| assistant_tag: assistant | |
| bf16: true | |
| content_tag: value | |
| cutoff_len: 2048 | |
| dataset: mlfoundations-dev/openthoughts1_short | |
| dataset_dir: ONLINE | |
| ddp_timeout: 180000000 | |
| deepspeed: dcft/train/zero3_dclm.json | |
| do_train: true | |
| enable_liger_kernel: true | |
| finetuning_type: full | |
| formatting: sharegpt | |
| global_batch_size: 16 | |
| gradient_accumulation_steps: 2 | |
| hub_model_id: mlfoundations-dev/dclm_baseline_openthoughts1 | |
| learning_rate: 1.0e-05 | |
| logging_steps: 1 | |
| lr_scheduler_type: cosine | |
| max_samples: 1000000 | |
| messages: conversations | |
| model_name_or_path: apple/DCLM-7B | |
| neat_packing: false | |
| num_train_epochs: 5.0 | |
| output_dir: /data/horse/ws/rehe951g-p_finetuning/checkpoints/dclm_baseline_openthoughts1 | |
| overwrite_cache: true | |
| overwrite_output_dir: true | |
| packing: false | |
| per_device_eval_batch_size: 2 | |
| per_device_train_batch_size: 2 | |
| plot_loss: true | |
| preprocessing_num_workers: 16 | |
| push_to_db: true | |
| push_to_hub: true | |
| report_to: wandb | |
| role_tag: from | |
| run_name: dclm_baseline_openthoughts1 | |
| save_steps: 10 | |
| stage: sft | |
| template: qwen25 | |
| user_tag: user | |
| warmup_ratio: 0.1 | |