert", "/layers.5/mlp/gate_proj/Linear", "/layers.5/mlp/gate_proj/Linear/post_convert", "/layers.5/mlp/up_proj/Linear/pre_reshape", "/layers.5/mlp/up_proj/Linear/pre_convert", "/layers.5/mlp/up_proj/Linear", "/layers.5/mlp/up_proj/Linear/post_convert", "/layers.5/mlp/down_proj/Linear/pre_reshape", "/layers.5/mlp/down_proj/Linear/pre_convert", "/layers.5/mlp/down_proj/Linear", "/layers.5/mlp/down_proj/Linear/post_convert", "/layers.6/self_attn/q_proj/Linear/pre_reshape", "/layers.6/self_attn/q_proj/Linear/pre_convert", "/layers.6/self_attn/q_proj/Linear", "/layers.6/self_attn/q_proj/Linear/post_convert", "/layers.6/self_attn/k_proj/Linear/pre_reshape", "/layers.6/self_attn/k_proj/Linear/pre_convert", "/layers.6/self_attn/k_proj/Linear", "/layers.6/self_attn/k_proj/Linear/post_convert", "/layers.6/self_attn/v_proj/Linear/pre_reshape", "/layers.6/self_attn/v_proj/Linear/pre_convert", "/layers.6/self_attn/v_proj/Linear", "/layers.6/self_attn/v_proj/Linear/post_convert", "/layers.6/self_attn/o_proj/Linear/pre_reshape", "/layers.6/self_attn/o_proj/Linear/pre_convert", "/layers.6/self_attn/o_proj/Linear", "/layers.6/self_attn/o_proj/Linear/post_convert", "/layers.6/mlp/gate_proj/Linear/pre_reshape", "/layers.6/mlp/gate_proj/Linear/pre_convert", "/layers.6/mlp/gate_proj/Linear", "/layers.6/mlp/gate_proj/Linear/post_convert", "/layers.6/mlp/up_proj/Linear/pre_reshape", "/layers.6/mlp/up_proj/Linear/pre_convert", "/layers.6/mlp/up_proj/Linear", "/layers.6/mlp/up_proj/Linear/post_convert", "/layers.6/mlp/down_proj/Linear/pre_reshape", "/layers.6/mlp/down_proj/Linear/pre_convert", "/layers.6/mlp/down_proj/Linear", "/layers.6/mlp/down_proj/Linear/post_convert", "/layers.7/self_attn/q_proj/Linear/pre_reshape", "/layers.7/self_attn/q_proj/Linear/pre_convert", "/layers.7/self_attn/q_proj/Linear", "/layers.7/self_attn/q_proj/Linear/post_convert", "/layers.7/self_attn/k_proj/Linear/pre_reshape", "/layers.7/self_attn/k_proj/Linear/pre_convert", "/layers.7/self_attn/k_proj/Linear", "/layers.7/self_attn/k_proj/Linear/post_convert", "/layers.7/self_attn/v_proj/Linear/pre_reshape", "/layers.7/self_attn/v_proj/Linear/pre_convert", "/layers.7/self_attn/v_proj/Linear", "/layers.7/self_attn/v_proj/Linear/post_convert", "/layers.7/self_attn/o_proj/Linear/pre_reshape", "/layers.7/self_attn/o_proj/Linear/pre_convert", "/layers.7/self_attn/o_proj/Linear", "/layers.7/self_attn/o_proj/Linear/post_convert", "/layers.7/mlp/gate_proj/Linear/pre_reshape", "/layers.7/mlp/gate_proj/Linear/pre_convert", "/layers.7/mlp/gate_proj/Linear", "/layers.7/mlp/gate_proj/Linear/post_convert", "/layers.7/mlp/up_proj/Linear/pre_reshape", "/layers.7/mlp/up_proj/Linear/pre_convert", "/layers.7/mlp/up_proj/Linear", "/layers.7/mlp/up_proj/Linear/post_convert", "/layers.7/mlp/down_proj/Linear/pre_reshape", "/layers.7/mlp/down_proj/Linear/pre_convert", "/layers.7/mlp/down_proj/Linear", "/layers.7/mlp/down_proj/Linear/post_convert", "/layers.8/self_attn/q_proj/Linear/pre_reshape", "/layers.8/self_attn/q_proj/Linear/pre_convert", "/layers.8/self_attn/q_proj/Linear", "/layers.8/self_attn/q_proj/Linear/post_convert", "/layers.8/self_attn/k_proj/Linear/pre_reshape", "/layers.8/self_attn/k_proj/Linear/pre_convert", "/layers.8/self_attn/k_proj/Linear", "/layers.8/self_attn/k_proj/Linear/post_convert", "/layers.8/self_attn/v_proj/Linear/pre_reshape", "/layers.8/self_attn/v_proj/Linear/pre_convert", "/layers.8/self_attn/v_proj/Linear", "/layers.8/self_attn/v_proj/Linear/post_convert", "/layers.8/self_attn/o_proj/Linear/pre_reshape", "/layers.8/self_attn/o_proj/Linear/pre_convert", "/layers.8/self_attn/o_proj/Linear", "/layers.8/self_attn/o_proj/Linear/post_convert", "/layers.8/mlp/gate_proj/Linear/pre_reshape", "/layers.8/mlp/gate_proj/Linear/pre_convert", "/layers.8/mlp/gate_proj/Linear", "/layers.8/mlp/gate_proj/Linear/post_convert", "/layers.8/mlp/up_proj/Linear/pre_reshape", "/layers.8/mlp/up_proj/Linear/pre_convert", "/layers.8/mlp/up_proj/Linear", "/layers.8/mlp/up_proj/Linear/post_convert", "/layers.8/mlp/down_proj/Linear/pre_reshape", "/layers.8/mlp/down_proj/Linear/pre_convert", "/layers.8/mlp/down_proj/Linear", "/layers.8/mlp/down_proj/Linear/post_convert", "/layers.9/self_attn/q_proj/Linear/pre_reshape", "/layers.9/self_attn/q_proj/Linear/pre_convert", "/layers.9/self_attn/q_proj/Linear", "/layers.9/self_attn/q_proj/Linear/post_convert", "/layers.9/self_attn/k_proj/Linear/pre_reshape", "/layers.9/self_attn/k_proj/Linear/pre_convert", "/layers.9/self_attn/k_proj/Linear", "/layers.9/self_attn/k_proj/Linear/post_convert", "/layers.9/self_attn/v_proj/Linear/pre_reshape", "/layers.9/self_attn/v_proj/Linear/pre_convert", "/layers.9/self_attn/v_proj/Linear", "/layers.9/self_attn/v_proj/Linear/post_convert", "/layers.9/self_attn/o_proj/Linear/pre_reshape", "/layers.9/self_attn/o_proj/Linear/pre_convert", "/layers.9/self_attn/o_proj/Linear", "/layers.9/self_attn/o_proj/Linear/post_convert", "/layers.9/mlp/gate_proj/Linear/pre_reshape", "/layers.9/mlp/gate_proj/Linear/pre_convert", "/layers.9/mlp/gate_proj/Linear", "/layers.9/mlp/gate_proj/Linear/post_convert", "/layers.9/mlp/up_proj/Linear/pre_reshape", "/layers.9/mlp/up_proj/Linear/pre_convert", "/layers.9/mlp/up_proj/Linear", "/layers.9/mlp/up_proj/Linear/post_convert", "/layers.9/mlp/down_proj/Linear/pre_reshape", "/layers.9/mlp/down_proj/Linear/pre_convert", "/layers.9/mlp/down_proj/Linear", "/layers.9/mlp/down_proj/Linear/post_convert", "/layers.10/self_attn/q_proj/Linear/pre_reshape", "/layers.10/self_attn/q_proj/Linear/pre_convert", "/layers.10/self_attn/q_proj/Linear", "/layers.10/self_attn/q_proj/Linear/post_convert", "/layers.10/self_attn/k_proj/Linear/pre_reshape", "/layers.10/self_attn/k_proj/Linear/pre_convert", "/layers.10/self_attn/k_proj/Linear", "/layers.10/self_attn/k_proj/Linear/post_convert", "/layers.10/self_attn/v_proj/Linear/pre_reshape", "/layers.10/self_attn/v_proj/Linear/pre_convert", "/layers.10/self_attn/v_proj/Linear", "/layers.10/self_attn/v_proj/Linear/post_convert", "/layers.10/self_attn/o_proj/Linear/pre_reshape", "/layers.10/self_attn/o_proj/Linear/pre_convert", "/layers.10/self_attn/o_proj/Linear", "/layers.10/self_attn/o_proj/Linear/post_convert", "/layers.10/mlp/gate_proj/Linear/pre_reshape", "/layers.10/mlp/gate_proj/Linear/pre_convert", "/layers.10/mlp/gate_proj/Linear", "/layers.10/mlp/gate_proj/Linear/post_convert", "/layers.10/mlp/up_proj/Linear/pre_reshape", "/layers.10/mlp/up_proj/Linear/pre_convert", "/layers.10/mlp/up_proj/Linear", "/layers.10/mlp/up_proj/Linear/post_convert", "/layers.10/mlp/down_proj/Linear/pre_reshape", "/layers.10/mlp/down_proj/Linear/pre_convert", "/layers.10/mlp/down_proj/Linear", "/layers.10/mlp/down_proj/Linear/post_convert", "/layers.11/self_attn/q_proj/Linear/pre_reshape", "/layers.11/self_attn/q_proj/Linear/pre_convert", "/layers.11/self_attn/q_proj/Linear", "/layers.11/self_attn/q_proj/Linear/post_convert", "/layers.11/self_attn/k_proj/Linear/pre_reshape", "/layers.11/self_attn/k_proj/Linear/pre_convert", "/layers.11/self_attn/k_proj/Linear", "/layers.11/self_attn/k_proj/Linear/post_convert", "/layers.11/self_attn/v_proj/Linear/pre_reshape", "/layers.11/self_attn/v_proj/Linear/pre_convert", "/layers.11/self_attn/v_proj/Linear", "/layers.11/self_attn/v_proj/Linear/post_convert", "/layers.11/self_attn/o_proj/Linear/pre_reshape", "/layers.11/self_attn/o_proj/Linear/pre_convert", "/layers.11/self_attn/o_proj/Linear", "/layers.11/self_attn/o_proj/Linear/post_convert", "/layers.11/mlp/gate_proj/Linear/pre_reshape", "/layers.11/mlp/gate_proj/Linear/pre_convert", "/layers.11/mlp/gate_proj/Linear", "/layers.11/mlp/gate_proj/Linear/post_convert", "/layers.11/mlp/up_proj/Linear/pre_reshape", "/layers.11/mlp/up_proj/Linear/pre_convert", "/layers.11/mlp/up_proj/Linear", "/layers.11/mlp/up_proj/Linear/post_convert", "/layers.11/mlp/down_proj/Linear/pre_reshape", "/layers.11/mlp/down_proj/Linear/pre_convert", "/layers.11/mlp/down_proj/Linear", "/layers.11/mlp/down_proj/Linear/post_convert", "/layers.12/self_attn/q_proj/Linear/pre_reshape", "/layers.12/self_attn/q_proj/Linear/pre_convert", "/layers.12/self_attn/q_proj/Linear", "/layers.12/self_attn/q_proj/Linear/post_convert", "/layers.12/self_attn/k_proj/Linear/pre_reshape", "/layers.12/self_attn/k_proj/Linear/pre_convert", "/layers.12/self_attn/k_proj/Linear", "/layers.12/self_attn/k_proj/Linear/post_convert", "/layers.12/self_attn/v_proj/Linear/pre_reshape", "/layers.12/self_attn/v_proj/Linear/pre_convert", "/layers.12/self_attn/v_proj/Linear", "/layers.12/self_attn/v_proj/Linear/post_convert", "/layers.12/self_attn/o_proj/Linear/pre_reshape", "/layers.12/self_attn/o_proj/Linear/pre_convert", "/layers.12/self_attn/o_proj/Linear", "/layers.12/self_attn/o_proj/Linear/post_convert", "/layers.12/mlp/gate_proj/Linear/pre_reshape", "/layers.12/mlp/gate_proj/Linear/pre_convert", "/layers.12/mlp/gate_proj/Linear", "/layers.12/mlp/gate_proj/Linear/post_convert", "/layers.12/mlp/up_proj/Linear/pre_reshape", "/layers.12/mlp/up_proj/Linear/pre_convert", "/layers.12/mlp/up_proj/Linear", "/layers.12/mlp/up_proj/Linear/post_convert", "/layers.12/mlp/down_proj/Linear/pre_reshape", "/layers.12/mlp/down_proj/Linear/pre_convert", "/layers.12/mlp/down_proj/Linear", "/layers.12/mlp/down_proj/Linear/post_convert", "/layers.13/self_attn/q_proj/Linear/pre_reshape", "/layers.13/self_attn/q