diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4984 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9987389659520807,
+  "eval_steps": 500,
+  "global_step": 495,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.00201765447667087,
+      "grad_norm": 42.0,
+      "learning_rate": 0.0,
+      "loss": 3.6982,
+      "memory/device_mem_reserved(gib)": 49.98,
+      "memory/max_mem_active(gib)": 45.11,
+      "memory/max_mem_allocated(gib)": 45.11,
+      "step": 1
+    },
+    {
+      "epoch": 0.00403530895334174,
+      "grad_norm": 43.0,
+      "learning_rate": 2.0000000000000002e-07,
+      "loss": 3.8197,
+      "memory/device_mem_reserved(gib)": 50.78,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 2
+    },
+    {
+      "epoch": 0.00605296343001261,
+      "grad_norm": 43.5,
+      "learning_rate": 4.0000000000000003e-07,
+      "loss": 3.8927,
+      "memory/device_mem_reserved(gib)": 50.78,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 3
+    },
+    {
+      "epoch": 0.00807061790668348,
+      "grad_norm": 44.5,
+      "learning_rate": 6.000000000000001e-07,
+      "loss": 3.7968,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 4
+    },
+    {
+      "epoch": 0.01008827238335435,
+      "grad_norm": 42.25,
+      "learning_rate": 8.000000000000001e-07,
+      "loss": 3.8159,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 5
+    },
+    {
+      "epoch": 0.01210592686002522,
+      "grad_norm": 39.5,
+      "learning_rate": 1.0000000000000002e-06,
+      "loss": 3.8933,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 6
+    },
+    {
+      "epoch": 0.01412358133669609,
+      "grad_norm": 42.0,
+      "learning_rate": 1.2000000000000002e-06,
+      "loss": 3.8945,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 7
+    },
+    {
+      "epoch": 0.01614123581336696,
+      "grad_norm": 44.5,
+      "learning_rate": 1.4000000000000001e-06,
+      "loss": 3.9202,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 8
+    },
+    {
+      "epoch": 0.01815889029003783,
+      "grad_norm": 44.5,
+      "learning_rate": 1.6000000000000001e-06,
+      "loss": 3.8472,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 9
+    },
+    {
+      "epoch": 0.0201765447667087,
+      "grad_norm": 44.25,
+      "learning_rate": 1.8000000000000001e-06,
+      "loss": 3.9382,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 10
+    },
+    {
+      "epoch": 0.02219419924337957,
+      "grad_norm": 46.75,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 3.8074,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 11
+    },
+    {
+      "epoch": 0.02421185372005044,
+      "grad_norm": 43.0,
+      "learning_rate": 2.2e-06,
+      "loss": 3.9049,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 12
+    },
+    {
+      "epoch": 0.02622950819672131,
+      "grad_norm": 46.25,
+      "learning_rate": 2.4000000000000003e-06,
+      "loss": 3.8222,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 13
+    },
+    {
+      "epoch": 0.02824716267339218,
+      "grad_norm": 39.75,
+      "learning_rate": 2.6e-06,
+      "loss": 3.7079,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 14
+    },
+    {
+      "epoch": 0.03026481715006305,
+      "grad_norm": 39.25,
+      "learning_rate": 2.8000000000000003e-06,
+      "loss": 3.6677,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 15
+    },
+    {
+      "epoch": 0.03228247162673392,
+      "grad_norm": 46.75,
+      "learning_rate": 3e-06,
+      "loss": 3.6726,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 16
+    },
+    {
+      "epoch": 0.03430012610340479,
+      "grad_norm": 43.0,
+      "learning_rate": 3.2000000000000003e-06,
+      "loss": 3.7075,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 17
+    },
+    {
+      "epoch": 0.03631778058007566,
+      "grad_norm": 47.25,
+      "learning_rate": 3.4000000000000005e-06,
+      "loss": 3.7656,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 18
+    },
+    {
+      "epoch": 0.03833543505674653,
+      "grad_norm": 42.25,
+      "learning_rate": 3.6000000000000003e-06,
+      "loss": 3.6015,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 19
+    },
+    {
+      "epoch": 0.0403530895334174,
+      "grad_norm": 45.25,
+      "learning_rate": 3.8000000000000005e-06,
+      "loss": 3.5949,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 20
+    },
+    {
+      "epoch": 0.04237074401008827,
+      "grad_norm": 43.5,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 3.514,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 21
+    },
+    {
+      "epoch": 0.04438839848675914,
+      "grad_norm": 44.75,
+      "learning_rate": 4.2000000000000004e-06,
+      "loss": 3.3632,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 22
+    },
+    {
+      "epoch": 0.04640605296343001,
+      "grad_norm": 40.75,
+      "learning_rate": 4.4e-06,
+      "loss": 3.4502,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 23
+    },
+    {
+      "epoch": 0.04842370744010088,
+      "grad_norm": 33.25,
+      "learning_rate": 4.600000000000001e-06,
+      "loss": 3.3191,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 24
+    },
+    {
+      "epoch": 0.05044136191677175,
+      "grad_norm": 39.0,
+      "learning_rate": 4.800000000000001e-06,
+      "loss": 3.3536,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 25
+    },
+    {
+      "epoch": 0.05245901639344262,
+      "grad_norm": 27.625,
+      "learning_rate": 5e-06,
+      "loss": 3.3645,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 26
+    },
+    {
+      "epoch": 0.05447667087011349,
+      "grad_norm": 26.375,
+      "learning_rate": 4.999944151382673e-06,
+      "loss": 3.2603,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 27
+    },
+    {
+      "epoch": 0.05649432534678436,
+      "grad_norm": 20.375,
+      "learning_rate": 4.999776608025946e-06,
+      "loss": 3.2394,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 28
+    },
+    {
+      "epoch": 0.05851197982345523,
+      "grad_norm": 20.375,
+      "learning_rate": 4.99949737741547e-06,
+      "loss": 3.237,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 29
+    },
+    {
+      "epoch": 0.0605296343001261,
+      "grad_norm": 14.25,
+      "learning_rate": 4.99910647202696e-06,
+      "loss": 3.0697,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 30
+    },
+    {
+      "epoch": 0.06254728877679698,
+      "grad_norm": 12.4375,
+      "learning_rate": 4.998603909325636e-06,
+      "loss": 3.2899,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 31
+    },
+    {
+      "epoch": 0.06456494325346784,
+      "grad_norm": 11.25,
+      "learning_rate": 4.997989711765447e-06,
+      "loss": 3.1902,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 32
+    },
+    {
+      "epoch": 0.06658259773013872,
+      "grad_norm": 7.5625,
+      "learning_rate": 4.9972639067880555e-06,
+      "loss": 3.0802,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 33
+    },
+    {
+      "epoch": 0.06860025220680958,
+      "grad_norm": 6.5,
+      "learning_rate": 4.996426526821629e-06,
+      "loss": 3.1234,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 34
+    },
+    {
+      "epoch": 0.07061790668348046,
+      "grad_norm": 6.25,
+      "learning_rate": 4.9954776092793755e-06,
+      "loss": 3.1298,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 35
+    },
+    {
+      "epoch": 0.07263556116015132,
+      "grad_norm": 5.9375,
+      "learning_rate": 4.994417196557884e-06,
+      "loss": 2.9955,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 36
+    },
+    {
+      "epoch": 0.0746532156368222,
+      "grad_norm": 5.46875,
+      "learning_rate": 4.993245336035219e-06,
+      "loss": 3.1321,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 37
+    },
+    {
+      "epoch": 0.07667087011349306,
+      "grad_norm": 3.875,
+      "learning_rate": 4.991962080068813e-06,
+      "loss": 3.0705,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 38
+    },
+    {
+      "epoch": 0.07868852459016394,
+      "grad_norm": 4.40625,
+      "learning_rate": 4.990567485993125e-06,
+      "loss": 3.072,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 39
+    },
+    {
+      "epoch": 0.0807061790668348,
+      "grad_norm": 3.859375,
+      "learning_rate": 4.989061616117073e-06,
+      "loss": 3.2298,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 40
+    },
+    {
+      "epoch": 0.08272383354350568,
+      "grad_norm": 3.46875,
+      "learning_rate": 4.98744453772126e-06,
+      "loss": 3.014,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 41
+    },
+    {
+      "epoch": 0.08474148802017654,
+      "grad_norm": 3.25,
+      "learning_rate": 4.985716323054959e-06,
+      "loss": 3.0271,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 42
+    },
+    {
+      "epoch": 0.08675914249684742,
+      "grad_norm": 3.40625,
+      "learning_rate": 4.983877049332889e-06,
+      "loss": 3.0976,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 43
+    },
+    {
+      "epoch": 0.08877679697351828,
+      "grad_norm": 2.671875,
+      "learning_rate": 4.981926798731767e-06,
+      "loss": 2.959,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 44
+    },
+    {
+      "epoch": 0.09079445145018916,
+      "grad_norm": 3.0,
+      "learning_rate": 4.97986565838663e-06,
+      "loss": 2.9811,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 45
+    },
+    {
+      "epoch": 0.09281210592686003,
+      "grad_norm": 2.6875,
+      "learning_rate": 4.977693720386951e-06,
+      "loss": 2.9736,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 46
+    },
+    {
+      "epoch": 0.0948297604035309,
+      "grad_norm": 2.203125,
+      "learning_rate": 4.975411081772516e-06,
+      "loss": 3.0368,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 47
+    },
+    {
+      "epoch": 0.09684741488020177,
+      "grad_norm": 1.96875,
+      "learning_rate": 4.9730178445290945e-06,
+      "loss": 3.0042,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 48
+    },
+    {
+      "epoch": 0.09886506935687264,
+      "grad_norm": 1.84375,
+      "learning_rate": 4.970514115583878e-06,
+      "loss": 3.0294,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 49
+    },
+    {
+      "epoch": 0.1008827238335435,
+      "grad_norm": 1.8359375,
+      "learning_rate": 4.967900006800708e-06,
+      "loss": 3.0115,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 50
+    },
+    {
+      "epoch": 0.10290037831021438,
+      "grad_norm": 1.78125,
+      "learning_rate": 4.965175634975072e-06,
+      "loss": 3.0661,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 51
+    },
+    {
+      "epoch": 0.10491803278688525,
+      "grad_norm": 1.6015625,
+      "learning_rate": 4.96234112182889e-06,
+      "loss": 3.0006,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 52
+    },
+    {
+      "epoch": 0.10693568726355612,
+      "grad_norm": 1.5859375,
+      "learning_rate": 4.959396594005073e-06,
+      "loss": 3.0469,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 53
+    },
+    {
+      "epoch": 0.10895334174022699,
+      "grad_norm": 1.453125,
+      "learning_rate": 4.95634218306187e-06,
+      "loss": 3.0325,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 54
+    },
+    {
+      "epoch": 0.11097099621689786,
+      "grad_norm": 1.296875,
+      "learning_rate": 4.953178025466981e-06,
+      "loss": 2.9788,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 55
+    },
+    {
+      "epoch": 0.11298865069356873,
+      "grad_norm": 1.3984375,
+      "learning_rate": 4.949904262591467e-06,
+      "loss": 3.0148,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 56
+    },
+    {
+      "epoch": 0.1150063051702396,
+      "grad_norm": 1.3828125,
+      "learning_rate": 4.946521040703434e-06,
+      "loss": 2.9016,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 57
+    },
+    {
+      "epoch": 0.11702395964691047,
+      "grad_norm": 1.21875,
+      "learning_rate": 4.943028510961492e-06,
+      "loss": 3.1604,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 58
+    },
+    {
+      "epoch": 0.11904161412358134,
+      "grad_norm": 1.2109375,
+      "learning_rate": 4.939426829408008e-06,
+      "loss": 2.9688,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 59
+    },
+    {
+      "epoch": 0.1210592686002522,
+      "grad_norm": 1.203125,
+      "learning_rate": 4.9357161569621275e-06,
+      "loss": 2.8605,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 60
+    },
+    {
+      "epoch": 0.12307692307692308,
+      "grad_norm": 0.93359375,
+      "learning_rate": 4.931896659412593e-06,
+      "loss": 3.0057,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 61
+    },
+    {
+      "epoch": 0.12509457755359396,
+      "grad_norm": 0.96484375,
+      "learning_rate": 4.92796850741033e-06,
+      "loss": 2.9342,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 62
+    },
+    {
