| { |
| "model_name_or_path": "output/sft", |
| "max_length": 512, |
| "trust_remote_code": true, |
| "train_datasets": [ |
| [ |
| "PKU-SafeRLHF/train", |
| { |
| "proportion": 1.0, |
| "path": "PKU-SafeRLHF-harmless-only-30k" |
| } |
| ] |
| ], |
| "eval_datasets": [ |
| [ |
| "PKU-SafeRLHF/test", |
| { |
| "proportion": 1.0 |
| } |
| ] |
| ], |
| "loss_type": "sequence-wise", |
| "epochs": 2, |
| "per_device_train_batch_size": 16, |
| "per_device_eval_batch_size": 16, |
| "gradient_accumulation_steps": 2, |
| "gradient_checkpointing": true, |
| "normalize_score_during_training": false, |
| "normalizer_type": "ExponentialMovingAverage", |
| "normalizer_momentum": 0.9, |
| "lr": 2e-05, |
| "lr_scheduler_type": "cosine", |
| "lr_warmup_ratio": 0.03, |
| "weight_decay": 0.1, |
| "seed": 42, |
| "fp16": false, |
| "bf16": true, |
| "tf32": true, |
| "eval_strategy": "epoch", |
| "eval_interval": 1000000, |
| "need_eval": false, |
| "eval_split_ratio": null, |
| "output_dir": "/data/jiongxiao_wang/rlhf_attack/safe-rlhf/output/rm_30k", |
| "log_type": "wandb", |
| "log_dir": "/data/jiongxiao_wang/rlhf_attack/safe-rlhf/output/rm_30k", |
| "log_project": "Safe-RLHF-RM", |
| "log_run_name": "reward-2024-01-05-20-03-25", |
| "save_16bit": false, |
| "save_interval": 1000000, |
| "local_rank": 0, |
| "zero_stage": 3, |
| "deepspeed": false, |
| "deepspeed_config": null, |
| "deepscale": false, |
| "deepscale_config": null, |
| "deepspeed_mpi": false, |
| "global_rank": 0, |
| "device": { |
| "type": "torch.device", |
| "repr": "device(type='cuda', index=0)" |
| }, |
| "num_update_steps_per_epoch": 210, |
| "total_training_steps": 420 |
| } |
