| data: | |
| dataset_name: manual_plus_executed_prompt_variance | |
| from_langfuse: true | |
| method: sft | |
| num_proc: 2 | |
| seed: 3407 | |
| split: train | |
| test_size: 0.2 | |
| logging: | |
| file: finetune.log | |
| model: | |
| adapter: null | |
| dtype: null | |
| load_in_4bit: true | |
| lora: | |
| alpha: 16 | |
| bias: none | |
| dropout: 0 | |
| gradient_checkpointing: unsloth | |
| loftq_config: null | |
| r: 16 | |
| random_state: 3407 | |
| target_modules: | |
| - q_proj | |
| - k_proj | |
| - v_proj | |
| - o_proj | |
| - gate_proj | |
| - up_proj | |
| - down_proj | |
| use_rslora: false | |
| max_seq_length: 2048 | |
| name: Qwen/Qwen2.5-7B-Instruct | |
| save: | |
| gguf_quantization: | |
| - q4_k_m | |
| - q8_0 | |
| - q5_k_m | |
| hub_repo: joaormedeiros/qwen-2.5-lite-recursive-v5-rules-iter-1 | |
| local_dir: lora_model | |
| private: true | |
| push_to_hub: true | |
| save_mode: all_training_files | |
| test_inference: | |
| enabled: true | |
| input: 1, 1, 2, 3, 5, 8 | |
| instruction: Continue the fibonnaci sequence. | |
| training: | |
| batch_size: | |
| eval: 2 | |
| train: 4 | |
| eval_steps: 100 | |
| gradient_accumulation_steps: 4 | |
| learning_rate: 0.0002 | |
| logging_steps: 1 | |
| lr_scheduler: linear | |
| num_epochs: 5 | |
| optimizer: adamw_8bit | |
| output_dir: outputs | |
| report_to: tensorboard | |
| seed: 3407 | |
| warmup_steps: 1 | |
| weight_decay: 0.01 | |