+      "epoch": 0.1271122320302648,
+      "grad_norm": 0.9921875,
+      "learning_rate": 4.9239318764608245e-06,
+      "loss": 2.949,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 63
+    },
+    {
+      "epoch": 0.1291298865069357,
+      "grad_norm": 0.95703125,
+      "learning_rate": 4.919786946916282e-06,
+      "loss": 2.9372,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 64
+    },
+    {
+      "epoch": 0.13114754098360656,
+      "grad_norm": 0.79296875,
+      "learning_rate": 4.91553390396757e-06,
+      "loss": 2.9564,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 65
+    },
+    {
+      "epoch": 0.13316519546027744,
+      "grad_norm": 0.90234375,
+      "learning_rate": 4.911172937635942e-06,
+      "loss": 2.9389,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 66
+    },
+    {
+      "epoch": 0.1351828499369483,
+      "grad_norm": 0.79296875,
+      "learning_rate": 4.906704242764551e-06,
+      "loss": 2.8962,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 67
+    },
+    {
+      "epoch": 0.13720050441361917,
+      "grad_norm": 0.76953125,
+      "learning_rate": 4.902128019009741e-06,
+      "loss": 2.9853,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 68
+    },
+    {
+      "epoch": 0.13921815889029004,
+      "grad_norm": 0.77734375,
+      "learning_rate": 4.8974444708321265e-06,
+      "loss": 2.9407,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 69
+    },
+    {
+      "epoch": 0.14123581336696092,
+      "grad_norm": 0.78125,
+      "learning_rate": 4.892653807487461e-06,
+      "loss": 2.9395,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 70
+    },
+    {
+      "epoch": 0.14325346784363177,
+      "grad_norm": 0.72265625,
+      "learning_rate": 4.887756243017282e-06,
+      "loss": 3.0041,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 71
+    },
+    {
+      "epoch": 0.14527112232030265,
+      "grad_norm": 0.63671875,
+      "learning_rate": 4.882751996239352e-06,
+      "loss": 2.9692,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 72
+    },
+    {
+      "epoch": 0.14728877679697353,
+      "grad_norm": 0.6875,
+      "learning_rate": 4.8776412907378845e-06,
+      "loss": 2.9279,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 73
+    },
+    {
+      "epoch": 0.1493064312736444,
+      "grad_norm": 0.7109375,
+      "learning_rate": 4.872424354853545e-06,
+      "loss": 2.9776,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 74
+    },
+    {
+      "epoch": 0.15132408575031525,
+      "grad_norm": 0.80859375,
+      "learning_rate": 4.867101421673261e-06,
+      "loss": 3.0093,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 75
+    },
+    {
+      "epoch": 0.15334174022698613,
+      "grad_norm": 0.62890625,
+      "learning_rate": 4.861672729019798e-06,
+      "loss": 2.9489,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 76
+    },
+    {
+      "epoch": 0.155359394703657,
+      "grad_norm": 0.6171875,
+      "learning_rate": 4.856138519441137e-06,
+      "loss": 2.8465,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 77
+    },
+    {
+      "epoch": 0.15737704918032788,
+      "grad_norm": 0.7265625,
+      "learning_rate": 4.8504990401996434e-06,
+      "loss": 2.9363,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 78
+    },
+    {
+      "epoch": 0.15939470365699873,
+      "grad_norm": 0.53125,
+      "learning_rate": 4.8447545432610095e-06,
+      "loss": 2.8965,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 79
+    },
+    {
+      "epoch": 0.1614123581336696,
+      "grad_norm": 0.61328125,
+      "learning_rate": 4.8389052852830055e-06,
+      "loss": 2.96,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 80
+    },
+    {
+      "epoch": 0.16343001261034049,
+      "grad_norm": 0.578125,
+      "learning_rate": 4.832951527604007e-06,
+      "loss": 2.9506,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 81
+    },
+    {
+      "epoch": 0.16544766708701136,
+      "grad_norm": 0.59375,
+      "learning_rate": 4.826893536231322e-06,
+      "loss": 2.8551,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 82
+    },
+    {
+      "epoch": 0.1674653215636822,
+      "grad_norm": 0.625,
+      "learning_rate": 4.820731581829303e-06,
+      "loss": 3.0159,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 83
+    },
+    {
+      "epoch": 0.1694829760403531,
+      "grad_norm": 0.5234375,
+      "learning_rate": 4.814465939707259e-06,
+      "loss": 2.9186,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 84
+    },
+    {
+      "epoch": 0.17150063051702397,
+      "grad_norm": 0.51171875,
+      "learning_rate": 4.808096889807147e-06,
+      "loss": 3.0161,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 85
+    },
+    {
+      "epoch": 0.17351828499369484,
+      "grad_norm": 0.57421875,
+      "learning_rate": 4.801624716691072e-06,
+      "loss": 2.9589,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 86
+    },
+    {
+      "epoch": 0.1755359394703657,
+      "grad_norm": 0.5625,
+      "learning_rate": 4.795049709528571e-06,
+      "loss": 2.9643,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 87
+    },
+    {
+      "epoch": 0.17755359394703657,
+      "grad_norm": 0.578125,
+      "learning_rate": 4.78837216208369e-06,
+      "loss": 2.9542,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 88
+    },
+    {
+      "epoch": 0.17957124842370745,
+      "grad_norm": 0.50390625,
+      "learning_rate": 4.7815923727018625e-06,
+      "loss": 2.9755,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 89
+    },
+    {
+      "epoch": 0.18158890290037832,
+      "grad_norm": 0.54296875,
+      "learning_rate": 4.774710644296579e-06,
+      "loss": 2.9595,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 90
+    },
+    {
+      "epoch": 0.18360655737704917,
+      "grad_norm": 0.48828125,
+      "learning_rate": 4.767727284335852e-06,
+      "loss": 2.9172,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 91
+    },
+    {
+      "epoch": 0.18562421185372005,
+      "grad_norm": 0.515625,
+      "learning_rate": 4.760642604828482e-06,
+      "loss": 2.9182,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 92
+    },
+    {
+      "epoch": 0.18764186633039093,
+      "grad_norm": 0.59375,
+      "learning_rate": 4.753456922310109e-06,
+      "loss": 2.9225,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 93
+    },
+    {
+      "epoch": 0.1896595208070618,
+      "grad_norm": 0.515625,
+      "learning_rate": 4.746170557829084e-06,
+      "loss": 2.8674,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 94
+    },
+    {
+      "epoch": 0.19167717528373265,
+      "grad_norm": 0.4921875,
+      "learning_rate": 4.738783836932109e-06,
+      "loss": 2.8713,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 95
+    },
+    {
+      "epoch": 0.19369482976040353,
+      "grad_norm": 0.478515625,
+      "learning_rate": 4.731297089649704e-06,
+      "loss": 3.0112,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 96
+    },
+    {
+      "epoch": 0.1957124842370744,
+      "grad_norm": 0.46875,
+      "learning_rate": 4.723710650481456e-06,
+      "loss": 2.9569,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 97
+    },
+    {
+      "epoch": 0.19773013871374528,
+      "grad_norm": 0.48828125,
+      "learning_rate": 4.7160248583810755e-06,
+      "loss": 2.9466,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 98
+    },
+    {
+      "epoch": 0.19974779319041613,
+      "grad_norm": 0.48828125,
+      "learning_rate": 4.708240056741253e-06,
+      "loss": 2.9399,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 99
+    },
+    {
+      "epoch": 0.201765447667087,
+      "grad_norm": 0.51171875,
+      "learning_rate": 4.700356593378312e-06,
+      "loss": 2.9445,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 100
+    },
+    {
+      "epoch": 0.2037831021437579,
+      "grad_norm": 0.51953125,
+      "learning_rate": 4.692374820516679e-06,
+      "loss": 2.9657,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 101
+    },
+    {
+      "epoch": 0.20580075662042877,
+      "grad_norm": 0.53125,
+      "learning_rate": 4.684295094773134e-06,
+      "loss": 3.0236,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 102
+    },
+    {
+      "epoch": 0.20781841109709961,
+      "grad_norm": 0.470703125,
+      "learning_rate": 4.676117777140887e-06,
+      "loss": 2.9442,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 103
+    },
+    {
+      "epoch": 0.2098360655737705,
+      "grad_norm": 0.490234375,
+      "learning_rate": 4.667843232973444e-06,
+      "loss": 2.9491,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 104
+    },
+    {
+      "epoch": 0.21185372005044137,
+      "grad_norm": 0.5390625,
+      "learning_rate": 4.659471831968285e-06,
+      "loss": 3.0061,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 105
+    },
+    {
+      "epoch": 0.21387137452711225,
+      "grad_norm": 0.474609375,
+      "learning_rate": 4.651003948150349e-06,
+      "loss": 2.8666,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 106
+    },
+    {
+      "epoch": 0.2158890290037831,
+      "grad_norm": 0.4375,
+      "learning_rate": 4.642439959855316e-06,
+      "loss": 2.9507,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 107
+    },
+    {
+      "epoch": 0.21790668348045397,
+      "grad_norm": 0.48828125,
+      "learning_rate": 4.633780249712712e-06,
+      "loss": 2.8984,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 108
+    },
+    {
+      "epoch": 0.21992433795712485,
+      "grad_norm": 0.421875,
+      "learning_rate": 4.625025204628806e-06,
+      "loss": 2.8593,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 109
+    },
+    {
+      "epoch": 0.22194199243379573,
+      "grad_norm": 0.470703125,
+      "learning_rate": 4.616175215769328e-06,
+      "loss": 2.8705,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 110
+    },
+    {
+      "epoch": 0.22395964691046658,
+      "grad_norm": 0.482421875,
+      "learning_rate": 4.607230678541993e-06,
+      "loss": 2.916,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 111
+    },
+    {
+      "epoch": 0.22597730138713745,
+      "grad_norm": 0.39453125,
+      "learning_rate": 4.5981919925788285e-06,
+      "loss": 2.921,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 112
+    },
+    {
+      "epoch": 0.22799495586380833,
+      "grad_norm": 0.40625,
+      "learning_rate": 4.5890595617183254e-06,
+      "loss": 2.8851,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 113
+    },
+    {
+      "epoch": 0.2300126103404792,
+      "grad_norm": 0.443359375,
+      "learning_rate": 4.579833793987393e-06,
+      "loss": 2.9397,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 114
+    },
+    {
+      "epoch": 0.23203026481715006,
+      "grad_norm": 0.47265625,
+      "learning_rate": 4.570515101583128e-06,
+      "loss": 2.8029,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 115
+    },
+    {
+      "epoch": 0.23404791929382093,
+      "grad_norm": 0.447265625,
+      "learning_rate": 4.561103900854401e-06,
+      "loss": 2.9667,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 116
+    },
+    {
+      "epoch": 0.2360655737704918,
+      "grad_norm": 0.435546875,
+      "learning_rate": 4.551600612283249e-06,
+      "loss": 3.009,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 117
+    },
+    {
+      "epoch": 0.2380832282471627,
+      "grad_norm": 0.427734375,
+      "learning_rate": 4.542005660466095e-06,
+      "loss": 2.7894,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 118
+    },
+    {
+      "epoch": 0.24010088272383354,
+      "grad_norm": 0.45703125,
+      "learning_rate": 4.532319474094769e-06,
+      "loss": 2.9377,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 119
+    },
+    {
+      "epoch": 0.2421185372005044,
+      "grad_norm": 0.4609375,
+      "learning_rate": 4.522542485937369e-06,
+      "loss": 2.8978,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 120
+    },
+    {
+      "epoch": 0.2441361916771753,
+      "grad_norm": 0.45703125,
+      "learning_rate": 4.512675132818908e-06,
+      "loss": 2.9091,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 121
+    },
+    {
+      "epoch": 0.24615384615384617,
+      "grad_norm": 0.4375,
+      "learning_rate": 4.5027178556018095e-06,
+      "loss": 2.9647,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 122
+    },
+    {
+      "epoch": 0.24817150063051702,
+      "grad_norm": 0.431640625,
+      "learning_rate": 4.492671099166204e-06,
+      "loss": 2.94,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 123
+    },
+    {
+      "epoch": 0.2501891551071879,
+      "grad_norm": 0.412109375,
+      "learning_rate": 4.482535312390059e-06,
+      "loss": 2.8902,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 124
+    },
+    {
+      "epoch": 0.25220680958385877,
+      "grad_norm": 0.421875,
