Files
llama-3-8b-base-new-dpo-ult…/all_results.json
ModelHub XC 1d0aa7f57e 初始化项目,由ModelHub XC社区提供模型
Model: W-61/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.3-20260428-045924
Source: Original Platform
2026-05-15 20:24:15 +08:00

27 lines
1020 B
JSON

{
"epoch": 0.9989528795811519,
"eval_KL/chosen_KL_mean": -467.01068115234375,
"eval_KL/mean": -554.3729858398438,
"eval_KL/rejected_KL_mean": -641.7353515625,
"eval_KL/std": 329.4678649902344,
"eval_fcm_dpo/beta": 0.0010197119554504752,
"eval_logits/chosen": -0.7331064939498901,
"eval_logits/rejected": -0.7050662040710449,
"eval_logps/chosen": -754.8375244140625,
"eval_logps/ref_chosen": -287.8268127441406,
"eval_logps/ref_rejected": -266.9300231933594,
"eval_logps/rejected": -908.6654663085938,
"eval_loss": 0.6327955722808838,
"eval_margin_dpo/margin_mean": 174.72463989257812,
"eval_margin_dpo/margin_std": 379.1971740722656,
"eval_runtime": 78.0079,
"eval_samples": 2000,
"eval_samples_per_second": 25.638,
"eval_steps_per_second": 3.205,
"total_flos": 0.0,
"train_loss": 5.002585172153369,
"train_runtime": 6534.5428,
"train_samples": 61135,
"train_samples_per_second": 9.356,
"train_steps_per_second": 0.073
}