_proj/Linear/pre_convert", "/layers.13/self_attn/q_proj/Linear", "/layers.13/self_attn/q_proj/Linear/post_convert", "/layers.13/self_attn/k_proj/Linear/pre_reshape", "/layers.13/self_attn/k_proj/Linear/pre_convert", "/layers.13/self_attn/k_proj/Linear", "/layers.13/self_attn/k_proj/Linear/post_convert", "/layers.13/self_attn/v_proj/Linear/pre_reshape", "/layers.13/self_attn/v_proj/Linear/pre_convert", "/layers.13/self_attn/v_proj/Linear", "/layers.13/self_attn/v_proj/Linear/post_convert", "/layers.13/self_attn/o_proj/Linear/pre_reshape", "/layers.13/self_attn/o_proj/Linear/pre_convert", "/layers.13/self_attn/o_proj/Linear", "/layers.13/self_attn/o_proj/Linear/post_convert", "/layers.13/mlp/gate_proj/Linear/pre_reshape", "/layers.13/mlp/gate_proj/Linear/pre_convert", "/layers.13/mlp/gate_proj/Linear", "/layers.13/mlp/gate_proj/Linear/post_convert", "/layers.13/mlp/up_proj/Linear/pre_reshape", "/layers.13/mlp/up_proj/Linear/pre_convert", "/layers.13/mlp/up_proj/Linear", "/layers.13/mlp/up_proj/Linear/post_convert", "/layers.13/mlp/down_proj/Linear/pre_reshape", "/layers.13/mlp/down_proj/Linear/pre_convert", "/layers.13/mlp/down_proj/Linear", "/layers.13/mlp/down_proj/Linear/post_convert", "/layers.14/self_attn/q_proj/Linear/pre_reshape", "/layers.14/self_attn/q_proj/Linear/pre_convert", "/layers.14/self_attn/q_proj/Linear", "/layers.14/self_attn/q_proj/Linear/post_convert", "/layers.14/self_attn/k_proj/Linear/pre_reshape", "/layers.14/self_attn/k_proj/Linear/pre_convert", "/layers.14/self_attn/k_proj/Linear", "/layers.14/self_attn/k_proj/Linear/post_convert", "/layers.14/self_attn/v_proj/Linear/pre_reshape", "/layers.14/self_attn/v_proj/Linear/pre_convert", "/layers.14/self_attn/v_proj/Linear", "/layers.14/self_attn/v_proj/Linear/post_convert", "/layers.14/self_attn/o_proj/Linear/pre_reshape", "/layers.14/self_attn/o_proj/Linear/pre_convert", "/layers.14/self_attn/o_proj/Linear", "/layers.14/self_attn/o_proj/Linear/post_convert", "/layers.14/mlp/gate_proj/Linear/pre_reshape", "/layers.14/mlp/gate_proj/Linear/pre_convert", "/layers.14/mlp/gate_proj/Linear", "/layers.14/mlp/gate_proj/Linear/post_convert", "/layers.14/mlp/up_proj/Linear/pre_reshape", "/layers.14/mlp/up_proj/Linear/pre_convert", "/layers.14/mlp/up_proj/Linear", "/layers.14/mlp/up_proj/Linear/post_convert", "/layers.14/mlp/down_proj/Linear/pre_reshape", "/layers.14/mlp/down_proj/Linear/pre_convert", "/layers.14/mlp/down_proj/Linear", "/layers.14/mlp/down_proj/Linear/post_convert", "/layers.15/self_attn/q_proj/Linear/pre_reshape", "/layers.15/self_attn/q_proj/Linear/pre_convert", "/layers.15/self_attn/q_proj/Linear", "/layers.15/self_attn/q_proj/Linear/post_convert", "/layers.15/self_attn/k_proj/Linear/pre_reshape", "/layers.15/self_attn/k_proj/Linear/pre_convert", "/layers.15/self_attn/k_proj/Linear", "/layers.15/self_attn/k_proj/Linear/post_convert", "/layers.15/self_attn/v_proj/Linear/pre_reshape", "/layers.15/self_attn/v_proj/Linear/pre_convert", "/layers.15/self_attn/v_proj/Linear", "/layers.15/self_attn/v_proj/Linear/post_convert", "/layers.15/self_attn/o_proj/Linear/pre_reshape", "/layers.15/self_attn/o_proj/Linear/pre_convert", "/layers.15/self_attn/o_proj/Linear", "/layers.15/self_attn/o_proj/Linear/post_convert", "/layers.15/mlp/gate_proj/Linear/pre_reshape", "/layers.15/mlp/gate_proj/Linear/pre_convert", "/layers.15/mlp/gate_proj/Linear", "/layers.15/mlp/gate_proj/Linear/post_convert", "/layers.15/mlp/up_proj/Linear/pre_reshape", "/layers.15/mlp/up_proj/Linear/pre_convert", "/layers.15/mlp/up_proj/Linear", "/layers.15/mlp/up_proj/Linear/post_convert", "/layers.15/mlp/down_proj/Linear/pre_reshape", "/layers.15/mlp/down_proj/Linear/pre_convert", "/layers.15/mlp/down_proj/Linear", "/layers.15/mlp/down_proj/Linear/post_convert", "/layers.16/self_attn/q_proj/Linear/pre_reshape", "/layers.16/self_attn/q_proj/Linear/pre_convert", "/layers.16/self_attn/q_proj/Linear", "/layers.16/self_attn/q_proj/Linear/post_convert", "/layers.16/self_attn/k_proj/Linear/pre_reshape", "/layers.16/self_attn/k_proj/Linear/pre_convert", "/layers.16/self_attn/k_proj/Linear", "/layers.16/self_attn/k_proj/Linear/post_convert", "/layers.16/self_attn/v_proj/Linear/pre_reshape", "/layers.16/self_attn/v_proj/Linear/pre_convert", "/layers.16/self_attn/v_proj/Linear", "/layers.16/self_attn/v_proj/Linear/post_convert", "/layers.16/self_attn/o_proj/Linear/pre_reshape", "/layers.16/self_attn/o_proj/Linear/pre_convert", "/layers.16/self_attn/o_proj/Linear", "/layers.16/self_attn/o_proj/Linear/post_convert", "/layers.16/mlp/gate_proj/Linear/pre_reshape", "/layers.16/mlp/gate_proj/Linear/pre_convert", "/layers.16/mlp/gate_proj/Linear", "/layers.16/mlp/gate_proj/Linear/post_convert", "/layers.16/mlp/up_proj/Linear/pre_reshape", "/layers.16/mlp/up_proj/Linear/pre_convert", "/layers.16/mlp/up_proj/Linear", "/layers.16/mlp/up_proj/Linear/post_convert", "/layers.16/mlp/down_proj/Linear/pre_reshape", "/layers.16/mlp/down_proj/Linear/pre_convert", "/layers.16/mlp/down_proj/Linear", "/layers.16/mlp/down_proj/Linear/post_convert", "/layers.17/self_attn/q_proj/Linear/pre_reshape", "/layers.17/self_attn/q_proj/Linear/pre_convert", "/layers.17/self_attn/q_proj/Linear", "/layers.17/self_attn/q_proj/Linear/post_convert", "/layers.17/self_attn/k_proj/Linear/pre_reshape", "/layers.17/self_attn/k_proj/Linear/pre_convert", "/layers.17/self_attn/k_proj/Linear", "/layers.17/self_attn/k_proj/Linear/post_convert", "/layers.17/self_attn/v_proj/Linear/pre_reshape", "/layers.17/self_attn/v_proj/Linear/pre_convert", "/layers.17/self_attn/v_proj/Linear", "/layers.17/self_attn/v_proj/Linear/post_convert", "/layers.17/self_attn/o_proj/Linear/pre_reshape", "/layers.17/self_attn/o_proj/Linear/pre_convert", "/layers.17/self_attn/o_proj/Linear", "/layers.17/self_attn/o_proj/Linear/post_convert", "/layers.17/mlp/gate_proj/Linear/pre_reshape", "/layers.17/mlp/gate_proj/Linear/pre_convert", "/layers.17/mlp/gate_proj/Linear", "/layers.17/mlp/gate_proj/Linear/post_convert", "/layers.17/mlp/up_proj/Linear/pre_reshape", "/layers.17/mlp/up_proj/Linear/pre_convert", "/layers.17/mlp/up_proj/Linear", "/layers.17/mlp/up_proj/Linear/post_convert", "/layers.17/mlp/down_proj/Linear/pre_reshape", "/layers.17/mlp/down_proj/Linear/pre_convert", "/layers.17/mlp/down_proj/Linear", "/layers.17/mlp/down_proj/Linear/post_convert", "/layers.18/self_attn/q_proj/Linear/pre_reshape", "/layers.18/self_attn/q_proj/Linear/pre_convert", "/layers.18/self_attn/q_proj/Linear", "/layers.18/self_attn/q_proj/Linear/post_convert", "/layers.18/self_attn/k_proj/Linear/pre_reshape", "/layers.18/self_attn/k_proj/Linear/pre_convert", "/layers.18/self_attn/k_proj/Linear", "/layers.18/self_attn/k_proj/Linear/post_convert", "/layers.18/self_attn/v_proj/Linear/pre_reshape", "/layers.18/self_attn/v_proj/Linear/pre_convert", "/layers.18/self_attn/v_proj/Linear", "/layers.18/self_attn/v_proj/Linear/post_convert", "/layers.18/self_attn/o_proj/Linear/pre_reshape", "/layers.18/self_attn/o_proj/Linear/pre_convert", "/layers.18/self_attn/o_proj/Linear", "/layers.18/self_attn/o_proj/Linear/post_convert", "/layers.18/mlp/gate_proj/Linear/pre_reshape", "/layers.18/mlp/gate_proj/Linear/pre_convert", "/layers.18/mlp/gate_proj/Linear", "/layers.18/mlp/gate_proj/Linear/post_convert", "/layers.18/mlp/up_proj/Linear/pre_reshape", "/layers.18/mlp/up_proj/Linear/pre_convert", "/layers.18/mlp/up_proj/Linear", "/layers.18/mlp/up_proj/Linear/post_convert", "/layers.18/mlp/down_proj/Linear/pre_reshape", "/layers.18/mlp/down_proj/Linear/pre_convert", "/layers.18/mlp/down_proj/Linear", "/layers.18/mlp/down_proj/Linear/post_convert", "/layers.19/self_attn/q_proj/Linear/pre_reshape", "/layers.19/self_attn/q_proj/Linear/pre_convert", "/layers.19/self_attn/q_proj/Linear", "/layers.19/self_attn/q_proj/Linear/post_convert", "/layers.19/self_attn/k_proj/Linear/pre_reshape", "/layers.19/self_attn/k_proj/Linear/pre_convert", "/layers.19/self_attn/k_proj/Linear", "/layers.19/self_attn/k_proj/Linear/post_convert", "/layers.19/self_attn/v_proj/Linear/pre_reshape", "/layers.19/self_attn/v_proj/Linear/pre_convert", "/layers.19/self_attn/v_proj/Linear", "/layers.19/self_attn/v_proj/Linear/post_convert", "/layers.19/self_attn/o_proj/Linear/pre_reshape", "/layers.19/self_attn/o_proj/Linear/pre_convert", "/layers.19/self_attn/o_proj/Linear", "/layers.19/self_attn/o_proj/Linear/post_convert", "/layers.19/mlp/gate_proj/Linear/pre_reshape", "/layers.19/mlp/gate_proj/Linear/pre_convert", "/layers.19/mlp/gate_proj/Linear", "/layers.19/mlp/gate_proj/Linear/post_convert", "/layers.19/mlp/up_proj/Linear/pre_reshape", "/layers.19/mlp/up_proj/Linear/pre_convert", "/layers.19/mlp/up_proj/Linear", "/layers.19/mlp/up_proj/Linear/post_convert", "/layers.19/mlp/down_proj/Linear/pre_reshape", "/layers.19/mlp/down_proj/Linear/pre_convert", "/layers.19/mlp/down_proj/Linear", "/layers.19/mlp/down_proj/Linear/post_convert", "/layers.20/self_attn/q_proj/Linear/pre_reshape", "/layers.20/self_attn/q_proj/Linear/pre_convert", "/layers.20/self_attn/q_proj/Linear", "/layers.20/self_attn/q_proj/Linear/post_convert", "/layers.20/self_attn/k_proj/Linear/pre_reshape", "/layers.20/self_attn/k_proj/Linear/pre_convert", "/layers.20/self_attn/k_proj/Linear", "/layers.20/self_attn/k_proj/Linear/post_convert", "/layers.20/self_attn/v_proj/Linear/pre_reshape", "/layers.20/self_attn/v_proj/Linear/pre_convert", "/layers.20/self_attn/v_proj/Linear", "/layers.20/self_attn/v_proj/Linear/post_convert", "/layers.20/self_attn/o_proj/Linear/pre_reshape", "/layers.20/self_attn/o_proj/Linear/pre_convert", "/layers.20/self_attn/o_proj/Linear", "/layers.20/self_attn/o_proj/Linear/post_convert", "/layers.20/mlp/gate_proj/Linear/pre_reshape", "/layers.20/mlp/gate_proj/Linear/pre_convert", "/layers.20/mlp/gate_proj/Linear", "/layers.20/mlp/gate_proj/Linear/post_convert", "/layers.20/mlp/up_proj/Linear/pre_reshape", "/layers.20/mlp/up_proj/Linear/pre_convert", "/layers.20/mlp/up_proj/Linear", "/layers.20/mlp/up_proj/Linear/post_convert", "/layers.20/mlp/down_proj/Linear/pre_reshape", "/layers.20/mlp/down_proj/Linear/pre_convert", "/layers.20/mlp/down_proj/Linear", "/layers.20/mlp/down_proj/Linear/post_convert", "/layers.21/self_attn/q_proj/Linear/pre_reshape", "/layers.21/self_attn/q_proj/Linear/pre_convert", "/layers.21/self_attn/q_proj/Linear", "/layers.21/self_attn/q_proj/Linear/post_convert", "/layers.21/self_attn/k_proj/Linear/pre_reshape", "/layers.21/self_attn/k_proj/Linear/pre_convert", "/layers.21/self_attn/k_proj/Linear", "/layers.21/self_attn/k_proj/Linear/post_convert", "/layers.21/self_attn/v_proj/Linear/pre_reshape", "/layers.21/self_attn/v_proj/Linear/pre_convert", "/layers.21/self_attn/v_proj/Linear", "/layers.21/self_attn/v_proj/Linear/post_convert", "/layers.21/self_attn/o_proj/Linear/pre_reshape", "/layers.21/self_attn/o_proj/Linear/pre_convert", "/layers.21/self_attn/o_proj/Linear", "/layers.21/self_attn/o_proj/Linear/post_convert", "/layers.21/mlp/gate_proj/Linear/pre_reshape", "/layers.21/mlp/gate_proj/Linear/pre_convert", "/layers.21/mlp/gate_proj/Linear", "/layers.21/mlp/gate_proj/Linear/post_convert", "/layers.21/mlp/up_proj/Linear/pre_reshape", "/layers.21/mlp/up_proj/Linear/pre_convert", "/layers.21/mlp/up_proj/Linear", "/layers.21/mlp/up_proj/Linear/post_convert", "/layers.21/mlp/down_proj/Linear/pre_reshape", "/layers.21/mlp/down_proj/Linear/pre_convert", "/layers.21/mlp/down_proj/Linear", "/layers.21/mlp/down_proj/Linear/post_convert", "/layers.22/self_attn/q_proj/Linear/pre_reshape", "/layers.22/self_attn/q_proj/Linear/pre_convert", "/layers.22/self_attn/q_proj/Linear", "/layers.22/self_attn/q_proj/Linear/post_convert", "/layers.22/self_attn/k_proj/Linear/pre_reshape", "/layers.22/self_attn/k_proj/Linear/pre_convert", "/layers.22/self_attn/k_proj/Linear", "/layers.22/self_attn/k_proj/Linear/post_convert", "/layers.22/self_attn/v_proj/Linear/pre_reshape", "/layers.22/self_attn/v_proj/Linear/pre_convert", "/layers.22/self_attn/v_proj/Linear", "/layers.22/self_attn/v_proj/Linear/post_convert", "/layers.22/self_attn/o_proj/Linear/pre_reshape", "/layers.22/self_attn/o_proj/Linear/pre_convert", "/layers.22/self_attn/o_proj/Linear", "/layers.22/self_attn/o_proj/Linear/post_convert", "/layers.22/mlp/gate_proj/Linear/pre_reshape", "/layers.22/mlp/gate_proj/Linear/pre_convert", "/layers.22/mlp/gate_proj/Linear", "/layers.22/mlp/gate_proj/Linear/post_convert", "/layers.22/mlp/up_proj/Linear/pre_reshape", "/layers.22/mlp/up_proj/Linear/pre_convert", "/layers.22/mlp/up_proj/Linear", "/layers.22/mlp/up_proj/Linear/post_convert", "/layers.22/mlp/down_proj/Linear/pre_reshape", "/layers.22/mlp/down_proj/Linear/pre_convert", "/layers.22/mlp/down_proj/Linear", "/layers.22/mlp/down_proj/Linear/post_convert", "/layers.23/self_attn/q_proj/Linear/pre_reshape", "/layers.23/self_attn/q_proj/Linear/pre_convert", "/layers.23/self_attn/q_proj/Linear", "/layers.23/self_attn/q_proj/Linear/post_convert", "/layers.23/self_attn/k_proj/Linear/pre_reshape", "/layers.23/self_attn/k_proj/Linear/pre_convert", "/layers.23/self_attn/k_proj/Linear", "/layers.23/self_attn/k_proj/Linear/post_convert", "/layers.23/self_attn/v_proj/Linear/pre_reshape", "/layers.23/self_attn/v_proj/Linear/pre_convert", "/layers.23/self_attn/v_proj/Linear", "/layers.23/self_attn/v_proj/Linear/post_convert", "/layers.23/self_attn/o_proj/Linear/pre_reshape", "/layers.23/self_attn/o_proj/Linear/pre_convert", "/layers.23/self_attn/o_proj/Linear", "/layers.23/self_attn/o_proj/Linear/post_convert", "/layers.23/mlp/gate_proj/Linear/pre_reshape", "/layers.23/mlp/gate_proj/Linear/pre_convert", "/layers.23/mlp/gate_proj/Linear", "/layers.23/mlp/gate_proj/Linear/post_convert", "/layers.23/mlp/up_proj/Linear/pre_reshape", "/layers.23/mlp/up_proj/Linear/pre_convert", "/layers.23/mlp/up_proj/Linear", "/layers.23/mlp/up_proj/Linear/post_convert", "/layers.23/mlp/down_proj/Linear/pre_reshape", "/layers.23/mlp/down_proj/Linear/pre_convert", "/layers.23/mlp/down_proj/Linear", "/layers.23/mlp/down_proj/Linear/post_convert", "/layers.24/self_attn/q_proj/Linear/pre_reshape", "/layers.24/self_attn/q_proj/Linear/pre_convert", "/layers.24/self_attn/q_proj/Linear", "/layers.24/self_attn/q_proj/Linear/post_convert", "/layers.24/self_attn/k_proj/Linear/pre_reshape", "/layers.24/self_attn/k_proj/Linear/pre_convert", "/layers.24/self_attn/k_proj/Linear", "/layers.24/self_attn/k_proj/Linear/post_convert", "/layers.24/self_attn/v_proj/Linear/pre_reshape", "/layers.24/self_attn/v_proj/Linear/pre_convert", "/layers.24/self_attn/v_proj/Linear", "/layers.24/self_attn/v_proj/Linear/post_convert", "/layers.24/self_attn/o_proj/Linear/pre_reshape", "/layers.24/self_attn/o_proj/Linear/pre_convert", "/layers.24/self_attn/o_proj/Linear", "/layers.24/self_attn/o_proj/Linear/post_convert", "/layers.24/mlp/gate_proj/Linear/pre_reshape", "/layers.24/mlp/gate_proj/Linear/pre_convert", "/layers.24/mlp/gate_proj/Linear", "/layers.24/mlp/gate_proj/Linear/post_convert", "/layers.24/mlp/up_proj/Linear/pre_reshape", "/layers.24/mlp/up_proj/Linear/pre_convert", "/layers.24/mlp/up_proj/Linear", "/layers.24/mlp/up_proj/Linear/post_convert", "/layers.24/mlp/down_proj/Linear/pre_reshape", "/layers.24/mlp/down_proj/Linear/pre_convert", "/layers.24/mlp/down_proj/Linear", "/layers.24/mlp/down_proj/Linear/post_convert", "/layers.25/self_attn/q_proj/Linear/pre_reshape", "/layers.25/self_attn/q_proj/Linear/pre_convert", "/layers.25/self_attn/q_proj/Linear", "/layers.25/self_attn/q_proj/Linear/post_convert", "/layers.25/self_attn/k_proj/Linear/pre_reshape", "/layers.25/self_attn/k_proj/Linear/pre_convert", "/layers.25/self_attn/k_proj/Linear", "/layers.25/self_attn/k_proj/Linear/post_convert", "/layers.25/self_attn/v_proj/Linear/pre_reshape", "/layers.25/self_attn/v_proj/Linear/pre_convert", "/layers.25/self_attn/v_proj/Linear", "/layers.25/self_attn/v_proj/Linear/post_convert", "/layers.25/self_attn/o_proj/Linear/pre_reshape", "/layers.25/self_attn/o_proj/Linear/pre_convert", "/layers.25/self_attn/o_proj/Linear", "/layers.25/self_attn/o_proj/Linear/post_convert", "/layers.25/mlp/gate_proj/Linear/pre_reshape", "/layers.25/mlp/gate_proj/Linear/pre_convert", "/layers.25/mlp/gate_proj/Linear", "/layers.25/mlp/gate_proj/Linear/post_convert", "/layers.25/mlp/up_proj/Linear/pre_reshape", "/layers.25/mlp/up_proj/Linear/pre_convert", "/layers.25/mlp/up_proj/Linear", "/layers.25/mlp/up_proj/Linear/post_convert", "/layers.25/mlp/down_proj/Linear/pre_reshape", "/layers.25/mlp/down_proj/Linear/pre_convert", "/layers.25/mlp/down_proj/Linear", "/layers.25/mlp/down_proj/Linear/post_convert", "/layers.26/self_attn/q_proj/Linear/pre_reshape", "/layers.26/self_attn/q_proj/Linear/pre_convert", "/layers.26/self_attn/q_proj/Linear", "/layers.26/self_attn/q_proj/Linear/post_convert", "/layers.26/self_attn/k_proj/Linear/pre_reshape", "/layers.26/self_attn/k_proj/Linear/pre_convert", "/layers.26/self_attn/k_proj/Linear", "/layers.26/self_attn/k_proj/Linear/post_convert", "/layers.26/self_attn/v_proj/Linear/pre_reshape", "/layers.26/self_attn/v_proj/Linear/pre_convert", "/layers.26/self_attn/v_proj/Linear", "/layers.26/self_attn/v_proj/Linear/post_convert", "/layers.26/self_attn/o_proj/Linear/pre_reshape", "/layers.26/self_attn/o_proj/Linear/pre_convert", "/layers.26/self_attn/o_proj/Linear", "/layers.26/self_attn/o_proj/Linear/post_convert", "/layers.26/mlp/gate_proj/Linear/pre_reshape", "/layers.26/mlp/gate_proj/Linear/pre_convert", "/layers.26/mlp/gate_proj/Linear", "/layers.26/mlp/gate_proj/Linear/post_convert", "/layers.26/mlp/up_proj/Linear/pre_reshape", "/layers.26/mlp/up_proj/Linear/pre_convert", "/layers.26/mlp/up_proj/Linear", "/layers.26/mlp/up_proj/Linear/post_convert", "/layers.26/mlp/down_proj/Linear/pre_reshape", "/layers.26/mlp/down_proj/Linear/pre_convert", "/layers.26/mlp/down_proj/Linear", "/layers.26/mlp/down_proj/Linear/post_convert", "/layers.27/self_attn/q_proj/Linear/pre_reshape", "/layers.27/self_attn/q_proj/Linear/pre_convert", "/layers.27/self_attn/q_proj/Linear", "/layers.27/self_attn/q_proj/Linear/post_convert", "/layers.27/self_attn/k_proj/Linear/pre_reshape", "/layers.27/self_attn/k_proj/Linear/pre_convert", "/layers.27/self_attn/k_proj/Linear", "/layers.27/self_attn/k_proj/Linear/post_convert", "/layers.27/self_attn/v_proj/Linear/pre_reshape", "/layers.27/self_attn/v_proj/Linear/pre_convert", "/layers.27/self_attn/v_proj/Linear", "/layers.27/self_attn/v_proj/Linear/post_convert", "/layers.27/self_attn/o_proj/Linear/pre_reshape", "/layers.27/self_attn/o_proj/Linear/pre_convert", "/layers.27/self_attn/o_proj/Linear", "/layers.27/self_attn/o_proj/Linear/post_convert", "/layers.27/mlp/gate_proj/Linear/pre_reshape", "/layers.27/mlp/gate_proj/Linear/pre_convert