| seed: 0 |
| exp_name: dpo_mistral_fsdp_bf16_hh_harmless |
| batch_size: 8 |
| eval_batch_size: 16 |
| debug: false |
| fsdp_port: 30287 |
| datasets: |
| - hh |
| wandb: |
| enabled: true |
| entity: renatz1224 |
| project: dpo |
| api_key: d41bb0d59da8688d3d8f5371b1e3b3cfcced0f74 |
| local_dirs: |
| - /mntcephfs/data/med/chuyi2/dpo |
| sample_during_eval: false |
| n_eval_model_samples: 16 |
| do_first_eval: true |
| local_run_dir: /mntcephfs/data/med/chuyi2/dpo/kongchuyi/dpo_mistral_fsdp_bf16_hh_harmless_2024-02-21_19-35-15_643724 |
| lr: 5.0e-07 |
| gradient_accumulation_steps: 1 |
| max_grad_norm: 10.0 |
| max_length: 4096 |
| max_prompt_length: 2048 |
| n_epochs: 1 |
| n_examples: null |
| n_eval_examples: 256 |
| trainer: FSDPTrainer |
| optimizer: RMSprop |
| warmup_steps: 50 |
| activation_checkpointing: true |
| eval_every: 2000 |
| minimum_log_interval_secs: 1.0 |
| model: |
| name_or_path: /mntnfs/med_data2/kcy/la2-mistral-30k/checkpoint-445 |
| tokenizer_name_or_path: /mntnfs/med_data2/kcy/la2-mistral-30k/checkpoint-445 |
| archive: null |
| block_name: MistralDecoderLayer |
| policy_dtype: bfloat16 |
| fsdp_policy_mp: bfloat16 |
| reference_dtype: float16 |
| loss: |
| name: dpo |
| beta: 0.1 |
| label_smoothing: 0 |
| reference_free: false |
|
|