| checkpoint_dir: checkpoints/meta-llama/Meta-Llama-3-8B-Instruct | |
| out_dir: /apdcephfs/share_300000800/user/wenlinyao/research2/AlphaCode/litgpt_trained_model/llama-3-8b-instruct_4k_alphaflow_v10 | |
| precision: bf16-true | |
| devices: 8 | |
| resume: false | |
| data: | |
| class_path: litgpt.data.JSON | |
| init_args: | |
| json_path: /apdcephfs/share_300000800/user/wenlinyao/research2/AlphaCode/litgpt_data/alphaflow_training_data_v10.json | |
| mask_prompt: false | |
| val_split_fraction: 0.005 | |
| prompt_style: llama3 | |
| ignore_index: -100 | |
| seed: 42 | |
| num_workers: 4 | |
| train: | |
| save_interval: 600 | |
| log_interval: 1 | |
| global_batch_size: 128 | |
| micro_batch_size: 1 | |
| lr_warmup_steps: 50 | |
| epochs: 4 | |
| max_seq_length: 4096 | |
| min_lr: 2.0e-06 | |
| eval: | |
| interval: 200 | |
| max_new_tokens: 2048 | |
| max_iters: 100 | |
| initial_validation: false | |
| optimizer: | |
| class_path: torch.optim.AdamW | |
| init_args: | |
| lr: 2.0e-05 | |
| weight_decay: 0.02 | |
| betas: | |
| - 0.9 | |
| - 0.95 | |
| logger_name: csv | |
| seed: 11 | |