", "/layers.27/mlp/gate_proj/Linear", "/layers.27/mlp/gate_proj/Linear/post_convert", "/layers.27/mlp/up_proj/Linear/pre_reshape", "/layers.27/mlp/up_proj/Linear/pre_convert", "/layers.27/mlp/up_proj/Linear", "/layers.27/mlp/up_proj/Linear/post_convert", "/layers.27/mlp/down_proj/Linear/pre_reshape", "/layers.27/mlp/down_proj/Linear/pre_convert", "/layers.27/mlp/down_proj/Linear", "/layers.27/mlp/down_proj/Linear/post_convert", "/layers.28/self_attn/q_proj/Linear/pre_reshape", "/layers.28/self_attn/q_proj/Linear/pre_convert", "/layers.28/self_attn/q_proj/Linear", "/layers.28/self_attn/q_proj/Linear/post_convert", "/layers.28/self_attn/k_proj/Linear/pre_reshape", "/layers.28/self_attn/k_proj/Linear/pre_convert", "/layers.28/self_attn/k_proj/Linear", "/layers.28/self_attn/k_proj/Linear/post_convert", "/layers.28/self_attn/v_proj/Linear/pre_reshape", "/layers.28/self_attn/v_proj/Linear/pre_convert", "/layers.28/self_attn/v_proj/Linear", "/layers.28/self_attn/v_proj/Linear/post_convert", "/layers.28/self_attn/o_proj/Linear/pre_reshape", "/layers.28/self_attn/o_proj/Linear/pre_convert", "/layers.28/self_attn/o_proj/Linear", "/layers.28/self_attn/o_proj/Linear/post_convert", "/layers.28/mlp/gate_proj/Linear/pre_reshape", "/layers.28/mlp/gate_proj/Linear/pre_convert", "/layers.28/mlp/gate_proj/Linear", "/layers.28/mlp/gate_proj/Linear/post_convert", "/layers.28/mlp/up_proj/Linear/pre_reshape", "/layers.28/mlp/up_proj/Linear/pre_convert", "/layers.28/mlp/up_proj/Linear", "/layers.28/mlp/up_proj/Linear/post_convert", "/layers.28/mlp/down_proj/Linear/pre_reshape", "/layers.28/mlp/down_proj/Linear/pre_convert", "/layers.28/mlp/down_proj/Linear", "/layers.28/mlp/down_proj/Linear/post_convert", "/layers.29/self_attn/q_proj/Linear/pre_reshape", "/layers.29/self_attn/q_proj/Linear/pre_convert", "/layers.29/self_attn/q_proj/Linear", "/layers.29/self_attn/q_proj/Linear/post_convert", "/layers.29/self_attn/k_proj/Linear/pre_reshape", "/layers.29/self_attn/k_proj/Linear/pre_convert", "/layers.29/self_attn/k_proj/Linear", "/layers.29/self_attn/k_proj/Linear/post_convert", "/layers.29/self_attn/v_proj/Linear/pre_reshape", "/layers.29/self_attn/v_proj/Linear/pre_convert", "/layers.29/self_attn/v_proj/Linear", "/layers.29/self_attn/v_proj/Linear/post_convert", "/layers.29/self_attn/o_proj/Linear/pre_reshape", "/layers.29/self_attn/o_proj/Linear/pre_convert", "/layers.29/self_attn/o_proj/Linear", "/layers.29/self_attn/o_proj/Linear/post_convert", "/layers.29/mlp/gate_proj/Linear/pre_reshape", "/layers.29/mlp/gate_proj/Linear/pre_convert", "/layers.29/mlp/gate_proj/Linear", "/layers.29/mlp/gate_proj/Linear/post_convert", "/layers.29/mlp/up_proj/Linear/pre_reshape", "/layers.29/mlp/up_proj/Linear/pre_convert", "/layers.29/mlp/up_proj/Linear", "/layers.29/mlp/up_proj/Linear/post_convert", "/layers.29/mlp/down_proj/Linear/pre_reshape", "/layers.29/mlp/down_proj/Linear/pre_convert", "/layers.29/mlp/down_proj/Linear", "/layers.29/mlp/down_proj/Linear/post_convert", "/layers.30/self_attn/q_proj/Linear/pre_reshape", "/layers.30/self_attn/q_proj/Linear/pre_convert", "/layers.30/self_attn/q_proj/Linear", "/layers.30/self_attn/q_proj/Linear/post_convert", "/layers.30/self_attn/k_proj/Linear/pre_reshape", "/layers.30/self_attn/k_proj/Linear/pre_convert", "/layers.30/self_attn/k_proj/Linear", "/layers.30/self_attn/k_proj/Linear/post_convert", "/layers.30/self_attn/v_proj/Linear/pre_reshape", "/layers.30/self_attn/v_proj/Linear/pre_convert", "/layers.30/self_attn/v_proj/Linear", "/layers.30/self_attn/v_proj/Linear/post_convert", "/layers.30/self_attn/o_proj/Linear/pre_reshape", "/layers.30/self_attn/o_proj/Linear/pre_convert", "/layers.30/self_attn/o_proj/Linear", "/layers.30/self_attn/o_proj/Linear/post_convert", "/layers.30/mlp/gate_proj/Linear/pre_reshape", "/layers.30/mlp/gate_proj/Linear/pre_convert", "/layers.30/mlp/gate_proj/Linear", "/layers.30/mlp/gate_proj/Linear/post_convert", "/layers.30/mlp/up_proj/Linear/pre_reshape", "/layers.30/mlp/up_proj/Linear/pre_convert", "/layers.30/mlp/up_proj/Linear", "/layers.30/mlp/up_proj/Linear/post_convert", "/layers.30/mlp/down_proj/Linear/pre_reshape", "/layers.30/mlp/down_proj/Linear/pre_convert", "/layers.30/mlp/down_proj/Linear", "/layers.30/mlp/down_proj/Linear/post_convert", "/layers.31/self_attn/q_proj/Linear/pre_reshape", "/layers.31/self_attn/q_proj/Linear/pre_convert", "/layers.31/self_attn/q_proj/Linear", "/layers.31/self_attn/q_proj/Linear/post_convert", "/layers.31/self_attn/k_proj/Linear/pre_reshape", "/layers.31/self_attn/k_proj/Linear/pre_convert", "/layers.31/self_attn/k_proj/Linear", "/layers.31/self_attn/k_proj/Linear/post_convert", "/layers.31/self_attn/v_proj/Linear/pre_reshape", "/layers.31/self_attn/v_proj/Linear/pre_convert", "/layers.31/self_attn/v_proj/Linear", "/layers.31/self_attn/v_proj/Linear/post_convert", "/layers.31/self_attn/o_proj/Linear/pre_reshape", "/layers.31/self_attn/o_proj/Linear/pre_convert", "/layers.31/self_attn/o_proj/Linear", "/layers.31/self_attn/o_proj/Linear/post_convert", "/layers.31/mlp/gate_proj/Linear/pre_reshape", "/layers.31/mlp/gate_proj/Linear/pre_convert", "/layers.31/mlp/gate_proj/Linear", "/layers.31/mlp/gate_proj/Linear/post_convert", "/layers.31/mlp/up_proj/Linear/pre_reshape", "/layers.31/mlp/up_proj/Linear/pre_convert", "/layers.31/mlp/up_proj/Linear", "/layers.31/mlp/up_proj/Linear/post_convert", "/layers.31/mlp/down_proj/Linear/pre_reshape", "/layers.31/mlp/down_proj/Linear/pre_convert", "/layers.31/mlp/down_proj/Linear", "/layers.31/mlp/down_proj/Linear/post_convert", "/layers.32/self_attn/q_proj/Linear/pre_reshape", "/layers.32/self_attn/q_proj/Linear/pre_convert", "/layers.32/self_attn/q_proj/Linear", "/layers.32/self_attn/q_proj/Linear/post_convert", "/layers.32/self_attn/k_proj/Linear/pre_reshape", "/layers.32/self_attn/k_proj/Linear/pre_convert", "/layers.32/self_attn/k_proj/Linear", "/layers.32/self_attn/k_proj/Linear/post_convert", "/layers.32/self_attn/v_proj/Linear/pre_reshape", "/layers.32/self_attn/v_proj/Linear/pre_convert", "/layers.32/self_attn/v_proj/Linear", "/layers.32/self_attn/v_proj/Linear/post_convert", "/layers.32/self_attn/o_proj/Linear/pre_reshape", "/layers.32/self_attn/o_proj/Linear/pre_convert", "/layers.32/self_attn/o_proj/Linear", "/layers.32/self_attn/o_proj/Linear/post_convert", "/layers.32/mlp/gate_proj/Linear/pre_reshape", "/layers.32/mlp/gate_proj/Linear/pre_convert", "/layers.32/mlp/gate_proj/Linear", "/layers.32/mlp/gate_proj/Linear/post_convert", "/layers.32/mlp/up_proj/Linear/pre_reshape", "/layers.32/mlp/up_proj/Linear/pre_convert", "/layers.32/mlp/up_proj/Linear", "/layers.32/mlp/up_proj/Linear/post_convert", "/layers.32/mlp/down_proj/Linear/pre_reshape", "/layers.32/mlp/down_proj/Linear/pre_convert", "/layers.32/mlp/down_proj/Linear", "/layers.32/mlp/down_proj/Linear/post_convert", "/layers.33/self_attn/q_proj/Linear/pre_reshape", "/layers.33/self_attn/q_proj/Linear/pre_convert", "/layers.33/self_attn/q_proj/Linear", "/layers.33/self_attn/q_proj/Linear/post_convert", "/layers.33/self_attn/k_proj/Linear/pre_reshape", "/layers.33/self_attn/k_proj/Linear/pre_convert", "/layers.33/self_attn/k_proj/Linear", "/layers.33/self_attn/k_proj/Linear/post_convert", "/layers.33/self_attn/v_proj/Linear/pre_reshape", "/layers.33/self_attn/v_proj/Linear/pre_convert", "/layers.33/self_attn/v_proj/Linear", "/layers.33/self_attn/v_proj/Linear/post_convert", "/layers.33/self_attn/o_proj/Linear/pre_reshape", "/layers.33/self_attn/o_proj/Linear/pre_convert", "/layers.33/self_attn/o_proj/Linear", "/layers.33/self_attn/o_proj/Linear/post_convert", "/layers.33/mlp/gate_proj/Linear/pre_reshape", "/layers.33/mlp/gate_proj/Linear/pre_convert", "/layers.33/mlp/gate_proj/Linear", "/layers.33/mlp/gate_proj/Linear/post_convert", "/layers.33/mlp/up_proj/Linear/pre_reshape", "/layers.33/mlp/up_proj/Linear/pre_convert", "/layers.33/mlp/up_proj/Linear", "/layers.33/mlp/up_proj/Linear/post_convert", "/layers.33/mlp/down_proj/Linear/pre_reshape", "/layers.33/mlp/down_proj/Linear/pre_convert", "/layers.33/mlp/down_proj/Linear", "/layers.33/mlp/down_proj/Linear/post_convert", "/layers.34/self_attn/q_proj/Linear/pre_reshape", "/layers.34/self_attn/q_proj/Linear/pre_convert", "/layers.34/self_attn/q_proj/Linear", "/layers.34/self_attn/q_proj/Linear/post_convert", "/layers.34/self_attn/k_proj/Linear/pre_reshape", "/layers.34/self_attn/k_proj/Linear/pre_convert", "/layers.34/self_attn/k_proj/Linear", "/layers.34/self_attn/k_proj/Linear/post_convert", "/layers.34/self_attn/v_proj/Linear/pre_reshape", "/layers.34/self_attn/v_proj/Linear/pre_convert", "/layers.34/self_attn/v_proj/Linear", "/layers.34/self_attn/v_proj/Linear/post_convert", "/layers.34/self_attn/o_proj/Linear/pre_reshape", "/layers.34/self_attn/o_proj/Linear/pre_convert", "/layers.34/self_attn/o_proj/Linear", "/layers.34/self_attn/o_proj/Linear/post_convert", "/layers.34/mlp/gate_proj/Linear/pre_reshape", "/layers.34/mlp/gate_proj/Linear/pre_convert", "/layers.34/mlp/gate_proj/Linear", "/layers.34/mlp/gate_proj/Linear/post_convert", "/layers.34/mlp/up_proj/Linear/pre_reshape", "/layers.34/mlp/up_proj/Linear/pre_convert", "/layers.34/mlp/up_proj/Linear", "/layers.34/mlp/up_proj/Linear/post_convert", "/layers.34/mlp/down_proj/Linear/pre_reshape", "/layers.34/mlp/down_proj/Linear/pre_convert", "/layers.34/mlp/down_proj/Linear", "/