| bf16: true | |
| cutoff_len: 128 | |
| # dataset: HNO1_train_wo_reasoning | |
| dataset: HNO1_train | |
| # dataset: HNO1_train_fake_reasoning | |
| # eval_dataset: | |
| dataset_dir: /workspace/LLaMA-Factory/data | |
| ddp_timeout: 180000000 | |
| # deepspeed: /workspace/LLaMA-Factory/examples/deepspeed/ds_z3_config.json | |
| do_train: true | |
| do_eval: false | |
| enable_thinking: false | |
| # eval_steps: 100 | |
| # eval_strategy: steps | |
| finetuning_type: lora | |
| lora_alpha: 16 | |
| lora_rank: 8 | |
| lora_dropout: 0.05 | |
| lora_target: all | |
| flash_attn: auto | |
| gradient_accumulation_steps: 1 | |
| include_num_input_tokens_seen: true | |
| learning_rate: 5e-5 | |
| logging_steps: 1 | |
| lr_scheduler_type: constant_with_warmup | |
| max_grad_norm: 2 | |
| max_samples: 100000000 | |
| model_name_or_path: /workspace/meta-llama/Llama-3.1-8B-Instruct | |
| num_train_epochs: 100000000 | |
| optim: adamw_torch | |
| output_dir: /workspace/v121rc_exp1/H | |
| packing: false | |
| # per_device_eval_batch_size: 64 | |
| per_device_train_batch_size: 64 | |
| plot_loss: true | |
| preprocessing_num_workers: 16 | |
| report_to: wandb | |
| save_steps: 1000 | |
| stage: sft | |
| template: llama3 | |
| trust_remote_code: true | |
| #val_size: 0.5 | |
| warmup_steps: 10 | |
| resize_vocab: true | |
| weight_decay: 1 | |
| adam_beta1: 0.9 | |
| adam_beta2: 0.98 | |
| # eval_on_each_dataset: true | |
| # compute_accuracy: true | |
| # accuracy_at_last_token: true | |
| # accuracy_with_generate: true | |
| # predict_with_generate: true | |
| # do_sample: false | |
| # temperature: 0.0 | |
| # top_p: 1.0 | |
| # max_new_tokens: 1024 | |
| # group_by_length: false | |
| # add_tokens: <MILLFIELD>,<Yes>,<No>,<think>,</think> | |