Model: W-61/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-s_star-0.4-20260425-111846 Source: Original Platform
17 lines
642 B
JSON
17 lines
642 B
JSON
{
|
|
"epoch": 0.9989528795811519,
|
|
"eval_fcm_dpo/beta": 0.0036579067818820477,
|
|
"eval_logits/chosen": -0.8699643611907959,
|
|
"eval_logits/rejected": -0.854047954082489,
|
|
"eval_logps/chosen": -504.4405212402344,
|
|
"eval_logps/ref_chosen": -287.8268127441406,
|
|
"eval_logps/ref_rejected": -266.9300231933594,
|
|
"eval_logps/rejected": -574.5838623046875,
|
|
"eval_loss": 0.582348108291626,
|
|
"eval_margin_dpo/margin_mean": 91.04013061523438,
|
|
"eval_margin_dpo/margin_std": 147.3343505859375,
|
|
"eval_runtime": 78.8341,
|
|
"eval_samples": 2000,
|
|
"eval_samples_per_second": 25.37,
|
|
"eval_steps_per_second": 3.171
|
|
} |