Files
qwen3-8b-base-margin-dpo-ul…/eval_results.json
ModelHub XC 27acccf3d4 初始化项目,由ModelHub XC社区提供模型
Model: W-61/qwen3-8b-base-margin-dpo-ultrafeedback-4xh200-batch-128-20260423-040315
Source: Original Platform
2026-06-01 09:31:36 +08:00

16 lines
597 B
JSON

{
"epoch": 0.9989528795811519,
"eval_logits/chosen": 1.1346925497055054,
"eval_logits/rejected": 1.1706666946411133,
"eval_logps/chosen": -313.02630615234375,
"eval_logps/ref_chosen": -281.4588928222656,
"eval_logps/ref_rejected": -261.84954833984375,
"eval_logps/rejected": -342.32696533203125,
"eval_loss": 0.5572407245635986,
"eval_margin_dpo/margin_mean": 48.91001510620117,
"eval_margin_dpo/margin_std": 68.51956176757812,
"eval_runtime": 92.7227,
"eval_samples": 2000,
"eval_samples_per_second": 21.57,
"eval_steps_per_second": 1.348
}