| base_model: teknium/OpenHermes-2.5-Mistral-7B | |
| model_type: MistralForCausalLM | |
| tokenizer_type: LlamaTokenizer | |
| is_mistral_derived_model: true | |
| load_in_8bit: false | |
| load_in_4bit: true | |
| strict: false | |
| rl: true | |
| datasets: | |
| - path: winglian/no_robots_rlhf | |
| split: train | |
| test_datasets: | |
| - path: winglian/no_robots_rlhf | |
| split: test | |
| dataset_prepared_path: | |
| val_set_size: 0.0 | |
| output_dir: ./qlora-out | |
| adapter: qlora | |
| lora_model_dir: | |
| sequence_len: 2048 | |
| sample_packing: false | |
| pad_to_sequence_len: true | |
| lora_r: 64 | |
| lora_alpha: 16 | |
| lora_dropout: 0.05 | |
| lora_target_modules: | |
| lora_target_linear: true | |
| lora_fan_in_fan_out: | |
| wandb_project: oh-ipo | |
| wandb_entity: oaaic | |
| wandb_watch: | |
| wandb_run_id: | |
| wandb_log_model: | |
| gradient_accumulation_steps: 4 | |
| micro_batch_size: 2 | |
| num_epochs: 3 | |
| optimizer: rmsprop | |
| lr_scheduler: cosine | |
| learning_rate: 5e-7 | |
| train_on_inputs: false | |
| group_by_length: false | |
| bf16: true | |
| fp16: false | |
| tf32: false | |
| gradient_checkpointing: true | |
| resume_from_checkpoint: | |
| logging_steps: 1 | |
| flash_attention: true | |
| warmup_steps: 10 | |
| eval_steps: 0.0 | |
| eval_table_size: | |
| save_steps: | |
| deepspeed: deepspeed/zero3.json | |
| weight_decay: 0.0 | |