+      "learning_rate": 4.472310948129113e-06,
+      "loss": 3.0486,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 125
+    },
+    {
+      "epoch": 0.2542244640605296,
+      "grad_norm": 0.47265625,
+      "learning_rate": 4.461998463196653e-06,
+      "loss": 2.848,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 126
+    },
+    {
+      "epoch": 0.2562421185372005,
+      "grad_norm": 0.447265625,
+      "learning_rate": 4.451598318343099e-06,
+      "loss": 2.9275,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 127
+    },
+    {
+      "epoch": 0.2582597730138714,
+      "grad_norm": 0.443359375,
+      "learning_rate": 4.441110978235419e-06,
+      "loss": 3.0023,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 128
+    },
+    {
+      "epoch": 0.2602774274905422,
+      "grad_norm": 0.421875,
+      "learning_rate": 4.430536911436368e-06,
+      "loss": 2.8752,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 129
+    },
+    {
+      "epoch": 0.26229508196721313,
+      "grad_norm": 0.447265625,
+      "learning_rate": 4.419876590383554e-06,
+      "loss": 2.9107,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 130
+    },
+    {
+      "epoch": 0.264312736443884,
+      "grad_norm": 0.376953125,
+      "learning_rate": 4.409130491368331e-06,
+      "loss": 2.9529,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 131
+    },
+    {
+      "epoch": 0.2663303909205549,
+      "grad_norm": 0.400390625,
+      "learning_rate": 4.398299094514515e-06,
+      "loss": 2.9129,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 132
+    },
+    {
+      "epoch": 0.26834804539722573,
+      "grad_norm": 0.376953125,
+      "learning_rate": 4.387382883756938e-06,
+      "loss": 2.9802,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 133
+    },
+    {
+      "epoch": 0.2703656998738966,
+      "grad_norm": 0.39453125,
+      "learning_rate": 4.37638234681982e-06,
+      "loss": 2.8098,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 134
+    },
+    {
+      "epoch": 0.2723833543505675,
+      "grad_norm": 0.365234375,
+      "learning_rate": 4.365297975194984e-06,
+      "loss": 2.9357,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 135
+    },
+    {
+      "epoch": 0.27440100882723834,
+      "grad_norm": 0.48828125,
+      "learning_rate": 4.354130264119894e-06,
+      "loss": 2.9512,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 136
+    },
+    {
+      "epoch": 0.2764186633039092,
+      "grad_norm": 0.52734375,
+      "learning_rate": 4.342879712555528e-06,
+      "loss": 2.9478,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 137
+    },
+    {
+      "epoch": 0.2784363177805801,
+      "grad_norm": 0.423828125,
+      "learning_rate": 4.331546823164083e-06,
+      "loss": 3.0044,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 138
+    },
+    {
+      "epoch": 0.28045397225725094,
+      "grad_norm": 0.388671875,
+      "learning_rate": 4.320132102286524e-06,
+      "loss": 2.9148,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 139
+    },
+    {
+      "epoch": 0.28247162673392184,
+      "grad_norm": 0.36328125,
+      "learning_rate": 4.308636059919952e-06,
+      "loss": 2.8966,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 140
+    },
+    {
+      "epoch": 0.2844892812105927,
+      "grad_norm": 0.40625,
+      "learning_rate": 4.297059209694824e-06,
+      "loss": 2.7564,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 141
+    },
+    {
+      "epoch": 0.28650693568726354,
+      "grad_norm": 0.400390625,
+      "learning_rate": 4.2854020688520025e-06,
+      "loss": 2.809,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 142
+    },
+    {
+      "epoch": 0.28852459016393445,
+      "grad_norm": 0.416015625,
+      "learning_rate": 4.273665158219645e-06,
+      "loss": 2.9237,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 143
+    },
+    {
+      "epoch": 0.2905422446406053,
+      "grad_norm": 0.3828125,
+      "learning_rate": 4.261849002189939e-06,
+      "loss": 2.9274,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 144
+    },
+    {
+      "epoch": 0.29255989911727615,
+      "grad_norm": 0.40234375,
+      "learning_rate": 4.249954128695662e-06,
+      "loss": 2.909,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 145
+    },
+    {
+      "epoch": 0.29457755359394705,
+      "grad_norm": 0.359375,
+      "learning_rate": 4.237981069186606e-06,
+      "loss": 2.8917,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 146
+    },
+    {
+      "epoch": 0.2965952080706179,
+      "grad_norm": 0.39453125,
+      "learning_rate": 4.225930358605827e-06,
+      "loss": 2.8702,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 147
+    },
+    {
+      "epoch": 0.2986128625472888,
+      "grad_norm": 0.373046875,
+      "learning_rate": 4.213802535365741e-06,
+      "loss": 2.9231,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 148
+    },
+    {
+      "epoch": 0.30063051702395965,
+      "grad_norm": 0.376953125,
+      "learning_rate": 4.201598141324078e-06,
+      "loss": 2.8554,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 149
+    },
+    {
+      "epoch": 0.3026481715006305,
+      "grad_norm": 0.376953125,
+      "learning_rate": 4.189317721759663e-06,
+      "loss": 2.8243,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 150
+    },
+    {
+      "epoch": 0.3046658259773014,
+      "grad_norm": 0.400390625,
+      "learning_rate": 4.176961825348059e-06,
+      "loss": 3.031,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 151
+    },
+    {
+      "epoch": 0.30668348045397226,
+      "grad_norm": 0.37109375,
+      "learning_rate": 4.16453100413705e-06,
+      "loss": 2.8808,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 152
+    },
+    {
+      "epoch": 0.3087011349306431,
+      "grad_norm": 0.36328125,
+      "learning_rate": 4.152025813521976e-06,
+      "loss": 2.8957,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 153
+    },
+    {
+      "epoch": 0.310718789407314,
+      "grad_norm": 0.400390625,
+      "learning_rate": 4.1394468122209245e-06,
+      "loss": 3.0874,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 154
+    },
+    {
+      "epoch": 0.31273644388398486,
+      "grad_norm": 0.392578125,
+      "learning_rate": 4.1267945622497566e-06,
+      "loss": 2.9214,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 155
+    },
+    {
+      "epoch": 0.31475409836065577,
+      "grad_norm": 0.37890625,
+      "learning_rate": 4.114069628897006e-06,
+      "loss": 2.9121,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 156
+    },
+    {
+      "epoch": 0.3167717528373266,
+      "grad_norm": 0.388671875,
+      "learning_rate": 4.101272580698621e-06,
+      "loss": 2.944,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 157
+    },
+    {
+      "epoch": 0.31878940731399746,
+      "grad_norm": 0.40234375,
+      "learning_rate": 4.08840398941256e-06,
+      "loss": 2.928,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 158
+    },
+    {
+      "epoch": 0.32080706179066837,
+      "grad_norm": 0.390625,
+      "learning_rate": 4.075464429993244e-06,
+      "loss": 2.8135,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 159
+    },
+    {
+      "epoch": 0.3228247162673392,
+      "grad_norm": 0.369140625,
+      "learning_rate": 4.0624544805658795e-06,
+      "loss": 2.93,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 160
+    },
+    {
+      "epoch": 0.32484237074401007,
+      "grad_norm": 0.3515625,
+      "learning_rate": 4.049374722400613e-06,
+      "loss": 2.7283,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 161
+    },
+    {
+      "epoch": 0.32686002522068097,
+      "grad_norm": 0.380859375,
+      "learning_rate": 4.0362257398865715e-06,
+      "loss": 2.9056,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 162
+    },
+    {
+      "epoch": 0.3288776796973518,
+      "grad_norm": 0.37890625,
+      "learning_rate": 4.02300812050575e-06,
+      "loss": 2.9232,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 163
+    },
+    {
+      "epoch": 0.3308953341740227,
+      "grad_norm": 0.380859375,
+      "learning_rate": 4.009722454806762e-06,
+      "loss": 2.8507,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 164
+    },
+    {
+      "epoch": 0.3329129886506936,
+      "grad_norm": 0.388671875,
+      "learning_rate": 3.9963693363784544e-06,
+      "loss": 2.8988,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 165
+    },
+    {
+      "epoch": 0.3349306431273644,
+      "grad_norm": 0.35546875,
+      "learning_rate": 3.982949361823388e-06,
+      "loss": 2.8836,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 166
+    },
+    {
+      "epoch": 0.33694829760403533,
+      "grad_norm": 0.369140625,
+      "learning_rate": 3.969463130731183e-06,
+      "loss": 2.8122,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 167
+    },
+    {
+      "epoch": 0.3389659520807062,
+      "grad_norm": 0.361328125,
+      "learning_rate": 3.955911245651726e-06,
+      "loss": 2.9137,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 168
+    },
+    {
+      "epoch": 0.34098360655737703,
+      "grad_norm": 0.359375,
+      "learning_rate": 3.942294312068252e-06,
+      "loss": 2.9557,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 169
+    },
+    {
+      "epoch": 0.34300126103404793,
+      "grad_norm": 0.380859375,
+      "learning_rate": 3.928612938370292e-06,
+      "loss": 2.8614,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 170
+    },
+    {
+      "epoch": 0.3450189155107188,
+      "grad_norm": 0.376953125,
+      "learning_rate": 3.914867735826489e-06,
+      "loss": 2.9181,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 171
+    },
+    {
+      "epoch": 0.3470365699873897,
+      "grad_norm": 0.35546875,
+      "learning_rate": 3.901059318557287e-06,
+      "loss": 2.8829,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 172
+    },
+    {
+      "epoch": 0.34905422446406054,
+      "grad_norm": 0.421875,
+      "learning_rate": 3.8871883035074975e-06,
+      "loss": 2.9295,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 173
+    },
+    {
+      "epoch": 0.3510718789407314,
+      "grad_norm": 0.36328125,
+      "learning_rate": 3.87325531041873e-06,
+      "loss": 2.8678,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 174
+    },
+    {
+      "epoch": 0.3530895334174023,
+      "grad_norm": 0.376953125,
+      "learning_rate": 3.859260961801702e-06,
+      "loss": 2.871,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 175
+    },
+    {
+      "epoch": 0.35510718789407314,
+      "grad_norm": 0.3515625,
+      "learning_rate": 3.845205882908432e-06,
+      "loss": 2.9155,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 176
+    },
+    {
+      "epoch": 0.357124842370744,
+      "grad_norm": 0.341796875,
+      "learning_rate": 3.8310907017042966e-06,
+      "loss": 2.8222,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 177
+    },
+    {
+      "epoch": 0.3591424968474149,
+      "grad_norm": 0.37890625,
+      "learning_rate": 3.816916048839979e-06,
+      "loss": 2.9431,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 178
+    },
+    {
+      "epoch": 0.36116015132408574,
+      "grad_norm": 0.421875,
+      "learning_rate": 3.8026825576232906e-06,
+      "loss": 2.8602,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 179
+    },
+    {
+      "epoch": 0.36317780580075665,
+      "grad_norm": 0.380859375,
+      "learning_rate": 3.7883908639908752e-06,
+      "loss": 2.9468,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 180
+    },
+    {
+      "epoch": 0.3651954602774275,
+      "grad_norm": 0.365234375,
+      "learning_rate": 3.774041606479794e-06,
+      "loss": 2.9297,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 181
+    },
+    {
+      "epoch": 0.36721311475409835,
+      "grad_norm": 0.404296875,
+      "learning_rate": 3.759635426199001e-06,
+      "loss": 2.8849,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 182
+    },
+    {
+      "epoch": 0.36923076923076925,
+      "grad_norm": 0.373046875,
+      "learning_rate": 3.7451729668006974e-06,
+      "loss": 2.8075,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 183
+    },
+    {
+      "epoch": 0.3712484237074401,
+      "grad_norm": 0.359375,
+      "learning_rate": 3.730654874451569e-06,
+      "loss": 2.7931,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 184
+    },
+    {
+      "epoch": 0.37326607818411095,
+      "grad_norm": 0.375,
+      "learning_rate": 3.7160817978039256e-06,
+      "loss": 2.9012,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 185
+    },
+    {
+      "epoch": 0.37528373266078185,
+      "grad_norm": 0.34765625,
+      "learning_rate": 3.7014543879667097e-06,
+      "loss": 2.8396,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 186
+    },
+    {
+      "epoch": 0.3773013871374527,
+      "grad_norm": 0.345703125,