layers.34/mlp/down_proj/Linear/post_convert", "/layers.35/self_attn/q_proj/Linear/pre_reshape", "/layers.35/self_attn/q_proj/Linear/pre_convert", "/layers.35/self_attn/q_proj/Linear", "/layers.35/self_attn/q_proj/Linear/post_convert", "/layers.35/self_attn/k_proj/Linear/pre_reshape", "/layers.35/self_attn/k_proj/Linear/pre_convert", "/layers.35/self_attn/k_proj/Linear", "/layers.35/self_attn/k_proj/Linear/post_convert", "/layers.35/self_attn/v_proj/Linear/pre_reshape", "/layers.35/self_attn/v_proj/Linear/pre_convert", "/layers.35/self_attn/v_proj/Linear", "/layers.35/self_attn/v_proj/Linear/post_convert", "/layers.35/self_attn/o_proj/Linear/pre_reshape", "/layers.35/self_attn/o_proj/Linear/pre_convert", "/layers.35/self_attn/o_proj/Linear", "/layers.35/self_attn/o_proj/Linear/post_convert", "/layers.35/mlp/gate_proj/Linear/pre_reshape", "/layers.35/mlp/gate_proj/Linear/pre_convert", "/layers.35/mlp/gate_proj/Linear", "/layers.35/mlp/gate_proj/Linear/post_convert", "/layers.35/mlp/up_proj/Linear/pre_reshape", "/layers.35/mlp/up_proj/Linear/pre_convert", "/layers.35/mlp/up_proj/Linear", "/layers.35/mlp/up_proj/Linear/post_convert", "/layers.35/mlp/down_proj/Linear/pre_reshape", "/layers.35/mlp/down_proj/Linear/pre_convert", "/layers.35/mlp/down_proj/Linear", "/layers.35/mlp/down_proj/Linear/post_convert", "/layers.36/self_attn/q_proj/Linear/pre_reshape", "/layers.36/self_attn/q_proj/Linear/pre_convert", "/layers.36/self_attn/q_proj/Linear", "/layers.36/self_attn/q_proj/Linear/post_convert", "/layers.36/self_attn/k_proj/Linear/pre_reshape", "/layers.36/self_attn/k_proj/Linear/pre_convert", "/layers.36/self_attn/k_proj/Linear", "/layers.36/self_attn/k_proj/Linear/post_convert", "/layers.36/self_attn/v_proj/Linear/pre_reshape", "/layers.36/self_attn/v_proj/Linear/pre_convert", "/layers.36/self_attn/v_proj/Linear", "/layers.36/self_attn/v_proj/Linear/post_convert", "/layers.36/self_attn/o_proj/Linear/pre_reshape", "/layers.36/self_attn/o_proj/Linear/pre_convert", "/layers.36/self_attn/o_proj/Linear", "/layers.36/self_attn/o_proj/Linear/post_convert", "/layers.36/mlp/gate_proj/Linear/pre_reshape", "/layers.36/mlp/gate_proj/Linear/pre_convert", "/layers.36/mlp/gate_proj/Linear", "/layers.36/mlp/gate_proj/Linear/post_convert", "/layers.36/mlp/up_proj/Linear/pre_reshape", "/layers.36/mlp/up_proj/Linear/pre_convert", "/layers.36/mlp/up_proj/Linear", "/layers.36/mlp/up_proj/Linear/post_convert", "/layers.36/mlp/down_proj/Linear/pre_reshape", "/layers.36/mlp/down_proj/Linear/pre_convert", "/layers.36/mlp/down_proj/Linear", "/layers.36/mlp/down_proj/Linear/post_convert", "/layers.37/self_attn/q_proj/Linear/pre_reshape", "/layers.37/self_attn/q_proj/Linear/pre_convert", "/layers.37/self_attn/q_proj/Linear", "/layers.37/self_attn/q_proj/Linear/post_convert", "/layers.37/self_attn/k_proj/Linear/pre_reshape", "/layers.37/self_attn/k_proj/Linear/pre_convert", "/layers.37/self_attn/k_proj/Linear", "/layers.37/self_attn/k_proj/Linear/post_convert", "/layers.37/self_attn/v_proj/Linear/pre_reshape", "/layers.37/self_attn/v_proj/Linear/pre_convert", "/layers.37/self_attn/v_proj/Linear", "/layers.37/self_attn/v_proj/Linear/post_convert", "/layers.37/self_attn/o_proj/Linear/pre_reshape", "/layers.37/self_attn/o_proj/Linear/pre_convert", "/layers.37/self_attn/o_proj/Linear", "/layers.37/self_attn/o_proj/Linear/post_convert", "/layers.37/mlp/gate_proj/Linear/pre_reshape", "/layers.37/mlp/gate_proj/Linear/pre_convert", "/layers.37/mlp/gate_proj/Linear", "/layers.37/mlp/gate_proj/Linear/post_convert", "/layers.37/mlp/up_proj/Linear/pre_reshape", "/layers.37/mlp/up_proj/Linear/pre_convert", "/layers.37/mlp/up_proj/Linear", "/layers.37/mlp/up_proj/Linear/post_convert", "/layers.37/mlp/down_proj/Linear/pre_reshape", "/layers.37/mlp/down_proj/Linear/pre_convert", "/layers.37/mlp/down_proj/Linear", "/layers.37/mlp/down_proj/Linear/post_convert", "/layers.38/self_attn/q_proj/Linear/pre_reshape", "/layers.38/self_attn/q_proj/Linear/pre_convert", "/layers.38/self_attn/q_proj/Linear", "/layers.38/self_attn/q_proj/Linear/post_convert", "/layers.38/self_attn/k_proj/Linear/pre_reshape", "/layers.38/self_attn/k_proj/Linear/pre_convert", "/layers.38/self_attn/k_proj/Linear", "/layers.38/self_attn/k_proj/Linear/post_convert", "/layers.38/self_attn/v_proj/Linear/pre_reshape", "/layers.38/self_attn/v_proj/Linear/pre_convert", "/layers.38/self_attn/v_proj/Linear", "/layers.38/self_attn/v_proj/Linear/post_convert", "/layers.38/self_attn/o_proj/Linear/pre_reshape", "/layers.38/self_attn/o_proj/Linear/pre_convert", "/layers.38/self_attn/o_proj/Linear", "/layers.38/self_attn/o_proj/Linear/post_convert", "/layers.38/mlp/gate_proj/Linear/pre_reshape", "/layers.38/mlp/gate_proj/Linear/pre_convert", "/layers.38/mlp/gate_proj/Linear", "/layers.38/mlp/gate_proj/Linear/post_convert", "/layers.38/mlp/up_proj/Linear/pre_reshape", "/layers.38/mlp/up_proj/Linear/pre_convert", "/layers.38/mlp/up_proj/Linear", "/layers.38/mlp/up_proj/Linear/post_convert", "/layers.38/mlp/down_proj/Linear/pre_reshape", "/layers.38/mlp/down_proj/Linear/pre_convert", "/layers.38/mlp/down_proj/Linear", "/layers.38/mlp/down_proj/Linear/post_convert", "/layers.39/self_attn/q_proj/Linear/pre_reshape", "/layers.39/self_attn/q_proj/Linear/pre_convert", "/layers.39/self_attn/q_proj/Linear", "/layers.39/self_attn/q_proj/Linear/post_convert", "/layers.39/self_attn/k_proj/Linear/pre_reshape", "/layers.39/self_attn/k_proj/Linear/pre_convert", "/layers.39/self_attn/k_proj/Linear", "/layers.39/self_attn/k_proj/Linear/post_convert", "/layers.39/self_attn/v_proj/Linear/pre_reshape", "/layers.39/self_attn/v_proj/Linear/pre_convert", "/layers.39/self_attn/v_proj/Linear", "/layers.39/self_attn/v_proj/Linear/post_convert", "/layers.39/self_attn/o_proj/Linear/pre_reshape", "/layers.39/self_attn/o_proj/Linear/pre_convert", "/layers.39/self_attn/o_proj/Linear", "/layers.39/self_attn/o_proj/Linear/post_convert", "/layers.39/mlp/gate_proj/Linear/pre_reshape", "/layers.39/mlp/gate_proj/Linear/pre_convert", "/layers.39/mlp/gate_proj/Linear", "/layers.39/mlp/gate_proj/Linear/post_convert", "/layers.39/mlp/up_proj/Linear/pre_reshape", "/layers.39/mlp/up_proj/Linear/pre_convert", "/layers.39/mlp/up_proj/Linear", "/layers.39/mlp/up_proj/Linear/post_convert", "/layers.39/mlp/down_proj/Linear/pre_reshape", "/layers.39/mlp/down_proj/Linear/pre_convert", "/layers.39/mlp/down_proj/Linear", "/layers.39/mlp/down_proj/Linear/post_convert", "/layers.40/self_attn/q_proj/Linear/pre_reshape", "/layers.40/self_attn/q_proj/Linear/pre_convert", "/layers.40/self_attn/q_proj/Linear", "/layers.40/self_attn/q_proj/Linear/post_convert", "/layers.40/self_attn/k_proj/Linear/pre_reshape", "/layers.40/self_attn/k_proj/Linear/pre_convert", "/layers.40/self_attn/k_proj/Linear", "/layers.40/self_attn/k_proj/Linear/post_convert", "/layers.40/self_attn/v_proj/Linear/pre_reshape", "/layers.40/self_attn/v_proj/Linear/pre_convert", "/layers.40/self_attn/v_proj/Linear", "/layers.40/self_attn/v_proj/Linear/post_convert", "/layers.40/self_attn/o_proj/Linear/pre_reshape", "/layers.40/self_attn/o_proj/Linear/pre_convert", "/layers.40/self_attn/o_proj/Linear", "/layers.40/self_attn/o_proj/Linear/post_convert", "/layers.40/mlp/gate_proj/Linear/pre_reshape", "/layers.40/mlp/gate_proj/Linear/pre_convert", "/layers.40/mlp/gate_proj/Linear", "/layers.40/mlp/gate_proj/Linear/post_convert", "/layers.40/mlp/up_proj/Linear/pre_reshape", "/layers.40/mlp/up_proj/Linear/pre_convert", "/layers.40/mlp/up_proj/Linear", "/layers.40/mlp/up_proj/Linear/post_convert", "/layers.40/mlp/down_proj/Linear/pre_reshape", "/layers.40/mlp/down_proj/Linear/pre_convert", "/layers.40/mlp/down_proj/Linear", "/layers.40/mlp/down_proj/Linear/post_convert", "/layers.41/self_attn/q_proj/Linear/pre_reshape", "/layers.41/self_attn/q_proj/Linear/pre_convert", "/layers.41/self_attn/q_proj/Linear", "/layers.41/self_attn/q_proj/Linear/post_convert", "/layers.41/self_attn/k_proj/Linear/pre_reshape", "/layers.41/self_attn/k_proj/Linear/pre_convert", "/layers.41/self_attn/k_proj/Linear", "/layers.41/self_attn/k_proj/Linear/post_convert", "/layers.41/self_attn/v_proj/Linear/pre_reshape", "/layers.41/self_attn/v_proj/Linear/pre_convert", "/layers.41/self_attn/v_proj/Linear", "/layers.41/self_attn/v_proj/Linear/post_convert", "/layers.41/self_attn/o_proj/Linear/pre_reshape", "/layers.41/self_attn/o_proj/Linear/pre_convert", "/layers.41/self_attn/o_proj/Linear", "/layers.41/self_attn/o_proj/Linear/post_convert", "/layers.41/mlp/gate_proj/Linear/pre_reshape", "/layers.41/mlp/gate_proj/Linear/pre_convert", "/layers.41/mlp/gate_proj/Linear", "/layers.41/mlp/gate_proj/Linear/post_convert", "/layers.41/mlp/up_proj/Linear/pre_reshape", "/layers.41/mlp/up_proj/Linear/pre_convert", "/layers.41/mlp/up_proj/Linear", "/layers.41/mlp/up_proj/Linear/post_convert", "/layers.41/mlp/down_proj/Linear/pre_reshape", "/layers.41/mlp/down_proj/Linear/pre_convert", "/layers.41/mlp/down_proj/Linear", "/layers.41/mlp/down_proj/Linear/post_convert", "/layers.42/self_attn/q_proj/Linear/pre_reshape", "/layers.42/self_attn/q_proj/Linear/pre_convert", "/layers.42/self_attn/q_proj/Linear", "/layers.42/self_attn/q_proj/Linear/post_convert", "/layers.42/self_attn/k_proj/Linear/pre_reshape", "/layers.42/self_attn/k_proj/Linear/pre_convert", "/layers.42/self_attn/k_proj/Linear", "/layers.42/self