| { | |
| "anchor_mode": "true", | |
| "base_model": "Qwen/Qwen2.5-7B", | |
| "batch_size": 4, | |
| "bce_tau": 12.0, | |
| "epochs": 5, | |
| "grad_accum": 8, | |
| "grad_ckpt": true, | |
| "grad_clip": 1.0, | |
| "human_class": false, | |
| "lam_anchor": 0.0, | |
| "lam_disc": 0.0, | |
| "lam_hum": 0.0, | |
| "lam_marg": 0.0, | |
| "lam_rej": 0.0, | |
| "lam_sft": 0.3, | |
| "lora_alpha": 32, | |
| "lora_dropout": 0.05, | |
| "lora_r": 16, | |
| "lora_target_modules": [ | |
| "q_proj", | |
| "k_proj", | |
| "v_proj", | |
| "o_proj" | |
| ], | |
| "loss": "ce", | |
| "lr": 0.0001, | |
| "marg_T": 1.0, | |
| "marg_detach": false, | |
| "max_len": 512, | |
| "per_class": 0, | |
| "per_source_lora_b": true, | |
| "prefix_len": 1, | |
| "rej_beta": 0.2, | |
| "rej_mode": "pair", | |
| "seed": 42, | |
| "sources": [ | |
| "gpt", | |
| "claude", | |
| "gemini", | |
| "mistral", | |
| "llama" | |
| ], | |
| "tau": 1.0, | |
| "warmup_ratio": 0.05, | |
| "weight_decay": 0.01 | |
| } |