+      "learning_rate": 3.6867732984764144e-06,
+      "loss": 2.8584,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 187
+    },
+    {
+      "epoch": 0.3793190416141236,
+      "grad_norm": 0.373046875,
+      "learning_rate": 3.6720391852678783e-06,
+      "loss": 2.9347,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 188
+    },
+    {
+      "epoch": 0.38133669609079446,
+      "grad_norm": 0.353515625,
+      "learning_rate": 3.657252706644982e-06,
+      "loss": 2.9129,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 189
+    },
+    {
+      "epoch": 0.3833543505674653,
+      "grad_norm": 0.365234375,
+      "learning_rate": 3.6424145232512337e-06,
+      "loss": 2.8445,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 190
+    },
+    {
+      "epoch": 0.3853720050441362,
+      "grad_norm": 0.365234375,
+      "learning_rate": 3.627525298040255e-06,
+      "loss": 2.8571,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 191
+    },
+    {
+      "epoch": 0.38738965952080706,
+      "grad_norm": 0.36328125,
+      "learning_rate": 3.612585696246158e-06,
+      "loss": 2.9072,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 192
+    },
+    {
+      "epoch": 0.3894073139974779,
+      "grad_norm": 0.35546875,
+      "learning_rate": 3.5975963853538273e-06,
+      "loss": 2.9549,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 193
+    },
+    {
+      "epoch": 0.3914249684741488,
+      "grad_norm": 0.365234375,
+      "learning_rate": 3.5825580350690914e-06,
+      "loss": 2.8663,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 194
+    },
+    {
+      "epoch": 0.39344262295081966,
+      "grad_norm": 0.3515625,
+      "learning_rate": 3.5674713172888075e-06,
+      "loss": 3.0238,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 195
+    },
+    {
+      "epoch": 0.39546027742749057,
+      "grad_norm": 0.3671875,
+      "learning_rate": 3.552336906070838e-06,
+      "loss": 2.922,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 196
+    },
+    {
+      "epoch": 0.3974779319041614,
+      "grad_norm": 0.34375,
+      "learning_rate": 3.5371554776039344e-06,
+      "loss": 2.8983,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 197
+    },
+    {
+      "epoch": 0.39949558638083227,
+      "grad_norm": 0.3515625,
+      "learning_rate": 3.52192771017753e-06,
+      "loss": 2.9008,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 198
+    },
+    {
+      "epoch": 0.4015132408575032,
+      "grad_norm": 0.357421875,
+      "learning_rate": 3.5066542841514275e-06,
+      "loss": 2.9236,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 199
+    },
+    {
+      "epoch": 0.403530895334174,
+      "grad_norm": 0.400390625,
+      "learning_rate": 3.491335881925407e-06,
+      "loss": 2.8329,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 200
+    },
+    {
+      "epoch": 0.40554854981084487,
+      "grad_norm": 0.37890625,
+      "learning_rate": 3.4759731879087373e-06,
+      "loss": 2.8975,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 201
+    },
+    {
+      "epoch": 0.4075662042875158,
+      "grad_norm": 0.359375,
+      "learning_rate": 3.460566888489593e-06,
+      "loss": 2.9363,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 202
+    },
+    {
+      "epoch": 0.4095838587641866,
+      "grad_norm": 0.35546875,
+      "learning_rate": 3.4451176720043906e-06,
+      "loss": 2.9014,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 203
+    },
+    {
+      "epoch": 0.41160151324085753,
+      "grad_norm": 0.35546875,
+      "learning_rate": 3.4296262287070337e-06,
+      "loss": 2.8582,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 204
+    },
+    {
+      "epoch": 0.4136191677175284,
+      "grad_norm": 0.375,
+      "learning_rate": 3.4140932507380727e-06,
+      "loss": 2.9568,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 205
+    },
+    {
+      "epoch": 0.41563682219419923,
+      "grad_norm": 0.359375,
+      "learning_rate": 3.398519432093782e-06,
+      "loss": 2.7844,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 206
+    },
+    {
+      "epoch": 0.41765447667087013,
+      "grad_norm": 0.341796875,
+      "learning_rate": 3.3829054685951535e-06,
+      "loss": 2.9679,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 207
+    },
+    {
+      "epoch": 0.419672131147541,
+      "grad_norm": 0.353515625,
+      "learning_rate": 3.3672520578568018e-06,
+      "loss": 3.0012,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 208
+    },
+    {
+      "epoch": 0.42168978562421183,
+      "grad_norm": 0.359375,
+      "learning_rate": 3.351559899255806e-06,
+      "loss": 2.8042,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 209
+    },
+    {
+      "epoch": 0.42370744010088274,
+      "grad_norm": 0.34765625,
+      "learning_rate": 3.335829693900455e-06,
+      "loss": 2.8814,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 210
+    },
+    {
+      "epoch": 0.4257250945775536,
+      "grad_norm": 0.36328125,
+      "learning_rate": 3.3200621445989227e-06,
+      "loss": 2.8684,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 211
+    },
+    {
+      "epoch": 0.4277427490542245,
+      "grad_norm": 0.3671875,
+      "learning_rate": 3.304257955827872e-06,
+      "loss": 2.8368,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 212
+    },
+    {
+      "epoch": 0.42976040353089534,
+      "grad_norm": 0.36328125,
+      "learning_rate": 3.2884178337009764e-06,
+      "loss": 2.8657,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 213
+    },
+    {
+      "epoch": 0.4317780580075662,
+      "grad_norm": 0.353515625,
+      "learning_rate": 3.272542485937369e-06,
+      "loss": 2.8492,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 214
+    },
+    {
+      "epoch": 0.4337957124842371,
+      "grad_norm": 0.33984375,
+      "learning_rate": 3.2566326218300287e-06,
+      "loss": 2.904,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 215
+    },
+    {
+      "epoch": 0.43581336696090794,
+      "grad_norm": 0.41796875,
+      "learning_rate": 3.2406889522140854e-06,
+      "loss": 2.9362,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 216
+    },
+    {
+      "epoch": 0.4378310214375788,
+      "grad_norm": 0.34375,
+      "learning_rate": 3.2247121894350614e-06,
+      "loss": 2.8333,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 217
+    },
+    {
+      "epoch": 0.4398486759142497,
+      "grad_norm": 0.3515625,
+      "learning_rate": 3.208703047317045e-06,
+      "loss": 2.9378,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 218
+    },
+    {
+      "epoch": 0.44186633039092055,
+      "grad_norm": 0.330078125,
+      "learning_rate": 3.1926622411307985e-06,
+      "loss": 2.8866,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 219
+    },
+    {
+      "epoch": 0.44388398486759145,
+      "grad_norm": 0.33203125,
+      "learning_rate": 3.1765904875617977e-06,
+      "loss": 2.8879,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 220
+    },
+    {
+      "epoch": 0.4459016393442623,
+      "grad_norm": 0.5390625,
+      "learning_rate": 3.1604885046782158e-06,
+      "loss": 2.8542,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 221
+    },
+    {
+      "epoch": 0.44791929382093315,
+      "grad_norm": 0.35546875,
+      "learning_rate": 3.1443570118988357e-06,
+      "loss": 2.9413,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 222
+    },
+    {
+      "epoch": 0.44993694829760406,
+      "grad_norm": 0.353515625,
+      "learning_rate": 3.128196729960912e-06,
+      "loss": 2.9176,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 223
+    },
+    {
+      "epoch": 0.4519546027742749,
+      "grad_norm": 0.396484375,
+      "learning_rate": 3.1120083808879666e-06,
+      "loss": 2.9672,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 224
+    },
+    {
+      "epoch": 0.45397225725094575,
+      "grad_norm": 0.33984375,
+      "learning_rate": 3.095792687957528e-06,
+      "loss": 2.9543,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 225
+    },
+    {
+      "epoch": 0.45598991172761666,
+      "grad_norm": 0.33984375,
+      "learning_rate": 3.0795503756688212e-06,
+      "loss": 2.7878,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 226
+    },
+    {
+      "epoch": 0.4580075662042875,
+      "grad_norm": 0.365234375,
+      "learning_rate": 3.063282169710392e-06,
+      "loss": 2.9394,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 227
+    },
+    {
+      "epoch": 0.4600252206809584,
+      "grad_norm": 0.373046875,
+      "learning_rate": 3.046988796927688e-06,
+      "loss": 2.838,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 228
+    },
+    {
+      "epoch": 0.46204287515762926,
+      "grad_norm": 0.357421875,
+      "learning_rate": 3.0306709852905824e-06,
+      "loss": 2.9316,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 229
+    },
+    {
+      "epoch": 0.4640605296343001,
+      "grad_norm": 0.337890625,
+      "learning_rate": 3.014329463860849e-06,
+      "loss": 2.8895,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 230
+    },
+    {
+      "epoch": 0.466078184110971,
+      "grad_norm": 0.375,
+      "learning_rate": 2.9979649627595904e-06,
+      "loss": 2.9382,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 231
+    },
+    {
+      "epoch": 0.46809583858764187,
+      "grad_norm": 0.375,
+      "learning_rate": 2.981578213134614e-06,
+      "loss": 2.9179,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 232
+    },
+    {
+      "epoch": 0.4701134930643127,
+      "grad_norm": 0.314453125,
+      "learning_rate": 2.9651699471277664e-06,
+      "loss": 2.803,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 233
+    },
+    {
+      "epoch": 0.4721311475409836,
+      "grad_norm": 0.453125,
+      "learning_rate": 2.9487408978422233e-06,
+      "loss": 2.8782,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 234
+    },
+    {
+      "epoch": 0.47414880201765447,
+      "grad_norm": 0.357421875,
+      "learning_rate": 2.932291799309734e-06,
+      "loss": 2.9224,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 235
+    },
+    {
+      "epoch": 0.4761664564943254,
+      "grad_norm": 0.322265625,
+      "learning_rate": 2.9158233864578256e-06,
+      "loss": 2.8951,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 236
+    },
+    {
+      "epoch": 0.4781841109709962,
+      "grad_norm": 0.400390625,
+      "learning_rate": 2.8993363950769685e-06,
+      "loss": 2.8369,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 237
+    },
+    {
+      "epoch": 0.4802017654476671,
+      "grad_norm": 0.357421875,
+      "learning_rate": 2.8828315617877006e-06,
+      "loss": 2.9335,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 238
+    },
+    {
+      "epoch": 0.482219419924338,
+      "grad_norm": 0.337890625,
+      "learning_rate": 2.866309624007717e-06,
+      "loss": 2.838,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 239
+    },
+    {
+      "epoch": 0.4842370744010088,
+      "grad_norm": 0.333984375,
+      "learning_rate": 2.849771319918922e-06,
+      "loss": 2.9317,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 240
+    },
+    {
+      "epoch": 0.4862547288776797,
+      "grad_norm": 0.3671875,
+      "learning_rate": 2.8332173884344477e-06,
+      "loss": 2.8651,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 241
+    },
+    {
+      "epoch": 0.4882723833543506,
+      "grad_norm": 0.349609375,
+      "learning_rate": 2.8166485691656425e-06,
+      "loss": 2.8215,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 242
+    },
+    {
+      "epoch": 0.49029003783102143,
+      "grad_norm": 0.33203125,
+      "learning_rate": 2.8000656023890245e-06,
+      "loss": 2.909,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 243
+    },
+    {
+      "epoch": 0.49230769230769234,
+      "grad_norm": 0.361328125,
+      "learning_rate": 2.7834692290132054e-06,
+      "loss": 2.8417,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 244
+    },
+    {
+      "epoch": 0.4943253467843632,
+      "grad_norm": 0.35546875,
+      "learning_rate": 2.766860190545791e-06,
+      "loss": 2.8922,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 245
+    },
+    {
+      "epoch": 0.49634300126103403,
+      "grad_norm": 0.3515625,
+      "learning_rate": 2.7502392290602463e-06,
+      "loss": 2.8867,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 246
+    },
+    {
+      "epoch": 0.49836065573770494,
+      "grad_norm": 0.3203125,
+      "learning_rate": 2.7336070871627467e-06,
+      "loss": 2.8258,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 247
+    },
+    {
+      "epoch": 0.5003783102143758,
+      "grad_norm": 0.345703125,
+      "learning_rate": 2.716964507958994e-06,
+      "loss": 2.8912,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 248
+    },
+    {
+      "epoch": 0.5023959646910466,
+      "grad_norm": 0.349609375,