_attn/k_proj/Linear/post_convert", "/layers.42/self_attn/v_proj/Linear/pre_reshape", "/layers.42/self_attn/v_proj/Linear/pre_convert", "/layers.42/self_attn/v_proj/Linear", "/layers.42/self_attn/v_proj/Linear/post_convert", "/layers.42/self_attn/o_proj/Linear/pre_reshape", "/layers.42/self_attn/o_proj/Linear/pre_convert", "/layers.42/self_attn/o_proj/Linear", "/layers.42/self_attn/o_proj/Linear/post_convert", "/layers.42/mlp/gate_proj/Linear/pre_reshape", "/layers.42/mlp/gate_proj/Linear/pre_convert", "/layers.42/mlp/gate_proj/Linear", "/layers.42/mlp/gate_proj/Linear/post_convert", "/layers.42/mlp/up_proj/Linear/pre_reshape", "/layers.42/mlp/up_proj/Linear/pre_convert", "/layers.42/mlp/up_proj/Linear", "/layers.42/mlp/up_proj/Linear/post_convert", "/layers.42/mlp/down_proj/Linear/pre_reshape", "/layers.42/mlp/down_proj/Linear/pre_convert", "/layers.42/mlp/down_proj/Linear", "/layers.42/mlp/down_proj/Linear/post_convert", "/layers.43/self_attn/q_proj/Linear/pre_reshape", "/layers.43/self_attn/q_proj/Linear/pre_convert", "/layers.43/self_attn/q_proj/Linear", "/layers.43/self_attn/q_proj/Linear/post_convert", "/layers.43/self_attn/k_proj/Linear/pre_reshape", "/layers.43/self_attn/k_proj/Linear/pre_convert", "/layers.43/self_attn/k_proj/Linear", "/layers.43/self_attn/k_proj/Linear/post_convert", "/layers.43/self_attn/v_proj/Linear/pre_reshape", "/layers.43/self_attn/v_proj/Linear/pre_convert", "/layers.43/self_attn/v_proj/Linear", "/layers.43/self_attn/v_proj/Linear/post_convert", "/layers.43/self_attn/o_proj/Linear/pre_reshape", "/layers.43/self_attn/o_proj/Linear/pre_convert", "/layers.43/self_attn/o_proj/Linear", "/layers.43/self_attn/o_proj/Linear/post_convert", "/layers.43/mlp/gate_proj/Linear/pre_reshape", "/layers.43/mlp/gate_proj/Linear/pre_convert", "/layers.43/mlp/gate_proj/Linear", "/layers.43/mlp/gate_proj/Linear/post_convert", "/layers.43/mlp/up_proj/Linear/pre_reshape", "/layers.43/mlp/up_proj/Linear/pre_convert", "/layers.43/mlp/up_proj/Linear", "/layers.43/mlp/up_proj/Linear/post_convert", "/layers.43/mlp/down_proj/Linear/pre_reshape", "/layers.43/mlp/down_proj/Linear/pre_convert", "/layers.43/mlp/down_proj/Linear", "/layers.43/mlp/down_proj/Linear/post_convert", "/layers.44/self_attn/q_proj/Linear/pre_reshape", "/layers.44/self_attn/q_proj/Linear/pre_convert", "/layers.44/self_attn/q_proj/Linear", "/layers.44/self_attn/q_proj/Linear/post_convert", "/layers.44/self_attn/k_proj/Linear/pre_reshape", "/layers.44/self_attn/k_proj/Linear/pre_convert", "/layers.44/self_attn/k_proj/Linear", "/layers.44/self_attn/k_proj/Linear/post_convert", "/layers.44/self_attn/v_proj/Linear/pre_reshape", "/layers.44/self_attn/v_proj/Linear/pre_convert", "/layers.44/self_attn/v_proj/Linear", "/layers.44/self_attn/v_proj/Linear/post_convert", "/layers.44/self_attn/o_proj/Linear/pre_reshape", "/layers.44/self_attn/o_proj/Linear/pre_convert", "/layers.44/self_attn/o_proj/Linear", "/layers.44/self_attn/o_proj/Linear/post_convert", "/layers.44/mlp/gate_proj/Linear/pre_reshape", "/layers.44/mlp/gate_proj/Linear/pre_convert", "/layers.44/mlp/gate_proj/Linear", "/layers.44/mlp/gate_proj/Linear/post_convert", "/layers.44/mlp/up_proj/Linear/pre_reshape", "/layers.44/mlp/up_proj/Linear/pre_convert", "/layers.44/mlp/up_proj/Linear", "/layers.44/mlp/up_proj/Linear/post_convert", "/layers.44/mlp/down_proj/Linear/pre_reshape", "/layers.44/mlp/down_proj/Linear/pre_convert", "/layers.44/mlp/down_proj/Linear", "/layers.44/mlp/down_proj/Linear/post_convert", "/layers.45/self_attn/q_proj/Linear/pre_reshape", "/layers.45/self_attn/q_proj/Linear/pre_convert", "/layers.45/self_attn/q_proj/Linear", "/layers.45/self_attn/q_proj/Linear/post_convert", "/layers.45/self_attn/k_proj/Linear/pre_reshape", "/layers.45/self_attn/k_proj/Linear/pre_convert", "/layers.45/self_attn/k_proj/Linear", "/layers.45/self_attn/k_proj/Linear/post_convert", "/layers.45/self_attn/v_proj/Linear/pre_reshape", "/layers.45/self_attn/v_proj/Linear/pre_convert", "/layers.45/self_attn/v_proj/Linear", "/layers.45/self_attn/v_proj/Linear/post_convert", "/layers.45/self_attn/o_proj/Linear/pre_reshape", "/layers.45/self_attn/o_proj/Linear/pre_convert", "/layers.45/self_attn/o_proj/Linear", "/layers.45/self_attn/o_proj/Linear/post_convert", "/layers.45/mlp/gate_proj/Linear/pre_reshape", "/layers.45/mlp/gate_proj/Linear/pre_convert", "/layers.45/mlp/gate_proj/Linear", "/layers.45/mlp/gate_proj/Linear/post_convert", "/layers.45/mlp/up_proj/Linear/pre_reshape", "/layers.45/mlp/up_proj/Linear/pre_convert", "/layers.45/mlp/up_proj/Linear", "/layers.45/mlp/up_proj/Linear/post_convert", "/layers.45/mlp/down_proj/Linear/pre_reshape", "/layers.45/mlp/down_proj/Linear/pre_convert", "/layers.45/mlp/down_proj/Linear", "/layers.45/mlp/down_proj/Linear/post_convert", "/layers.46/self_attn/q_proj/Linear/pre_reshape", "/layers.46/self_attn/q_proj/Linear/pre_convert", "/layers.46/self_attn/q_proj/Linear", "/layers.46/self_attn/q_proj/Linear/post_convert", "/layers.46/self_attn/k_proj/Linear/pre_reshape", "/layers.46/self_attn/k_proj/Linear/pre_convert", "/layers.46/self_attn/k_proj/Linear", "/layers.46/self_attn/k_proj/Linear/post_convert", "/layers.46/self_attn/v_proj/Linear/pre_reshape", "/layers.46/self_attn/v_proj/Linear/pre_convert", "/layers.46/self_attn/v_proj/Linear", "/layers.46/self_attn/v_proj/Linear/post_convert", "/layers.46/self_attn/o_proj/Linear/pre_reshape", "/layers.46/self_attn/o_proj/Linear/pre_convert", "/layers.46/self_attn/o_proj/Linear", "/layers.46/self_attn/o_proj/Linear/post_convert", "/layers.46/mlp/gate_proj/Linear/pre_reshape", "/layers.46/mlp/gate_proj/Linear/pre_convert", "/layers.46/mlp/gate_proj/Linear", "/layers.46/mlp/gate_proj/Linear/post_convert", "/layers.46/mlp/up_proj/Linear/pre_reshape", "/layers.46/mlp/up_proj/Linear/pre_convert", "/layers.46/mlp/up_proj/Linear", "/layers.46/mlp/up_proj/Linear/post_convert", "/layers.46/mlp/down_proj/Linear/pre_reshape", "/layers.46/mlp/down_proj/Linear/pre_convert", "/layers.46/mlp/down_proj/Linear", "/layers.46/mlp/down_proj/Linear/post_convert", "/layers.47/self_attn/q_proj/Linear/pre_reshape", "/layers.47/self_attn/q_proj/Linear/pre_convert", "/layers.47/self_attn/q_proj/Linear", "/layers.47/self_attn/q_proj/Linear/post_convert", "/layers.47/self_attn/k_proj/Linear/pre_reshape", "/layers.47/self_attn/k_proj/Linear/pre_convert", "/layers.47/self_attn/k_proj/Linear", "/layers.47/self_attn/k_proj/Linear/post_convert", "/layers.47/self_attn/v_proj/Linear/pre_reshape", "/layers.47/self_attn/v_proj/Linear/pre_convert", "/layers.47/self_attn/v_proj/Linear", "/layers.47/self_attn/v_proj/Linear/post_convert", "/layers.47/self_attn/o_proj/Linear/pre_reshape", "/layers.47/self_attn/o_proj/Linear/pre_convert", "/layers.47/self_attn/o_proj/Linear", "/layers.47/self_attn/o_proj/Linear/post_convert", "/layers.47/mlp/gate_proj/Linear/pre_reshape", "/layers.47/mlp/gate_proj/Linear/pre_convert", "/layers.47/mlp/gate_proj/Linear", "/layers.47/mlp/gate_proj/Linear/post_convert", "/layers.47/mlp/up_proj/Linear/pre_reshape", "/layers.47/mlp/up_proj/Linear/pre_convert", "/layers.47/mlp/up_proj/Linear", "/layers.47/mlp/up_proj/Linear/post_convert", "/layers.47/mlp/down_proj/Linear/pre_reshape", "/layers.47/mlp/down_proj/Linear/pre_convert", "/layers.47/mlp/down_proj/Linear", "/layers.47/mlp/down_proj/Linear/post_convert", "/layers.48/self_attn/q_proj/Linear/pre_reshape", "/layers.48/self_attn/q_proj/Linear/pre_convert", "/layers.48/self_attn/q_proj/Linear", "/layers.48/self_attn/q_proj/Linear/post_convert", "/layers.48/self_attn/k_proj/Linear/pre_reshape", "/layers.48/self_attn/k_proj/Linear/pre_convert", "/layers.48/self_attn/k_proj/Linear", "/layers.48/self_attn/k_proj/Linear/post_convert", "/layers.48/self_attn/v_proj/Linear/pre_reshape", "/layers.48/self_attn/v_proj/Linear/pre_convert", "/layers.48/self_attn/v_proj/Linear", "/layers.48/self_attn/v_proj/Linear/post_convert", "/layers.48/self_attn/o_proj/Linear/pre_reshape", "/layers.48/self_attn/o_proj/Linear/pre_convert", "/layers.48/self_attn/o_proj/Linear", "/layers.48/self_attn/o_proj/Linear/post_convert", "/layers.48/mlp/gate_proj/Linear/pre_reshape", "/layers.48/mlp/gate_proj/Linear/pre_convert", "/layers.48/mlp/gate_proj/Linear", "/layers.48/mlp/gate_proj/Linear/post_convert", "/layers.48/mlp/up_proj/Linear/pre_reshape", "/layers.48/mlp/up_proj/Linear/pre_convert", "/layers.48/mlp/up_proj/Linear", "/layers.48/mlp/up_proj/Linear/post_convert", "/layers.48/mlp/down_proj/Linear/pre_reshape", "/layers.48/mlp/down_proj/Linear/pre_convert", "/layers.48/mlp/down_proj/Linear", "/layers.48/mlp/down_proj/Linear/post_convert", "/layers.49/self_attn/q_proj/Linear/pre_reshape", "/layers.49/self_attn/q_proj/Linear/pre_convert", "/layers.49/self_attn/q_proj/Linear", "/layers.49/self_attn/q_proj/Linear/post_convert", "/layers.49/self_attn/k_proj/Linear/pre_reshape", "/layers.49/self_attn/k_proj/Linear/pre_convert", "/layers.49/self_attn/k_proj/Linear", "/layers.49/self_attn/k_proj/Linear/post_convert", "/layers.49/self_attn/v_proj/Linear/pre_reshape", "/layers.49/self_attn/v_proj/Linear/pre_convert", "/layers.49/self_attn/v_proj/Linear", "/layers.49/self_attn/v_proj/Linear/post_convert", "/layers.49/self_attn/o_proj/Linear/pre_reshape", "/layers.49/self_attn/o_proj/Linear/pre_convert", "/layers.49/self_attn/o_proj/Linear", "/layers.49/self_attn