| adam_beta1: '0.9' | |
| adam_beta2: '0.999' | |
| bf16: 'True' | |
| cutoff_len: '2048' | |
| dataset: mlfoundations-dev/oh-dcft-v3.1-llama-3.1-8b | |
| dataset_dir: ONLINE | |
| ddp_timeout: '180000000' | |
| deepspeed: /opt/ml/code/zero3.json | |
| do_train: 'True' | |
| enable_liger_kernel: 'False' | |
| eval_strategy: epoch | |
| finetuning_type: full | |
| formatting: sharegpt | |
| global_batch_size: '512' | |
| gradient_accumulation_steps: '8' | |
| gradient_checkpointing: 'True' | |
| hub_model_id: mlfoundations-dev/oh-dcft-v3.1-llama-3.1-8b | |
| learning_rate: 5e-06 | |
| logging_steps: '10' | |
| lr_scheduler_type: constant | |
| max_grad_norm: '1' | |
| messages: conversations | |
| model_name_or_path: meta-llama/Meta-Llama-3.1-8B | |
| neat_packing: 'True' | |
| num_train_epochs: '3.0' | |
| output_dir: /opt/ml/model | |
| overwrite_cache: 'True' | |
| overwrite_output_dir: 'True' | |
| packing: 'True' | |
| per_device_train_batch_size: '8' | |
| plot_loss: 'True' | |
| preprocessing_num_workers: '16' | |
| push_to_db: 'True' | |
| push_to_hub: 'True' | |
| report_to: wandb | |
| run_name: oh-dcft-v3.1-llama-3.1-8b | |
| save_strategy: epoch | |
| stage: sft | |
| template: llama3 | |
| val_size: '0.05' | |
| warmup_ratio: '0.1' | |
| warmup_steps: '1738' | |
| weight_decay: '0.1' | |