+      "learning_rate": 2.7003122350210185e-06,
+      "loss": 2.9522,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 249
+    },
+    {
+      "epoch": 0.5044136191677175,
+      "grad_norm": 0.33984375,
+      "learning_rate": 2.6836510123539556e-06,
+      "loss": 2.8601,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 250
+    },
+    {
+      "epoch": 0.5064312736443884,
+      "grad_norm": 0.341796875,
+      "learning_rate": 2.6669815843628043e-06,
+      "loss": 2.8434,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 251
+    },
+    {
+      "epoch": 0.5084489281210592,
+      "grad_norm": 0.3359375,
+      "learning_rate": 2.650304695819168e-06,
+      "loss": 2.9957,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 252
+    },
+    {
+      "epoch": 0.5104665825977301,
+      "grad_norm": 0.388671875,
+      "learning_rate": 2.6336210918279807e-06,
+      "loss": 2.9345,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 253
+    },
+    {
+      "epoch": 0.512484237074401,
+      "grad_norm": 0.373046875,
+      "learning_rate": 2.6169315177942134e-06,
+      "loss": 2.8346,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 254
+    },
+    {
+      "epoch": 0.5145018915510718,
+      "grad_norm": 0.341796875,
+      "learning_rate": 2.6002367193895733e-06,
+      "loss": 2.8084,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 255
+    },
+    {
+      "epoch": 0.5165195460277427,
+      "grad_norm": 0.333984375,
+      "learning_rate": 2.5835374425191867e-06,
+      "loss": 2.9052,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 256
+    },
+    {
+      "epoch": 0.5185372005044137,
+      "grad_norm": 0.3359375,
+      "learning_rate": 2.566834433288272e-06,
+      "loss": 2.8175,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 257
+    },
+    {
+      "epoch": 0.5205548549810844,
+      "grad_norm": 0.341796875,
+      "learning_rate": 2.5501284379688067e-06,
+      "loss": 2.8486,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 258
+    },
+    {
+      "epoch": 0.5225725094577554,
+      "grad_norm": 0.35546875,
+      "learning_rate": 2.533420202966182e-06,
+      "loss": 2.8192,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 259
+    },
+    {
+      "epoch": 0.5245901639344263,
+      "grad_norm": 0.46484375,
+      "learning_rate": 2.516710474785856e-06,
+      "loss": 2.9291,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 260
+    },
+    {
+      "epoch": 0.526607818411097,
+      "grad_norm": 0.34765625,
+      "learning_rate": 2.5e-06,
+      "loss": 2.9709,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 261
+    },
+    {
+      "epoch": 0.528625472887768,
+      "grad_norm": 0.310546875,
+      "learning_rate": 2.483289525214145e-06,
+      "loss": 2.8498,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 262
+    },
+    {
+      "epoch": 0.5306431273644389,
+      "grad_norm": 0.353515625,
+      "learning_rate": 2.4665797970338183e-06,
+      "loss": 2.9485,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 263
+    },
+    {
+      "epoch": 0.5326607818411098,
+      "grad_norm": 0.33984375,
+      "learning_rate": 2.4498715620311937e-06,
+      "loss": 2.9009,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 264
+    },
+    {
+      "epoch": 0.5346784363177806,
+      "grad_norm": 0.337890625,
+      "learning_rate": 2.4331655667117284e-06,
+      "loss": 2.8577,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 265
+    },
+    {
+      "epoch": 0.5366960907944515,
+      "grad_norm": 0.38671875,
+      "learning_rate": 2.4164625574808145e-06,
+      "loss": 2.8691,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 266
+    },
+    {
+      "epoch": 0.5387137452711224,
+      "grad_norm": 0.447265625,
+      "learning_rate": 2.3997632806104275e-06,
+      "loss": 2.8994,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 267
+    },
+    {
+      "epoch": 0.5407313997477932,
+      "grad_norm": 0.330078125,
+      "learning_rate": 2.383068482205788e-06,
+      "loss": 2.8892,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 268
+    },
+    {
+      "epoch": 0.5427490542244641,
+      "grad_norm": 0.373046875,
+      "learning_rate": 2.36637890817202e-06,
+      "loss": 2.9277,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 269
+    },
+    {
+      "epoch": 0.544766708701135,
+      "grad_norm": 0.361328125,
+      "learning_rate": 2.3496953041808327e-06,
+      "loss": 2.7727,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 270
+    },
+    {
+      "epoch": 0.5467843631778058,
+      "grad_norm": 0.310546875,
+      "learning_rate": 2.333018415637196e-06,
+      "loss": 2.8423,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 271
+    },
+    {
+      "epoch": 0.5488020176544767,
+      "grad_norm": 0.328125,
+      "learning_rate": 2.3163489876460453e-06,
+      "loss": 2.7515,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 272
+    },
+    {
+      "epoch": 0.5508196721311476,
+      "grad_norm": 0.369140625,
+      "learning_rate": 2.2996877649789815e-06,
+      "loss": 3.039,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 273
+    },
+    {
+      "epoch": 0.5528373266078184,
+      "grad_norm": 0.33203125,
+      "learning_rate": 2.2830354920410066e-06,
+      "loss": 2.8529,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 274
+    },
+    {
+      "epoch": 0.5548549810844893,
+      "grad_norm": 0.33203125,
+      "learning_rate": 2.2663929128372537e-06,
+      "loss": 2.8775,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 275
+    },
+    {
+      "epoch": 0.5568726355611602,
+      "grad_norm": 0.333984375,
+      "learning_rate": 2.249760770939754e-06,
+      "loss": 2.9064,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 276
+    },
+    {
+      "epoch": 0.558890290037831,
+      "grad_norm": 0.3359375,
+      "learning_rate": 2.2331398094542097e-06,
+      "loss": 2.8687,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 277
+    },
+    {
+      "epoch": 0.5609079445145019,
+      "grad_norm": 0.333984375,
+      "learning_rate": 2.2165307709867954e-06,
+      "loss": 2.9269,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 278
+    },
+    {
+      "epoch": 0.5629255989911728,
+      "grad_norm": 0.34375,
+      "learning_rate": 2.199934397610976e-06,
+      "loss": 2.877,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 279
+    },
+    {
+      "epoch": 0.5649432534678437,
+      "grad_norm": 0.3203125,
+      "learning_rate": 2.1833514308343583e-06,
+      "loss": 2.9292,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 280
+    },
+    {
+      "epoch": 0.5669609079445145,
+      "grad_norm": 0.345703125,
+      "learning_rate": 2.1667826115655536e-06,
+      "loss": 2.8412,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 281
+    },
+    {
+      "epoch": 0.5689785624211854,
+      "grad_norm": 0.330078125,
+      "learning_rate": 2.150228680081079e-06,
+      "loss": 2.88,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 282
+    },
+    {
+      "epoch": 0.5709962168978563,
+      "grad_norm": 0.3515625,
+      "learning_rate": 2.1336903759922838e-06,
+      "loss": 2.7502,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 283
+    },
+    {
+      "epoch": 0.5730138713745271,
+      "grad_norm": 0.408203125,
+      "learning_rate": 2.1171684382123002e-06,
+      "loss": 2.8315,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 284
+    },
+    {
+      "epoch": 0.575031525851198,
+      "grad_norm": 0.322265625,
+      "learning_rate": 2.1006636049230327e-06,
+      "loss": 2.941,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 285
+    },
+    {
+      "epoch": 0.5770491803278689,
+      "grad_norm": 0.328125,
+      "learning_rate": 2.0841766135421753e-06,
+      "loss": 2.9317,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 286
+    },
+    {
+      "epoch": 0.5790668348045397,
+      "grad_norm": 0.330078125,
+      "learning_rate": 2.0677082006902673e-06,
+      "loss": 2.931,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 287
+    },
+    {
+      "epoch": 0.5810844892812106,
+      "grad_norm": 0.349609375,
+      "learning_rate": 2.0512591021577775e-06,
+      "loss": 2.8567,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 288
+    },
+    {
+      "epoch": 0.5831021437578815,
+      "grad_norm": 0.345703125,
+      "learning_rate": 2.034830052872235e-06,
+      "loss": 2.7867,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 289
+    },
+    {
+      "epoch": 0.5851197982345523,
+      "grad_norm": 0.34765625,
+      "learning_rate": 2.018421786865387e-06,
+      "loss": 2.9489,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 290
+    },
+    {
+      "epoch": 0.5871374527112232,
+      "grad_norm": 0.34765625,
+      "learning_rate": 2.0020350372404104e-06,
+      "loss": 2.8389,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 291
+    },
+    {
+      "epoch": 0.5891551071878941,
+      "grad_norm": 0.33203125,
+      "learning_rate": 1.985670536139151e-06,
+      "loss": 2.9196,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 292
+    },
+    {
+      "epoch": 0.5911727616645649,
+      "grad_norm": 0.32421875,
+      "learning_rate": 1.9693290147094184e-06,
+      "loss": 2.9489,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 293
+    },
+    {
+      "epoch": 0.5931904161412358,
+      "grad_norm": 0.392578125,
+      "learning_rate": 1.9530112030723123e-06,
+      "loss": 2.958,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 294
+    },
+    {
+      "epoch": 0.5952080706179067,
+      "grad_norm": 0.34765625,
+      "learning_rate": 1.9367178302896087e-06,
+      "loss": 2.7799,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 295
+    },
+    {
+      "epoch": 0.5972257250945776,
+      "grad_norm": 0.376953125,
+      "learning_rate": 1.920449624331179e-06,
+      "loss": 2.9718,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 296
+    },
+    {
+      "epoch": 0.5992433795712484,
+      "grad_norm": 0.337890625,
+      "learning_rate": 1.9042073120424727e-06,
+      "loss": 2.9414,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 297
+    },
+    {
+      "epoch": 0.6012610340479193,
+      "grad_norm": 0.33203125,
+      "learning_rate": 1.887991619112035e-06,
+      "loss": 2.9257,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 298
+    },
+    {
+      "epoch": 0.6032786885245902,
+      "grad_norm": 0.349609375,
+      "learning_rate": 1.8718032700390887e-06,
+      "loss": 2.9388,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 299
+    },
+    {
+      "epoch": 0.605296343001261,
+      "grad_norm": 0.353515625,
+      "learning_rate": 1.8556429881011655e-06,
+      "loss": 2.8914,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 300
+    },
+    {
+      "epoch": 0.6073139974779319,
+      "grad_norm": 0.37109375,
+      "learning_rate": 1.8395114953217853e-06,
+      "loss": 2.8707,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 301
+    },
+    {
+      "epoch": 0.6093316519546028,
+      "grad_norm": 0.310546875,
+      "learning_rate": 1.8234095124382031e-06,
+      "loss": 2.8361,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 302
+    },
+    {
+      "epoch": 0.6113493064312736,
+      "grad_norm": 0.357421875,
+      "learning_rate": 1.8073377588692026e-06,
+      "loss": 2.8958,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 303
+    },
+    {
+      "epoch": 0.6133669609079445,
+      "grad_norm": 0.34375,
+      "learning_rate": 1.791296952682956e-06,
+      "loss": 2.9175,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 304
+    },
+    {
+      "epoch": 0.6153846153846154,
+      "grad_norm": 0.333984375,
+      "learning_rate": 1.775287810564939e-06,
+      "loss": 2.8774,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 305
+    },
+    {
+      "epoch": 0.6174022698612862,
+      "grad_norm": 0.326171875,
+      "learning_rate": 1.7593110477859155e-06,
+      "loss": 2.9021,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 306
+    },
+    {
+      "epoch": 0.6194199243379571,
+      "grad_norm": 0.349609375,
+      "learning_rate": 1.7433673781699717e-06,
+      "loss": 2.919,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 307
+    },
+    {
+      "epoch": 0.621437578814628,
+      "grad_norm": 0.357421875,
+      "learning_rate": 1.7274575140626318e-06,
+      "loss": 2.9148,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 308
+    },
+    {
+      "epoch": 0.6234552332912988,
+      "grad_norm": 0.35546875,
+      "learning_rate": 1.7115821662990246e-06,
+      "loss": 2.8786,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 309
+    },
+    {
+      "epoch": 0.6254728877679697,
+      "grad_norm": 0.361328125,
+      "learning_rate": 1.6957420441721285e-06,
+      "loss": 2.9553,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 310
+    },
+    {
+      "epoch": 0.6274905422446406,
+      "grad_norm": 0.328125,
