Model: W-61/llama-3-8b-base-new-dpo-hh-harmless-s_star0.85-4xh200-batch-64-20260421-213851 Source: Original Platform
20 lines
781 B
JSON
20 lines
781 B
JSON
{
|
|
"epoch": 0.999244142101285,
|
|
"eval_fcm_dpo/beta": 0.05573597922921181,
|
|
"eval_fcm_dpo/delta": 0.0020733291748911142,
|
|
"eval_fcm_dpo/margin": 15.057168006896973,
|
|
"eval_fcm_dpo/q_t": 0.349899560213089,
|
|
"eval_logits/chosen": 0.4825221300125122,
|
|
"eval_logits/rejected": 0.4322907030582428,
|
|
"eval_logps/chosen": -98.41519927978516,
|
|
"eval_logps/ref_chosen": -74.85946655273438,
|
|
"eval_logps/ref_rejected": -79.54898834228516,
|
|
"eval_logps/rejected": -118.1618881225586,
|
|
"eval_loss": 0.5171682834625244,
|
|
"eval_margin_dpo/margin_mean": 15.057168006896973,
|
|
"eval_margin_dpo/margin_std": 22.871326446533203,
|
|
"eval_runtime": 38.4532,
|
|
"eval_samples": 2303,
|
|
"eval_samples_per_second": 59.891,
|
|
"eval_steps_per_second": 1.872
|
|
} |