Model: W-61/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260428-045924 Source: Original Platform
21 lines
829 B
JSON
21 lines
829 B
JSON
{
|
|
"epoch": 0.9989528795811519,
|
|
"eval_KL/chosen_KL_mean": -268.62713623046875,
|
|
"eval_KL/mean": -320.5470886230469,
|
|
"eval_KL/rejected_KL_mean": -372.4670715332031,
|
|
"eval_KL/std": 159.96376037597656,
|
|
"eval_fcm_dpo/beta": 0.0025054924190044403,
|
|
"eval_logits/chosen": -0.8551161289215088,
|
|
"eval_logits/rejected": -0.8376473784446716,
|
|
"eval_logps/chosen": -556.4539184570312,
|
|
"eval_logps/ref_chosen": -287.8268127441406,
|
|
"eval_logps/ref_rejected": -266.9300231933594,
|
|
"eval_logps/rejected": -639.3970947265625,
|
|
"eval_loss": 0.6023811101913452,
|
|
"eval_margin_dpo/margin_mean": 103.83995056152344,
|
|
"eval_margin_dpo/margin_std": 181.04119873046875,
|
|
"eval_runtime": 77.9291,
|
|
"eval_samples": 2000,
|
|
"eval_samples_per_second": 25.664,
|
|
"eval_steps_per_second": 3.208
|
|
} |