Model: W-61/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725 Source: Original Platform
21 lines
821 B
JSON
21 lines
821 B
JSON
{
|
|
"epoch": 0.9989528795811519,
|
|
"eval_KL/chosen_KL_mean": -41.074798583984375,
|
|
"eval_KL/mean": -67.0528793334961,
|
|
"eval_KL/rejected_KL_mean": -93.03097534179688,
|
|
"eval_KL/std": 70.80274963378906,
|
|
"eval_fcm_dpo/beta": 0.005364115349948406,
|
|
"eval_logits/chosen": 1.4891531467437744,
|
|
"eval_logits/rejected": 1.5578936338424683,
|
|
"eval_logps/chosen": -321.4914855957031,
|
|
"eval_logps/ref_chosen": -280.41668701171875,
|
|
"eval_logps/ref_rejected": -264.6455383300781,
|
|
"eval_logps/rejected": -357.676513671875,
|
|
"eval_loss": 0.5913359522819519,
|
|
"eval_margin_dpo/margin_mean": 51.956180572509766,
|
|
"eval_margin_dpo/margin_std": 81.990966796875,
|
|
"eval_runtime": 88.3876,
|
|
"eval_samples": 2000,
|
|
"eval_samples_per_second": 22.628,
|
|
"eval_steps_per_second": 2.828
|
|
} |