Model: jackf857/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.4 Source: Original Platform
21 lines
820 B
JSON
21 lines
820 B
JSON
{
|
|
"epoch": 0.9989528795811519,
|
|
"eval_KL/chosen_KL_mean": -32.86964416503906,
|
|
"eval_KL/mean": -56.908538818359375,
|
|
"eval_KL/rejected_KL_mean": -80.94744110107422,
|
|
"eval_KL/std": 63.9189453125,
|
|
"eval_fcm_dpo/beta": 0.006948863156139851,
|
|
"eval_logits/chosen": 1.465262770652771,
|
|
"eval_logits/rejected": 1.5254052877426147,
|
|
"eval_logps/chosen": -313.286376953125,
|
|
"eval_logps/ref_chosen": -280.41668701171875,
|
|
"eval_logps/ref_rejected": -264.6455383300781,
|
|
"eval_logps/rejected": -345.59295654296875,
|
|
"eval_loss": 0.5767351388931274,
|
|
"eval_margin_dpo/margin_mean": 48.077789306640625,
|
|
"eval_margin_dpo/margin_std": 73.08638000488281,
|
|
"eval_runtime": 89.0979,
|
|
"eval_samples": 2000,
|
|
"eval_samples_per_second": 22.447,
|
|
"eval_steps_per_second": 2.806
|
|
} |