| bf16: true | |
| cutoff_len: 512 | |
| dataset: STE | |
| dataset_dir: data | |
| ddp_timeout: 180000000 | |
| do_train: true | |
| finetuning_type: lora | |
| flash_attn: auto | |
| gradient_accumulation_steps: 8 | |
| include_num_input_tokens_seen: true | |
| learning_rate: 0.0002 | |
| logging_steps: 5 | |
| lora_alpha: 16 | |
| lora_dropout: 0 | |
| lora_rank: 8 | |
| lora_target: all | |
| lr_scheduler_type: cosine | |
| max_grad_norm: 0.3 | |
| max_samples: 10000 | |
| model_name_or_path: meta-llama/Llama-3.2-3B-Instruct | |
| num_train_epochs: 1.0 | |
| optim: adamw_torch | |
| output_dir: saves/Llama-3.2-3B-Instruct/lora/train_2025-04-17-02-04-38 | |
| packing: false | |
| per_device_train_batch_size: 16 | |
| plot_loss: true | |
| preprocessing_num_workers: 16 | |
| report_to: none | |
| save_steps: 100 | |
| stage: sft | |
| template: alpaca | |
| trust_remote_code: true | |
| warmup_steps: 0 | |