Model: jackf857/llama-3-8b-base-new-dpo-hh-helpful-s_star0.85-4xh200-batch-64-20260421-233802 Source: Original Platform
20 lines
770 B
JSON
20 lines
770 B
JSON
{
|
|
"epoch": 1.0,
|
|
"eval_fcm_dpo/beta": 0.005598581861704588,
|
|
"eval_fcm_dpo/delta": 0.061232246458530426,
|
|
"eval_fcm_dpo/margin": 140.12789916992188,
|
|
"eval_fcm_dpo/q_t": 0.35402458906173706,
|
|
"eval_logits/chosen": 0.21182169020175934,
|
|
"eval_logits/rejected": 0.23247361183166504,
|
|
"eval_logps/chosen": -505.7603759765625,
|
|
"eval_logps/ref_chosen": -79.05104064941406,
|
|
"eval_logps/ref_rejected": -86.79793548583984,
|
|
"eval_logps/rejected": -653.6351928710938,
|
|
"eval_loss": 0.5076749920845032,
|
|
"eval_margin_dpo/margin_mean": 140.12789916992188,
|
|
"eval_margin_dpo/margin_std": 204.84616088867188,
|
|
"eval_runtime": 39.5033,
|
|
"eval_samples": 2339,
|
|
"eval_samples_per_second": 59.21,
|
|
"eval_steps_per_second": 1.873
|
|
} |