/o_proj/Linear/post_convert", "/layers.49/mlp/gate_proj/Linear/pre_reshape", "/layers.49/mlp/gate_proj/Linear/pre_convert", "/layers.49/mlp/gate_proj/Linear", "/layers.49/mlp/gate_proj/Linear/post_convert", "/layers.49/mlp/up_proj/Linear/pre_reshape", "/layers.49/mlp/up_proj/Linear/pre_convert", "/layers.49/mlp/up_proj/Linear", "/layers.49/mlp/up_proj/Linear/post_convert", "/layers.49/mlp/down_proj/Linear/pre_reshape", "/layers.49/mlp/down_proj/Linear/pre_convert", "/layers.49/mlp/down_proj/Linear", "/layers.49/mlp/down_proj/Linear/post_convert", "/layers.50/self_attn/q_proj/Linear/pre_reshape", "/layers.50/self_attn/q_proj/Linear/pre_convert", "/layers.50/self_attn/q_proj/Linear", "/layers.50/self_attn/q_proj/Linear/post_convert", "/layers.50/self_attn/k_proj/Linear/pre_reshape", "/layers.50/self_attn/k_proj/Linear/pre_convert", "/layers.50/self_attn/k_proj/Linear", "/layers.50/self_attn/k_proj/Linear/post_convert", "/layers.50/self_attn/v_proj/Linear/pre_reshape", "/layers.50/self_attn/v_proj/Linear/pre_convert", "/layers.50/self_attn/v_proj/Linear", "/layers.50/self_attn/v_proj/Linear/post_convert", "/layers.50/self_attn/o_proj/Linear/pre_reshape", "/layers.50/self_attn/o_proj/Linear/pre_convert", "/layers.50/self_attn/o_proj/Linear", "/layers.50/self_attn/o_proj/Linear/post_convert", "/layers.50/mlp/gate_proj/Linear/pre_reshape", "/layers.50/mlp/gate_proj/Linear/pre_convert", "/layers.50/mlp/gate_proj/Linear", "/layers.50/mlp/gate_proj/Linear/post_convert", "/layers.50/mlp/up_proj/Linear/pre_reshape", "/layers.50/mlp/up_proj/Linear/pre_convert", "/layers.50/mlp/up_proj/Linear", "/layers.50/mlp/up_proj/Linear/post_convert", "/layers.50/mlp/down_proj/Linear/pre_reshape", "/layers.50/mlp/down_proj/Linear/pre_convert", "/layers.50/mlp/down_proj/Linear", "/layers.50/mlp/down_proj/Linear/post_convert", "/layers.51/self_attn/q_proj/Linear/pre_reshape", "/layers.51/self_attn/q_proj/Linear/pre_convert", "/layers.51/self_attn/q_proj/Linear", "/layers.51/self_attn/q_proj/Linear/post_convert", "/layers.51/self_attn/k_proj/Linear/pre_reshape", "/layers.51/self_attn/k_proj/Linear/pre_convert", "/layers.51/self_attn/k_proj/Linear", "/layers.51/self_attn/k_proj/Linear/post_convert", "/layers.51/self_attn/v_proj/Linear/pre_reshape", "/layers.51/self_attn/v_proj/Linear/pre_convert", "/layers.51/self_attn/v_proj/Linear", "/layers.51/self_attn/v_proj/Linear/post_convert", "/layers.51/self_attn/o_proj/Linear/pre_reshape", "/layers.51/self_attn/o_proj/Linear/pre_convert", "/layers.51/self_attn/o_proj/Linear", "/layers.51/self_attn/o_proj/Linear/post_convert", "/layers.51/mlp/gate_proj/Linear/pre_reshape", "/layers.51/mlp/gate_proj/Linear/pre_convert", "/layers.51/mlp/gate_proj/Linear", "/layers.51/mlp/gate_proj/Linear/post_convert", "/layers.51/mlp/up_proj/Linear/pre_reshape", "/layers.51/mlp/up_proj/Linear/pre_convert", "/layers.51/mlp/up_proj/Linear", "/layers.51/mlp/up_proj/Linear/post_convert", "/layers.51/mlp/down_proj/Linear/pre_reshape", "/layers.51/mlp/down_proj/Linear/pre_convert", "/layers.51/mlp/down_proj/Linear", "/layers.51/mlp/down_proj/Linear/post_convert", "/layers.52/self_attn/q_proj/Linear/pre_reshape", "/layers.52/self_attn/q_proj/Linear/pre_convert", "/layers.52/self_attn/q_proj/Linear", "/layers.52/self_attn/q_proj/Linear/post_convert", "/layers.52/self_attn/k_proj/Linear/pre_reshape", "/layers.52/self_attn/k_proj/Linear/pre_convert", "/layers.52/self_attn/k_proj/Linear", "/layers.52/self_attn/k_proj/Linear/post_convert", "/layers.52/self_attn/v_proj/Linear/pre_reshape", "/layers.52/self_attn/v_proj/Linear/pre_convert", "/layers.52/self_attn/v_proj/Linear", "/layers.52/self_attn/v_proj/Linear/post_convert", "/layers.52/self_attn/o_proj/Linear/pre_reshape", "/layers.52/self_attn/o_proj/Linear/pre_convert", "/layers.52/self_attn/o_proj/Linear", "/layers.52/self_attn/o_proj/Linear/post_convert", "/layers.52/mlp/gate_proj/Linear/pre_reshape", "/layers.52/mlp/gate_proj/Linear/pre_convert", "/layers.52/mlp/gate_proj/Linear", "/layers.52/mlp/gate_proj/Linear/post_convert", "/layers.52/mlp/up_proj/Linear/pre_reshape", "/layers.52/mlp/up_proj/Linear/pre_convert", "/layers.52/mlp/up_proj/Linear", "/layers.52/mlp/up_proj/Linear/post_convert", "/layers.52/mlp/down_proj/Linear/pre_reshape", "/layers.52/mlp/down_proj/Linear/pre_convert", "/layers.52/mlp/down_proj/Linear", "/layers.52/mlp/down_proj/Linear/post_convert", "/layers.53/self_attn/q_proj/Linear/pre_reshape", "/layers.53/self_attn/q_proj/Linear/pre_convert", "/layers.53/self_attn/q_proj/Linear", "/layers.53/self_attn/q_proj/Linear/post_convert", "/layers.53/self_attn/k_proj/Linear/pre_reshape", "/layers.53/self_attn/k_proj/Linear/pre_convert", "/layers.53/self_attn/k_proj/Linear", "/layers.53/self_attn/k_proj/Linear/post_convert", "/layers.53/self_attn/v_proj/Linear/pre_reshape", "/layers.53/self_attn/v_proj/Linear/pre_convert", "/layers.53/self_attn/v_proj/Linear", "/layers.53/self_attn/v_proj/Linear/post_convert", "/layers.53/self_attn/o_proj/Linear/pre_reshape", "/layers.53/self_attn/o_proj/Linear/pre_convert", "/layers.53/self_attn/o_proj/Linear", "/layers.53/self_attn/o_proj/Linear/post_convert", "/layers.53/mlp/gate_proj/Linear/pre_reshape", "/layers.53/mlp/gate_proj/Linear/pre_convert", "/layers.53/mlp/gate_proj/Linear", "/layers.53/mlp/gate_proj/Linear/post_convert", "/layers.53/mlp/up_proj/Linear/pre_reshape", "/layers.53/mlp/up_proj/Linear/pre_convert", "/layers.53/mlp/up_proj/Linear", "/layers.53/mlp/up_proj/Linear/post_convert", "/layers.53/mlp/down_proj/Linear/pre_reshape", "/layers.53/mlp/down_proj/Linear/pre_convert", "/layers.53/mlp/down_proj/Linear", "/layers.53/mlp/down_proj/Linear/post_convert", "/layers.54/self_attn/q_proj/Linear/pre_reshape", "/layers.54/self_attn/q_proj/Linear/pre_convert", "/layers.54/self_attn/q_proj/Linear", "/layers.54/self_attn/q_proj/Linear/post_convert", "/layers.54/self_attn/k_proj/Linear/pre_reshape", "/layers.54/self_attn/k_proj/Linear/pre_convert", "/layers.54/self_attn/k_proj/Linear", "/layers.54/self_attn/k_proj/Linear/post_convert", "/layers.54/self_attn/v_proj/Linear/pre_reshape", "/layers.54/self_attn/v_proj/Linear/pre_convert", "/layers.54/self_attn/v_proj/Linear", "/layers.54/self_attn/v_proj/Linear/post_convert", "/layers.54/self_attn/o_proj/Linear/pre_reshape", "/layers.54/self_attn/o_proj/Linear/pre_convert", "/layers.54/self_attn/o_proj/Linear", "/layers.54/self_attn/o_proj/Linear/post_convert", "/layers.54/mlp/gate_proj/Linear/pre_reshape", "/layers.54/mlp/gate_proj/Linear/pre_convert", "/layers.54/mlp/gate_proj/Linear", "/layers.54/mlp/gate_proj/Linear/post_convert", "/layers.54/mlp/up_proj/Linear/pre_reshape", "/layers.54/mlp/up_proj/Linear/pre_convert", "/layers.54/mlp/up_proj/Linear", "/layers.54/mlp/up_proj/Linear/post_convert", "/layers.54/mlp/down_proj/Linear/pre_reshape", "/layers.54/mlp/down_proj/Linear/pre_convert", "/layers.54/mlp/down_proj/Linear", "/layers.54/mlp/down_proj/Linear/post_convert", "/layers.55/self_attn/q_proj/Linear/pre_reshape", "/layers.55/self_attn/q_proj/Linear/pre_convert", "/layers.55/self_attn/q_proj/Linear", "/layers.55/self_attn/q_proj/Linear/post_convert", "/layers.55/self_attn/k_proj/Linear/pre_reshape", "/layers.55/self_attn/k_proj/Linear/pre_convert", "/layers.55/self_attn/k_proj/Linear", "/layers.55/self_attn/k_proj/Linear/post_convert", "/layers.55/self_attn/v_proj/Linear/pre_reshape", "/layers.55/self_attn/v_proj/Linear/pre_convert", "/layers.55/self_attn/v_proj/Linear", "/layers.55/self_attn/v_proj/Linear/post_convert", "/layers.55/self_attn/o_proj/Linear/pre_reshape", "/layers.55/self_attn/o_proj/Linear/pre_convert", "/layers.55/self_attn/o_proj/Linear", "/layers.55/self_attn/o_proj/Linear/post_convert", "/layers.55/mlp/gate_proj/Linear/pre_reshape", "/layers.55/mlp/gate_proj/Linear/pre_convert", "/layers.55/mlp/gate_proj/Linear", "/layers.55/mlp/gate_proj/Linear/post_convert", "/layers.55/mlp/up_proj/Linear/pre_reshape", "/layers.55/mlp/up_proj/Linear/pre_convert", "/layers.55/mlp/up_proj/Linear", "/layers.55/mlp/up_proj/Linear/post_convert", "/layers.55/mlp/down_proj/Linear/pre_reshape", "/layers.55/mlp/down_proj/Linear/pre_convert", "/layers.55/mlp/down_proj/Linear", "/layers.55/mlp/down_proj/Linear/post_convert", "/layers.56/self_attn/q_proj/Linear/pre_reshape", "/layers.56/self_attn/q_proj/Linear/pre_convert", "/layers.56/self_attn/q_proj/Linear", "/layers.56/self_attn/q_proj/Linear/post_convert", "/layers.56/self_attn/k_proj/Linear/pre_reshape", "/layers.56/self_attn/k_proj/Linear/pre_convert", "/layers.56/self_attn/k_proj/Linear", "/layers.56/self_attn/k_proj/Linear/post_convert", "/layers.56/self_attn/v_proj/Linear/pre_reshape", "/layers.56/self_attn/v_proj/Linear/pre_convert", "/layers.56/self_attn/v_proj/Linear", "/layers.56/self_attn/v_proj/Linear/post_convert", "/layers.56/self_attn/o_proj/Linear/pre_reshape", "/layers.56/self_attn/o_proj/Linear/pre_convert", "/layers.56/self_attn/o_proj/Linear", "/layers.56/self_attn/o_proj/Linear/post_convert", "/layers.56/mlp/gate_proj/Linear/pre_reshape", "/layers.56/mlp/gate_proj/Linear/pre_convert", "/layers.56/mlp/gate_proj/Linear", "/layers.56/mlp/gate_proj/Linear/post_convert", "/layers.56/mlp/up_proj/Linear/pre_reshape", "/layers.56/mlp/up_proj/Linear/pre_convert", "/layers.56/mlp/up_proj/Linear", "/layers.56/mlp/up_proj/Linear/post_convert", "/layers.56