+      "learning_rate": 1.6799378554010773e-06,
+      "loss": 2.8745,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 311
+    },
+    {
+      "epoch": 0.6295081967213115,
+      "grad_norm": 0.400390625,
+      "learning_rate": 1.6641703060995456e-06,
+      "loss": 2.8396,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 312
+    },
+    {
+      "epoch": 0.6315258511979823,
+      "grad_norm": 0.33984375,
+      "learning_rate": 1.6484401007441938e-06,
+      "loss": 2.8615,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 313
+    },
+    {
+      "epoch": 0.6335435056746532,
+      "grad_norm": 0.349609375,
+      "learning_rate": 1.6327479421431984e-06,
+      "loss": 2.8957,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 314
+    },
+    {
+      "epoch": 0.6355611601513241,
+      "grad_norm": 0.35546875,
+      "learning_rate": 1.6170945314048476e-06,
+      "loss": 2.9217,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 315
+    },
+    {
+      "epoch": 0.6375788146279949,
+      "grad_norm": 0.34765625,
+      "learning_rate": 1.6014805679062185e-06,
+      "loss": 2.8061,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 316
+    },
+    {
+      "epoch": 0.6395964691046658,
+      "grad_norm": 0.31640625,
+      "learning_rate": 1.5859067492619284e-06,
+      "loss": 2.7589,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 317
+    },
+    {
+      "epoch": 0.6416141235813367,
+      "grad_norm": 0.306640625,
+      "learning_rate": 1.5703737712929674e-06,
+      "loss": 2.8038,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 318
+    },
+    {
+      "epoch": 0.6436317780580075,
+      "grad_norm": 0.345703125,
+      "learning_rate": 1.5548823279956104e-06,
+      "loss": 2.9054,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 319
+    },
+    {
+      "epoch": 0.6456494325346784,
+      "grad_norm": 0.37109375,
+      "learning_rate": 1.5394331115104074e-06,
+      "loss": 3.0088,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 320
+    },
+    {
+      "epoch": 0.6476670870113493,
+      "grad_norm": 0.33984375,
+      "learning_rate": 1.5240268120912631e-06,
+      "loss": 2.9129,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 321
+    },
+    {
+      "epoch": 0.6496847414880201,
+      "grad_norm": 0.34375,
+      "learning_rate": 1.5086641180745934e-06,
+      "loss": 2.9135,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 322
+    },
+    {
+      "epoch": 0.651702395964691,
+      "grad_norm": 0.33203125,
+      "learning_rate": 1.493345715848574e-06,
+      "loss": 2.9974,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 323
+    },
+    {
+      "epoch": 0.6537200504413619,
+      "grad_norm": 0.3359375,
+      "learning_rate": 1.478072289822471e-06,
+      "loss": 2.8818,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 324
+    },
+    {
+      "epoch": 0.6557377049180327,
+      "grad_norm": 0.34375,
+      "learning_rate": 1.462844522396066e-06,
+      "loss": 2.9075,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 325
+    },
+    {
+      "epoch": 0.6577553593947036,
+      "grad_norm": 0.33984375,
+      "learning_rate": 1.4476630939291631e-06,
+      "loss": 2.8859,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 326
+    },
+    {
+      "epoch": 0.6597730138713745,
+      "grad_norm": 0.32421875,
+      "learning_rate": 1.4325286827111931e-06,
+      "loss": 2.8487,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 327
+    },
+    {
+      "epoch": 0.6617906683480455,
+      "grad_norm": 0.328125,
+      "learning_rate": 1.417441964930909e-06,
+      "loss": 2.9071,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 328
+    },
+    {
+      "epoch": 0.6638083228247162,
+      "grad_norm": 0.328125,
+      "learning_rate": 1.4024036146461734e-06,
+      "loss": 2.8911,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 329
+    },
+    {
+      "epoch": 0.6658259773013872,
+      "grad_norm": 0.3671875,
+      "learning_rate": 1.3874143037538417e-06,
+      "loss": 2.8569,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 330
+    },
+    {
+      "epoch": 0.6678436317780581,
+      "grad_norm": 0.3359375,
+      "learning_rate": 1.372474701959745e-06,
+      "loss": 2.9294,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 331
+    },
+    {
+      "epoch": 0.6698612862547288,
+      "grad_norm": 0.341796875,
+      "learning_rate": 1.357585476748766e-06,
+      "loss": 2.9338,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 332
+    },
+    {
+      "epoch": 0.6718789407313998,
+      "grad_norm": 0.318359375,
+      "learning_rate": 1.342747293355019e-06,
+      "loss": 2.7899,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 333
+    },
+    {
+      "epoch": 0.6738965952080707,
+      "grad_norm": 0.34765625,
+      "learning_rate": 1.3279608147321223e-06,
+      "loss": 2.8819,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 334
+    },
+    {
+      "epoch": 0.6759142496847415,
+      "grad_norm": 0.369140625,
+      "learning_rate": 1.3132267015235862e-06,
+      "loss": 2.9289,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 335
+    },
+    {
+      "epoch": 0.6779319041614124,
+      "grad_norm": 0.349609375,
+      "learning_rate": 1.2985456120332907e-06,
+      "loss": 2.8636,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 336
+    },
+    {
+      "epoch": 0.6799495586380833,
+      "grad_norm": 0.30078125,
+      "learning_rate": 1.2839182021960753e-06,
+      "loss": 2.9377,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 337
+    },
+    {
+      "epoch": 0.6819672131147541,
+      "grad_norm": 0.36328125,
+      "learning_rate": 1.2693451255484314e-06,
+      "loss": 2.884,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 338
+    },
+    {
+      "epoch": 0.683984867591425,
+      "grad_norm": 0.333984375,
+      "learning_rate": 1.2548270331993034e-06,
+      "loss": 2.87,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 339
+    },
+    {
+      "epoch": 0.6860025220680959,
+      "grad_norm": 0.306640625,
+      "learning_rate": 1.2403645738009998e-06,
+      "loss": 2.9084,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 340
+    },
+    {
+      "epoch": 0.6880201765447667,
+      "grad_norm": 0.333984375,
+      "learning_rate": 1.2259583935202063e-06,
+      "loss": 2.9395,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 341
+    },
+    {
+      "epoch": 0.6900378310214376,
+      "grad_norm": 0.34375,
+      "learning_rate": 1.2116091360091262e-06,
+      "loss": 2.9742,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 342
+    },
+    {
+      "epoch": 0.6920554854981085,
+      "grad_norm": 0.3515625,
+      "learning_rate": 1.1973174423767098e-06,
+      "loss": 2.9324,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 343
+    },
+    {
+      "epoch": 0.6940731399747794,
+      "grad_norm": 0.326171875,
+      "learning_rate": 1.1830839511600211e-06,
+      "loss": 2.8262,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 344
+    },
+    {
+      "epoch": 0.6960907944514502,
+      "grad_norm": 0.365234375,
+      "learning_rate": 1.168909298295704e-06,
+      "loss": 2.8697,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 345
+    },
+    {
+      "epoch": 0.6981084489281211,
+      "grad_norm": 0.419921875,
+      "learning_rate": 1.1547941170915686e-06,
+      "loss": 2.922,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 346
+    },
+    {
+      "epoch": 0.700126103404792,
+      "grad_norm": 0.376953125,
+      "learning_rate": 1.140739038198298e-06,
+      "loss": 2.8969,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 347
+    },
+    {
+      "epoch": 0.7021437578814628,
+      "grad_norm": 0.333984375,
+      "learning_rate": 1.1267446895812704e-06,
+      "loss": 2.8233,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 348
+    },
+    {
+      "epoch": 0.7041614123581337,
+      "grad_norm": 0.330078125,
+      "learning_rate": 1.1128116964925023e-06,
+      "loss": 2.9209,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 349
+    },
+    {
+      "epoch": 0.7061790668348046,
+      "grad_norm": 0.306640625,
+      "learning_rate": 1.098940681442713e-06,
+      "loss": 2.855,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 350
+    },
+    {
+      "epoch": 0.7081967213114754,
+      "grad_norm": 0.341796875,
+      "learning_rate": 1.0851322641735119e-06,
+      "loss": 2.8907,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 351
+    },
+    {
+      "epoch": 0.7102143757881463,
+      "grad_norm": 0.333984375,
+      "learning_rate": 1.0713870616297093e-06,
+      "loss": 2.7215,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 352
+    },
+    {
+      "epoch": 0.7122320302648172,
+      "grad_norm": 0.341796875,
+      "learning_rate": 1.0577056879317486e-06,
+      "loss": 2.8413,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 353
+    },
+    {
+      "epoch": 0.714249684741488,
+      "grad_norm": 0.3515625,
+      "learning_rate": 1.0440887543482747e-06,
+      "loss": 2.8578,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 354
+    },
+    {
+      "epoch": 0.7162673392181589,
+      "grad_norm": 0.3359375,
+      "learning_rate": 1.0305368692688175e-06,
+      "loss": 2.9202,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 355
+    },
+    {
+      "epoch": 0.7182849936948298,
+      "grad_norm": 0.412109375,
+      "learning_rate": 1.0170506381766121e-06,
+      "loss": 2.962,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 356
+    },
+    {
+      "epoch": 0.7203026481715006,
+      "grad_norm": 0.330078125,
+      "learning_rate": 1.0036306636215462e-06,
+      "loss": 2.9256,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 357
+    },
+    {
+      "epoch": 0.7223203026481715,
+      "grad_norm": 0.302734375,
+      "learning_rate": 9.902775451932387e-07,
+      "loss": 2.9143,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 358
+    },
+    {
+      "epoch": 0.7243379571248424,
+      "grad_norm": 0.322265625,
+      "learning_rate": 9.769918794942511e-07,
+      "loss": 2.8691,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 359
+    },
+    {
+      "epoch": 0.7263556116015133,
+      "grad_norm": 0.298828125,
+      "learning_rate": 9.637742601134287e-07,
+      "loss": 2.8137,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 360
+    },
+    {
+      "epoch": 0.7283732660781841,
+      "grad_norm": 0.337890625,
+      "learning_rate": 9.506252775993882e-07,
+      "loss": 2.8396,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 361
+    },
+    {
+      "epoch": 0.730390920554855,
+      "grad_norm": 0.3203125,
+      "learning_rate": 9.375455194341215e-07,
+      "loss": 2.8566,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 362
+    },
+    {
+      "epoch": 0.7324085750315259,
+      "grad_norm": 0.318359375,
+      "learning_rate": 9.24535570006756e-07,
+      "loss": 2.8701,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 363
+    },
+    {
+      "epoch": 0.7344262295081967,
+      "grad_norm": 0.330078125,
+      "learning_rate": 9.115960105874411e-07,
+      "loss": 2.8288,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 364
+    },
+    {
+      "epoch": 0.7364438839848676,
+      "grad_norm": 0.287109375,
+      "learning_rate": 8.987274193013792e-07,
+      "loss": 2.8787,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 365
+    },
+    {
+      "epoch": 0.7384615384615385,
+      "grad_norm": 0.34765625,
+      "learning_rate": 8.85930371102994e-07,
+      "loss": 2.7868,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 366
+    },
+    {
+      "epoch": 0.7404791929382093,
+      "grad_norm": 0.32421875,
+      "learning_rate": 8.732054377502442e-07,
+      "loss": 2.9876,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 367
+    },
+    {
+      "epoch": 0.7424968474148802,
+      "grad_norm": 0.33984375,
+      "learning_rate": 8.605531877790762e-07,
+      "loss": 2.9011,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 368
+    },
+    {
+      "epoch": 0.7445145018915511,
+      "grad_norm": 0.36328125,
+      "learning_rate": 8.479741864780236e-07,
+      "loss": 2.8392,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 369
+    },
+    {
+      "epoch": 0.7465321563682219,
+      "grad_norm": 0.310546875,
+      "learning_rate": 8.354689958629514e-07,
+      "loss": 2.8629,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 370
+    },
+    {
+      "epoch": 0.7485498108448928,
+      "grad_norm": 0.3046875,
+      "learning_rate": 8.23038174651942e-07,
+      "loss": 2.9016,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 371
+    },
+    {
+      "epoch": 0.7505674653215637,
+      "grad_norm": 0.328125,
+      "learning_rate": 8.106822782403376e-07,
+      "loss": 2.9677,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 372
+    },
+    {
+      "epoch": 0.7525851197982345,
+      "grad_norm": 0.373046875,
