Files
llama3-8b-base-new-method-s…/eval_results.json
ModelHub XC 1e65d1513d 初始化项目,由ModelHub XC社区提供模型
Model: W-61/llama3-8b-base-new-method-s_star0.6-20260425-180936
Source: Original Platform
2026-05-05 20:48:43 +08:00

17 lines
644 B
JSON

{
"epoch": 0.9989528795811519,
"eval_fcm_dpo/beta": 0.009593765251338482,
"eval_logits/chosen": -0.8410288691520691,
"eval_logits/rejected": -0.8335962891578674,
"eval_logps/chosen": -391.4317626953125,
"eval_logps/ref_chosen": -287.8267517089844,
"eval_logps/ref_rejected": -266.9313659667969,
"eval_logps/rejected": -426.3018493652344,
"eval_loss": 0.5413669347763062,
"eval_margin_dpo/margin_mean": 55.76554870605469,
"eval_margin_dpo/margin_std": 85.69520568847656,
"eval_runtime": 81.3525,
"eval_samples": 2000,
"eval_samples_per_second": 24.584,
"eval_steps_per_second": 1.537
}