/mlp/down_proj/Linear/pre_reshape", "/layers.56/mlp/down_proj/Linear/pre_convert", "/layers.56/mlp/down_proj/Linear", "/layers.56/mlp/down_proj/Linear/post_convert", "/layers.57/self_attn/q_proj/Linear/pre_reshape", "/layers.57/self_attn/q_proj/Linear/pre_convert", "/layers.57/self_attn/q_proj/Linear", "/layers.57/self_attn/q_proj/Linear/post_convert", "/layers.57/self_attn/k_proj/Linear/pre_reshape", "/layers.57/self_attn/k_proj/Linear/pre_convert", "/layers.57/self_attn/k_proj/Linear", "/layers.57/self_attn/k_proj/Linear/post_convert", "/layers.57/self_attn/v_proj/Linear/pre_reshape", "/layers.57/self_attn/v_proj/Linear/pre_convert", "/layers.57/self_attn/v_proj/Linear", "/layers.57/self_attn/v_proj/Linear/post_convert", "/layers.57/self_attn/o_proj/Linear/pre_reshape", "/layers.57/self_attn/o_proj/Linear/pre_convert", "/layers.57/self_attn/o_proj/Linear", "/layers.57/self_attn/o_proj/Linear/post_convert", "/layers.57/mlp/gate_proj/Linear/pre_reshape", "/layers.57/mlp/gate_proj/Linear/pre_convert", "/layers.57/mlp/gate_proj/Linear", "/layers.57/mlp/gate_proj/Linear/post_convert", "/layers.57/mlp/up_proj/Linear/pre_reshape", "/layers.57/mlp/up_proj/Linear/pre_convert", "/layers.57/mlp/up_proj/Linear", "/layers.57/mlp/up_proj/Linear/post_convert", "/layers.57/mlp/down_proj/Linear/pre_reshape", "/layers.57/mlp/down_proj/Linear/pre_convert", "/layers.57/mlp/down_proj/Linear", "/layers.57/mlp/down_proj/Linear/post_convert", "/layers.58/self_attn/q_proj/Linear/pre_reshape", "/layers.58/self_attn/q_proj/Linear/pre_convert", "/layers.58/self_attn/q_proj/Linear", "/layers.58/self_attn/q_proj/Linear/post_convert", "/layers.58/self_attn/k_proj/Linear/pre_reshape", "/layers.58/self_attn/k_proj/Linear/pre_convert", "/layers.58/self_attn/k_proj/Linear", "/layers.58/self_attn/k_proj/Linear/post_convert", "/layers.58/self_attn/v_proj/Linear/pre_reshape", "/layers.58/self_attn/v_proj/Linear/pre_convert", "/layers.58/self_attn/v_proj/Linear", "/layers.58/self_attn/v_proj/Linear/post_convert", "/layers.58/self_attn/o_proj/Linear/pre_reshape", "/layers.58/self_attn/o_proj/Linear/pre_convert", "/layers.58/self_attn/o_proj/Linear", "/layers.58/self_attn/o_proj/Linear/post_convert", "/layers.58/mlp/gate_proj/Linear/pre_reshape", "/layers.58/mlp/gate_proj/Linear/pre_convert", "/layers.58/mlp/gate_proj/Linear", "/layers.58/mlp/gate_proj/Linear/post_convert", "/layers.58/mlp/up_proj/Linear/pre_reshape", "/layers.58/mlp/up_proj/Linear/pre_convert", "/layers.58/mlp/up_proj/Linear", "/layers.58/mlp/up_proj/Linear/post_convert", "/layers.58/mlp/down_proj/Linear/pre_reshape", "/layers.58/mlp/down_proj/Linear/pre_convert", "/layers.58/mlp/down_proj/Linear", "/layers.58/mlp/down_proj/Linear/post_convert", "/layers.59/self_attn/q_proj/Linear/pre_reshape", "/layers.59/self_attn/q_proj/Linear/pre_convert", "/layers.59/self_attn/q_proj/Linear", "/layers.59/self_attn/q_proj/Linear/post_convert", "/layers.59/self_attn/k_proj/Linear/pre_reshape", "/layers.59/self_attn/k_proj/Linear/pre_convert", "/layers.59/self_attn/k_proj/Linear", "/layers.59/self_attn/k_proj/Linear/post_convert", "/layers.59/self_attn/v_proj/Linear/pre_reshape", "/layers.59/self_attn/v_proj/Linear/pre_convert", "/layers.59/self_attn/v_proj/Linear", "/layers.59/self_attn/v_proj/Linear/post_convert", "/layers.59/self_attn/o_proj/Linear/pre_reshape", "/layers.59/self_attn/o_proj/Linear/pre_convert", "/layers.59/self_attn/o_proj/Linear", "/layers.59/self_attn/o_proj/Linear/post_convert", "/layers.59/mlp/gate_proj/Linear/pre_reshape", "/layers.59/mlp/gate_proj/Linear/pre_convert", "/layers.59/mlp/gate_proj/Linear", "/layers.59/mlp/gate_proj/Linear/post_convert", "/layers.59/mlp/up_proj/Linear/pre_reshape", "/layers.59/mlp/up_proj/Linear/pre_convert", "/layers.59/mlp/up_proj/Linear", "/layers.59/mlp/up_proj/Linear/post_convert", "/layers.59/mlp/down_proj/Linear/pre_reshape", "/layers.59/mlp/down_proj/Linear/pre_convert", "/layers.59/mlp/down_proj/Linear", "/layers.59/mlp/down_proj/Linear/post_convert", "/layers.60/self_attn/q_proj/Linear/pre_reshape", "/layers.60/self_attn/q_proj/Linear/pre_convert", "/layers.60/self_attn/q_proj/Linear", "/layers.60/self_attn/q_proj/Linear/post_convert", "/layers.60/self_attn/k_proj/Linear/pre_reshape", "/layers.60/self_attn/k_proj/Linear/pre_convert", "/layers.60/self_attn/k_proj/Linear", "/layers.60/self_attn/k_proj/Linear/post_convert", "/layers.60/self_attn/v_proj/Linear/pre_reshape", "/layers.60/self_attn/v_proj/Linear/pre_convert", "/layers.60/self_attn/v_proj/Linear", "/layers.60/self_attn/v_proj/Linear/post_convert", "/layers.60/self_attn/o_proj/Linear/pre_reshape", "/layers.60/self_attn/o_proj/Linear/pre_convert", "/layers.60/self_attn/o_proj/Linear", "/layers.60/self_attn/o_proj/Linear/post_convert", "/layers.60/mlp/gate_proj/Linear/pre_reshape", "/layers.60/mlp/gate_proj/Linear/pre_convert", "/layers.60/mlp/gate_proj/Linear", "/layers.60/mlp/gate_proj/Linear/post_convert", "/layers.60/mlp/up_proj/Linear/pre_reshape", "/layers.60/mlp/up_proj/Linear/pre_convert", "/layers.60/mlp/up_proj/Linear", "/layers.60/mlp/up_proj/Linear/post_convert", "/layers.60/mlp/down_proj/Linear/pre_reshape", "/layers.60/mlp/down_proj/Linear/pre_convert", "/layers.60/mlp/down_proj/Linear", "/layers.60/mlp/down_proj/Linear/post_convert", "/layers.61/self_attn/q_proj/Linear/pre_reshape", "/layers.61/self_attn/q_proj/Linear/pre_convert", "/layers.61/self_attn/q_proj/Linear", "/layers.61/self_attn/q_proj/Linear/post_convert", "/layers.61/self_attn/k_proj/Linear/pre_reshape", "/layers.61/self_attn/k_proj/Linear/pre_convert", "/layers.61/self_attn/k_proj/Linear", "/layers.61/self_attn/k_proj/Linear/post_convert", "/layers.61/self_attn/v_proj/Linear/pre_reshape", "/layers.61/self_attn/v_proj/Linear/pre_convert", "/layers.61/self_attn/v_proj/Linear", "/layers.61/self_attn/v_proj/Linear/post_convert", "/layers.61/self_attn/o_proj/Linear/pre_reshape", "/layers.61/self_attn/o_proj/Linear/pre_convert", "/layers.61/self_attn/o_proj/Linear", "/layers.61/self_attn/o_proj/Linear/post_convert", "/layers.61/mlp/gate_proj/Linear/pre_reshape", "/layers.61/mlp/gate_proj/Linear/pre_convert", "/layers.61/mlp/gate_proj/Linear", "/layers.61/mlp/gate_proj/Linear/post_convert", "/layers.61/mlp/up_proj/Linear/pre_reshape", "/layers.61/mlp/up_proj/Linear/pre_convert", "/layers.61/mlp/up_proj/Linear", "/layers.61/mlp/up_proj/Linear/post_convert", "/layers.61/mlp/down_proj/Linear/pre_reshape", "/layers.61/mlp/down_proj/Linear/pre_convert", "/layers.61/mlp/down_proj/Linear", "/layers.61/mlp/down_proj/Linear/post_convert", "/layers.62/self_attn/q_proj/Linear/pre_reshape", "/layers.62/self_attn/q_proj/Linear/pre_convert", "/layers.62/self_attn/q_proj/Linear", "/layers.62/self_attn/q_proj/Linear/post_convert", "/layers.62/self_attn/k_proj/Linear/pre_reshape", "/layers.62/self_attn/k_proj/Linear/pre_convert", "/layers.62/self_attn/k_proj/Linear", "/layers.62/self_attn/k_proj/Linear/post_convert", "/layers.62/self_attn/v_proj/Linear/pre_reshape", "/layers.62/self_attn/v_proj/Linear/pre_convert", "/layers.62/self_attn/v_proj/Linear", "/layers.62/self_attn/v_proj/Linear/post_convert", "/layers.62/self_attn/o_proj/Linear/pre_reshape", "/layers.62/self_attn/o_proj/Linear/pre_convert", "/layers.62/self_attn/o_proj/Linear", "/layers.62/self_attn/o_proj/Linear/post_convert", "/layers.62/mlp/gate_proj/Linear/pre_reshape", "/layers.62/mlp/gate_proj/Linear/pre_convert", "/layers.62/mlp/gate_proj/Linear", "/layers.62/mlp/gate_proj/Linear/post_convert", "/layers.62/mlp/up_proj/Linear/pre_reshape", "/layers.62/mlp/up_proj/Linear/pre_convert", "/layers.62/mlp/up_proj/Linear", "/layers.62/mlp/up_proj/Linear/post_convert", "/layers.62/mlp/down_proj/Linear/pre_reshape", "/layers.62/mlp/down_proj/Linear/pre_convert", "/layers.62/mlp/down_proj/Linear", "/layers.62/mlp/down_proj/Linear/post_convert", "/layers.63/self_attn/q_proj/Linear/pre_reshape", "/layers.63/self_attn/q_proj/Linear/pre_convert", "/layers.63/self_attn/q_proj/Linear", "/layers.63/self_attn/q_proj/Linear/post_convert", "/layers.63/self_attn/k_proj/Linear/pre_reshape", "/layers.63/self_attn/k_proj/Linear/pre_convert", "/layers.63/self_attn/k_proj/Linear", "/layers.63/self_attn/k_proj/Linear/post_convert", "/layers.63/self_attn/v_proj/Linear/pre_reshape", "/layers.63/self_attn/v_proj/Linear/pre_convert", "/layers.63/self_attn/v_proj/Linear", "/layers.63/self_attn/v_proj/Linear/post_convert", "/layers.63/self_attn/o_proj/Linear/pre_reshape", "/layers.63/self_attn/o_proj/Linear/pre_convert", "/layers.63/self_attn/o_proj/Linear", "/layers.63/self_attn/o_proj/Linear/post_convert", "/layers.63/mlp/gate_proj/Linear/pre_reshape", "/layers.63/mlp/gate_proj/Linear/pre_convert", "/layers.63/mlp/gate_proj/Linear", "/layers.63/mlp/gate_proj/Linear/post_convert", "/layers.63/mlp/up_proj/Linear/pre_reshape", "/layers.63/mlp/up_proj/Linear/pre_convert", "/layers.63/mlp/up_proj/Linear", "/layers.63/mlp/up_proj/Linear/post_convert", "/layers.63/mlp/down_proj/Linear/pre_reshape", "/layers.63/mlp/down_proj/Linear/pre_convert", "/layers.63/mlp/down_proj/Linear", "/layers.63/mlp/down_proj/Linear/post_convert", "/lm/lm_head/Linear/pre_reshape", "/lm/lm_head/Linear/pre_convert", "/lm/lm_head/Linear", "/lm/lm_head/Linear/post_convert" ], "tensorNumber": 0, "usage": "INFERENCE", "mnn_uuid": "0b39ed4a-f466-4850-9a5b-971b628c229f" }