+      "learning_rate": 7.984018586759227e-07,
+      "loss": 2.8286,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 373
+    },
+    {
+      "epoch": 0.7546027742749054,
+      "grad_norm": 0.326171875,
+      "learning_rate": 7.861974646342596e-07,
+      "loss": 2.826,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 374
+    },
+    {
+      "epoch": 0.7566204287515763,
+      "grad_norm": 0.333984375,
+      "learning_rate": 7.740696413941745e-07,
+      "loss": 2.8649,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 375
+    },
+    {
+      "epoch": 0.7586380832282472,
+      "grad_norm": 0.33984375,
+      "learning_rate": 7.620189308133943e-07,
+      "loss": 2.8342,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 376
+    },
+    {
+      "epoch": 0.760655737704918,
+      "grad_norm": 0.330078125,
+      "learning_rate": 7.500458713043385e-07,
+      "loss": 2.8452,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 377
+    },
+    {
+      "epoch": 0.7626733921815889,
+      "grad_norm": 0.345703125,
+      "learning_rate": 7.381509978100626e-07,
+      "loss": 2.7791,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 378
+    },
+    {
+      "epoch": 0.7646910466582598,
+      "grad_norm": 0.306640625,
+      "learning_rate": 7.263348417803545e-07,
+      "loss": 2.8265,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 379
+    },
+    {
+      "epoch": 0.7667087011349306,
+      "grad_norm": 0.314453125,
+      "learning_rate": 7.145979311479986e-07,
+      "loss": 2.9179,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 380
+    },
+    {
+      "epoch": 0.7687263556116015,
+      "grad_norm": 0.3046875,
+      "learning_rate": 7.029407903051771e-07,
+      "loss": 2.9605,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 381
+    },
+    {
+      "epoch": 0.7707440100882724,
+      "grad_norm": 0.306640625,
+      "learning_rate": 6.91363940080049e-07,
+      "loss": 2.8854,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 382
+    },
+    {
+      "epoch": 0.7727616645649432,
+      "grad_norm": 0.32421875,
+      "learning_rate": 6.798678977134768e-07,
+      "loss": 2.8363,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 383
+    },
+    {
+      "epoch": 0.7747793190416141,
+      "grad_norm": 0.318359375,
+      "learning_rate": 6.684531768359173e-07,
+      "loss": 2.9301,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 384
+    },
+    {
+      "epoch": 0.776796973518285,
+      "grad_norm": 0.345703125,
+      "learning_rate": 6.57120287444473e-07,
+      "loss": 2.8542,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 385
+    },
+    {
+      "epoch": 0.7788146279949558,
+      "grad_norm": 0.345703125,
+      "learning_rate": 6.458697358801061e-07,
+      "loss": 2.8322,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 386
+    },
+    {
+      "epoch": 0.7808322824716267,
+      "grad_norm": 0.314453125,
+      "learning_rate": 6.34702024805016e-07,
+      "loss": 2.8331,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 387
+    },
+    {
+      "epoch": 0.7828499369482976,
+      "grad_norm": 0.314453125,
+      "learning_rate": 6.236176531801813e-07,
+      "loss": 2.8607,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 388
+    },
+    {
+      "epoch": 0.7848675914249684,
+      "grad_norm": 0.345703125,
+      "learning_rate": 6.126171162430636e-07,
+      "loss": 2.9638,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 389
+    },
+    {
+      "epoch": 0.7868852459016393,
+      "grad_norm": 0.33984375,
+      "learning_rate": 6.017009054854858e-07,
+      "loss": 2.9402,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 390
+    },
+    {
+      "epoch": 0.7889029003783102,
+      "grad_norm": 0.3125,
+      "learning_rate": 5.908695086316701e-07,
+      "loss": 2.8842,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 391
+    },
+    {
+      "epoch": 0.7909205548549811,
+      "grad_norm": 0.427734375,
+      "learning_rate": 5.801234096164468e-07,
+      "loss": 2.87,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 392
+    },
+    {
+      "epoch": 0.7929382093316519,
+      "grad_norm": 0.34765625,
+      "learning_rate": 5.694630885636332e-07,
+      "loss": 2.8928,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 393
+    },
+    {
+      "epoch": 0.7949558638083228,
+      "grad_norm": 0.322265625,
+      "learning_rate": 5.588890217645821e-07,
+      "loss": 2.7904,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 394
+    },
+    {
+      "epoch": 0.7969735182849937,
+      "grad_norm": 0.341796875,
+      "learning_rate": 5.484016816569015e-07,
+      "loss": 2.9444,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 395
+    },
+    {
+      "epoch": 0.7989911727616645,
+      "grad_norm": 0.3203125,
+      "learning_rate": 5.380015368033476e-07,
+      "loss": 2.9772,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 396
+    },
+    {
+      "epoch": 0.8010088272383354,
+      "grad_norm": 0.34375,
+      "learning_rate": 5.276890518708885e-07,
+      "loss": 2.7561,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 397
+    },
+    {
+      "epoch": 0.8030264817150063,
+      "grad_norm": 0.32421875,
+      "learning_rate": 5.174646876099421e-07,
+      "loss": 2.8828,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 398
+    },
+    {
+      "epoch": 0.8050441361916771,
+      "grad_norm": 0.302734375,
+      "learning_rate": 5.073289008337967e-07,
+      "loss": 2.8515,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 399
+    },
+    {
+      "epoch": 0.807061790668348,
+      "grad_norm": 0.32421875,
+      "learning_rate": 4.972821443981921e-07,
+      "loss": 2.8422,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 400
+    },
+    {
+      "epoch": 0.809079445145019,
+      "grad_norm": 0.31640625,
+      "learning_rate": 4.873248671810929e-07,
+      "loss": 2.9368,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 401
+    },
+    {
+      "epoch": 0.8110970996216897,
+      "grad_norm": 0.373046875,
+      "learning_rate": 4.774575140626317e-07,
+      "loss": 2.955,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 402
+    },
+    {
+      "epoch": 0.8131147540983606,
+      "grad_norm": 0.302734375,
+      "learning_rate": 4.6768052590523053e-07,
+      "loss": 2.9267,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 403
+    },
+    {
+      "epoch": 0.8151324085750316,
+      "grad_norm": 0.328125,
+      "learning_rate": 4.579943395339062e-07,
+      "loss": 2.8792,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 404
+    },
+    {
+      "epoch": 0.8171500630517023,
+      "grad_norm": 0.328125,
+      "learning_rate": 4.4839938771675115e-07,
+      "loss": 2.9025,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 405
+    },
+    {
+      "epoch": 0.8191677175283733,
+      "grad_norm": 0.328125,
+      "learning_rate": 4.388960991455998e-07,
+      "loss": 2.9083,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 406
+    },
+    {
+      "epoch": 0.8211853720050442,
+      "grad_norm": 0.326171875,
+      "learning_rate": 4.294848984168723e-07,
+      "loss": 2.9084,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 407
+    },
+    {
+      "epoch": 0.8232030264817151,
+      "grad_norm": 0.310546875,
+      "learning_rate": 4.20166206012608e-07,
+      "loss": 2.8086,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 408
+    },
+    {
+      "epoch": 0.8252206809583859,
+      "grad_norm": 0.3359375,
+      "learning_rate": 4.109404382816756e-07,
+      "loss": 2.8263,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 409
+    },
+    {
+      "epoch": 0.8272383354350568,
+      "grad_norm": 0.326171875,
+      "learning_rate": 4.0180800742117246e-07,
+      "loss": 2.9327,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 410
+    },
+    {
+      "epoch": 0.8292559899117277,
+      "grad_norm": 0.333984375,
+      "learning_rate": 3.927693214580075e-07,
+      "loss": 2.8724,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 411
+    },
+    {
+      "epoch": 0.8312736443883985,
+      "grad_norm": 0.3359375,
+      "learning_rate": 3.8382478423067163e-07,
+      "loss": 2.8089,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 412
+    },
+    {
+      "epoch": 0.8332912988650694,
+      "grad_norm": 0.35546875,
+      "learning_rate": 3.7497479537119435e-07,
+      "loss": 2.9261,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 413
+    },
+    {
+      "epoch": 0.8353089533417403,
+      "grad_norm": 0.34765625,
+      "learning_rate": 3.662197502872886e-07,
+      "loss": 2.9533,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 414
+    },
+    {
+      "epoch": 0.8373266078184111,
+      "grad_norm": 0.3359375,
+      "learning_rate": 3.575600401446841e-07,
+      "loss": 2.9823,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 415
+    },
+    {
+      "epoch": 0.839344262295082,
+      "grad_norm": 0.349609375,
+      "learning_rate": 3.489960518496521e-07,
+      "loss": 2.9184,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 416
+    },
+    {
+      "epoch": 0.8413619167717529,
+      "grad_norm": 0.357421875,
+      "learning_rate": 3.405281680317149e-07,
+      "loss": 2.8453,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 417
+    },
+    {
+      "epoch": 0.8433795712484237,
+      "grad_norm": 0.314453125,
+      "learning_rate": 3.3215676702655687e-07,
+      "loss": 2.8601,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 418
+    },
+    {
+      "epoch": 0.8453972257250946,
+      "grad_norm": 0.32421875,
+      "learning_rate": 3.2388222285911373e-07,
+      "loss": 2.9082,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 419
+    },
+    {
+      "epoch": 0.8474148802017655,
+      "grad_norm": 0.322265625,
+      "learning_rate": 3.1570490522686624e-07,
+      "loss": 2.8188,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 420
+    },
+    {
+      "epoch": 0.8494325346784363,
+      "grad_norm": 0.3515625,
+      "learning_rate": 3.076251794833213e-07,
+      "loss": 2.8224,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 421
+    },
+    {
+      "epoch": 0.8514501891551072,
+      "grad_norm": 0.32421875,
+      "learning_rate": 2.9964340662168774e-07,
+      "loss": 2.8893,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 422
+    },
+    {
+      "epoch": 0.8534678436317781,
+      "grad_norm": 0.3203125,
+      "learning_rate": 2.9175994325874783e-07,
+      "loss": 2.8318,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 423
+    },
+    {
+      "epoch": 0.855485498108449,
+      "grad_norm": 0.31640625,
+      "learning_rate": 2.8397514161892484e-07,
+      "loss": 2.8396,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 424
+    },
+    {
+      "epoch": 0.8575031525851198,
+      "grad_norm": 0.31640625,
+      "learning_rate": 2.7628934951854506e-07,
+      "loss": 2.7336,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 425
+    },
+    {
+      "epoch": 0.8595208070617907,
+      "grad_norm": 0.326171875,
+      "learning_rate": 2.6870291035029724e-07,
+      "loss": 2.8688,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 426
+    },
+    {
+      "epoch": 0.8615384615384616,
+      "grad_norm": 0.30859375,
+      "learning_rate": 2.612161630678922e-07,
+      "loss": 2.7664,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 427
+    },
+    {
+      "epoch": 0.8635561160151324,
+      "grad_norm": 0.3515625,
+      "learning_rate": 2.5382944217091725e-07,
+      "loss": 2.8153,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 428
+    },
+    {
+      "epoch": 0.8655737704918033,
+      "grad_norm": 0.33984375,
+      "learning_rate": 2.465430776898911e-07,
+      "loss": 2.818,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 429
+    },
+    {
+      "epoch": 0.8675914249684742,
+      "grad_norm": 0.29296875,
+      "learning_rate": 2.3935739517151916e-07,
+      "loss": 2.9183,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 430
+    },
+    {
+      "epoch": 0.869609079445145,
+      "grad_norm": 0.359375,
+      "learning_rate": 2.3227271566414827e-07,
+      "loss": 2.874,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 431
+    },
+    {
+      "epoch": 0.8716267339218159,
+      "grad_norm": 0.3359375,
+      "learning_rate": 2.2528935570342165e-07,
+      "loss": 2.9182,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 432
+    },
+    {
+      "epoch": 0.8736443883984868,
+      "grad_norm": 0.341796875,
+      "learning_rate": 2.1840762729813808e-07,
+      "loss": 2.8589,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 433
+    },
+    {
+      "epoch": 0.8756620428751576,
+      "grad_norm": 0.31640625,
+      "learning_rate": 2.116278379163106e-07,
+      "loss": 2.938,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 434
+    },
+    {
+      "epoch": 0.8776796973518285,
+      "grad_norm": 0.353515625,
+      "learning_rate": 2.0495029047142983e-07,
+      "loss": 2.9281,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 435
+    },
+    {
+      "epoch": 0.8796973518284994,
+      "grad_norm": 0.330078125,
+      "learning_rate": 1.9837528330892781e-07,
+      "loss": 2.9188,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 436
+    },
+    {
+      "epoch": 0.8817150063051702,
+      "grad_norm": 0.32421875,
+      "learning_rate": 1.9190311019285368e-07,
+      "loss": 2.9296,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 437
+    },
+    {
+      "epoch": 0.8837326607818411,
+      "grad_norm": 0.31640625,
+      "learning_rate": 1.855340602927419e-07,
+      "loss": 2.8826,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 438
+    },
+    {
+      "epoch": 0.885750315258512,
+      "grad_norm": 0.359375,
+      "learning_rate": 1.7926841817069717e-07,
+      "loss": 2.8862,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 439
+    },
+    {
+      "epoch": 0.8877679697351829,
+      "grad_norm": 0.37890625,
+      "learning_rate": 1.7310646376867885e-07,
+      "loss": 2.9719,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 440
+    },
+    {
+      "epoch": 0.8897856242118537,
+      "grad_norm": 0.326171875,
+      "learning_rate": 1.6704847239599364e-07,
+      "loss": 2.8783,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 441
+    },
+    {
+      "epoch": 0.8918032786885246,
+      "grad_norm": 0.36328125,
+      "learning_rate": 1.6109471471699557e-07,
+      "loss": 2.9328,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 442
+    },
+    {
+      "epoch": 0.8938209331651955,
+      "grad_norm": 0.314453125,
+      "learning_rate": 1.5524545673899106e-07,
+      "loss": 2.8429,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 443
+    },
+    {
+      "epoch": 0.8958385876418663,
+      "grad_norm": 0.3515625,
+      "learning_rate": 1.4950095980035772e-07,
+      "loss": 2.9344,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 444
+    },
+    {
+      "epoch": 0.8978562421185372,
+      "grad_norm": 0.330078125,
+      "learning_rate": 1.438614805588634e-07,
+      "loss": 2.8765,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 445
+    },
+    {
+      "epoch": 0.8998738965952081,
+      "grad_norm": 0.37109375,
+      "learning_rate": 1.3832727098020333e-07,
+      "loss": 2.7767,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 446
+    },
+    {
+      "epoch": 0.9018915510718789,
+      "grad_norm": 0.330078125,
+      "learning_rate": 1.3289857832673947e-07,
+      "loss": 2.9776,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 447
+    },
+    {
+      "epoch": 0.9039092055485498,
+      "grad_norm": 0.404296875,
+      "learning_rate": 1.2757564514645492e-07,
+      "loss": 2.8624,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 448
+    },
+    {
+      "epoch": 0.9059268600252207,
+      "grad_norm": 0.330078125,
+      "learning_rate": 1.223587092621162e-07,
+      "loss": 2.993,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 449
+    },
+    {
+      "epoch": 0.9079445145018915,
+      "grad_norm": 0.31640625,
+      "learning_rate": 1.1724800376064799e-07,
+      "loss": 2.8225,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 450
+    },
+    {
+      "epoch": 0.9099621689785624,
+      "grad_norm": 0.353515625,
+      "learning_rate": 1.1224375698271894e-07,
+      "loss": 2.928,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 451
+    },
+    {
+      "epoch": 0.9119798234552333,
+      "grad_norm": 0.32421875,
+      "learning_rate": 1.0734619251253963e-07,
+      "loss": 2.8997,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 452
+    },
+    {
+      "epoch": 0.9139974779319041,
+      "grad_norm": 0.337890625,
+      "learning_rate": 1.0255552916787343e-07,
+      "loss": 2.9446,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 453
+    },
+    {
+      "epoch": 0.916015132408575,
+      "grad_norm": 0.326171875,
+      "learning_rate": 9.78719809902598e-08,
+      "loss": 2.9681,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 454
+    },
+    {
+      "epoch": 0.9180327868852459,
+      "grad_norm": 0.33984375,
+      "learning_rate": 9.329575723544925e-08,
+      "loss": 2.9225,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 455
+    },
+    {
+      "epoch": 0.9200504413619168,
+      "grad_norm": 0.314453125,
+      "learning_rate": 8.882706236405886e-08,
+      "loss": 2.885,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 456
+    },
+    {
+      "epoch": 0.9220680958385876,
+      "grad_norm": 0.296875,
+      "learning_rate": 8.446609603243117e-08,
+      "loss": 2.8799,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 457
+    },
+    {
+      "epoch": 0.9240857503152585,
+      "grad_norm": 0.328125,
+      "learning_rate": 8.021305308371891e-08,
+      "loss": 2.8873,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 458
+    },
+    {
+      "epoch": 0.9261034047919294,
+      "grad_norm": 0.333984375,
+      "learning_rate": 7.606812353917636e-08,
+      "loss": 2.8536,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 459
+    },
+    {
+      "epoch": 0.9281210592686002,
+      "grad_norm": 0.32421875,
+      "learning_rate": 7.203149258967035e-08,
+      "loss": 2.8754,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 460
+    },
+    {
+      "epoch": 0.9301387137452711,
+      "grad_norm": 0.341796875,
+      "learning_rate": 6.810334058740736e-08,
+      "loss": 2.804,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 461
+    },
+    {
+      "epoch": 0.932156368221942,
+      "grad_norm": 0.357421875,
+      "learning_rate": 6.428384303787282e-08,
+      "loss": 2.8765,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 462
+    },
+    {
+      "epoch": 0.9341740226986128,
+      "grad_norm": 0.3359375,
+      "learning_rate": 6.05731705919932e-08,
+      "loss": 2.8596,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 463
+    },
+    {
+      "epoch": 0.9361916771752837,
+      "grad_norm": 0.326171875,
+      "learning_rate": 5.697148903850869e-08,
+      "loss": 2.8129,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 464
+    },
+    {
+      "epoch": 0.9382093316519546,
+      "grad_norm": 0.34765625,
+      "learning_rate": 5.347895929656649e-08,
+      "loss": 2.8172,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 465
+    },
+    {
+      "epoch": 0.9402269861286254,
+      "grad_norm": 0.31640625,
+      "learning_rate": 5.009573740853313e-08,
+      "loss": 2.8026,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 466
+    },
+    {
+      "epoch": 0.9422446406052963,
+      "grad_norm": 0.376953125,
+      "learning_rate": 4.682197453301951e-08,
+      "loss": 2.8827,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 467
+    },
+    {
+      "epoch": 0.9442622950819672,
+      "grad_norm": 0.330078125,
+      "learning_rate": 4.365781693813048e-08,
+      "loss": 2.8079,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 468
+    },
+    {
+      "epoch": 0.946279949558638,
+      "grad_norm": 0.4609375,
+      "learning_rate": 4.060340599492646e-08,
+      "loss": 2.8384,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 469
+    },
+    {
+      "epoch": 0.9482976040353089,
+      "grad_norm": 0.330078125,
+      "learning_rate": 3.765887817111069e-08,
+      "loss": 2.8274,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 470
+    },
+    {
+      "epoch": 0.9503152585119798,
+      "grad_norm": 0.328125,
+      "learning_rate": 3.4824365024928585e-08,
+      "loss": 3.0094,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 471
+    },
+    {
+      "epoch": 0.9523329129886507,
+      "grad_norm": 0.326171875,
+      "learning_rate": 3.209999319929269e-08,
+      "loss": 2.8901,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 472
+    },
+    {
+      "epoch": 0.9543505674653215,
+      "grad_norm": 0.345703125,
+      "learning_rate": 2.9485884416122213e-08,
+      "loss": 2.9069,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 473
+    },
+    {
+      "epoch": 0.9563682219419924,
+      "grad_norm": 0.34375,
+      "learning_rate": 2.698215547090599e-08,
+      "loss": 2.8147,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 474
+    },
+    {
+      "epoch": 0.9583858764186634,
+      "grad_norm": 0.33984375,
+      "learning_rate": 2.458891822748444e-08,
+      "loss": 2.8549,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 475
+    },
+    {
+      "epoch": 0.9604035308953341,
+      "grad_norm": 0.349609375,
+      "learning_rate": 2.230627961304993e-08,
+      "loss": 2.7994,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 476
+    },
+    {
+      "epoch": 0.962421185372005,
+      "grad_norm": 0.310546875,
+      "learning_rate": 2.0134341613370633e-08,
+      "loss": 2.9219,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 477
+    },
+    {
+      "epoch": 0.964438839848676,
+      "grad_norm": 0.330078125,
+      "learning_rate": 1.8073201268234142e-08,
+      "loss": 2.8302,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 478
+    },
+    {
+      "epoch": 0.9664564943253467,
+      "grad_norm": 0.31640625,
+      "learning_rate": 1.612295066711095e-08,
+      "loss": 2.8299,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 479
+    },
+    {
+      "epoch": 0.9684741488020177,
+      "grad_norm": 0.37890625,
+      "learning_rate": 1.4283676945041348e-08,
+      "loss": 2.9729,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 480
+    },
+    {
+      "epoch": 0.9704918032786886,
+      "grad_norm": 0.314453125,
+      "learning_rate": 1.255546227873966e-08,
+      "loss": 2.8149,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 481
+    },
+    {
+      "epoch": 0.9725094577553594,
+      "grad_norm": 0.318359375,
+      "learning_rate": 1.0938383882926618e-08,
+      "loss": 2.848,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 482
+    },
+    {
+      "epoch": 0.9745271122320303,
+      "grad_norm": 0.310546875,
+      "learning_rate": 9.432514006875725e-09,
+      "loss": 2.8566,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 483
+    },
+    {
+      "epoch": 0.9765447667087012,
+      "grad_norm": 0.349609375,
+      "learning_rate": 8.037919931187243e-09,
+      "loss": 2.8315,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 484
+    },
+    {
+      "epoch": 0.978562421185372,
+      "grad_norm": 0.34765625,
+      "learning_rate": 6.754663964781971e-09,
+      "loss": 2.9492,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 485
+    },
+    {
+      "epoch": 0.9805800756620429,
+      "grad_norm": 0.326171875,
+      "learning_rate": 5.582803442117091e-09,
+      "loss": 2.7673,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 486
+    },
+    {
+      "epoch": 0.9825977301387138,
+      "grad_norm": 0.310546875,
+      "learning_rate": 4.522390720624603e-09,
+      "loss": 2.8186,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 487
+    },
+    {
+      "epoch": 0.9846153846153847,
+      "grad_norm": 0.31640625,
+      "learning_rate": 3.573473178371534e-09,
+      "loss": 2.8907,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 488
+    },
+    {
+      "epoch": 0.9866330390920555,
+      "grad_norm": 0.349609375,
+      "learning_rate": 2.736093211944679e-09,
+      "loss": 2.8521,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 489
+    },
+    {
+      "epoch": 0.9886506935687264,
+      "grad_norm": 0.3671875,
+      "learning_rate": 2.0102882345540696e-09,
+      "loss": 2.9154,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 490
+    },
+    {
+      "epoch": 0.9906683480453973,
+      "grad_norm": 0.333984375,
+      "learning_rate": 1.3960906743634706e-09,
+      "loss": 2.9166,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 491
+    },
+    {
+      "epoch": 0.9926860025220681,
+      "grad_norm": 0.349609375,
+      "learning_rate": 8.935279730407087e-10,
+      "loss": 2.8178,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 492
+    },
+    {
+      "epoch": 0.994703656998739,
+      "grad_norm": 0.3203125,
+      "learning_rate": 5.026225845308763e-10,
+      "loss": 2.8598,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 493
+    },
+    {
+      "epoch": 0.9967213114754099,
+      "grad_norm": 0.310546875,
+      "learning_rate": 2.2339197405490952e-10,
+      "loss": 2.9274,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 494
+    },
+    {
+      "epoch": 0.9987389659520807,
+      "grad_norm": 0.34765625,
+      "learning_rate": 5.5848617327436404e-11,
+      "loss": 3.005,
+      "memory/device_mem_reserved(gib)": 50.82,
+      "memory/max_mem_active(gib)": 45.14,
+      "memory/max_mem_allocated(gib)": 45.14,
+      "step": 495
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 495,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 248,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 6.453534295012147e+18,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}