Model: jackf857/llama-3-8b-base-new-dpo-hh-helpful-s_star0.6-4xh200-batch-64-20260421-214335-rerun Source: Original Platform
20 lines
768 B
JSON
20 lines
768 B
JSON
{
|
|
"epoch": 1.0,
|
|
"eval_fcm_dpo/beta": 0.003359581343829632,
|
|
"eval_fcm_dpo/delta": 0.061418402940034866,
|
|
"eval_fcm_dpo/margin": 157.58413696289062,
|
|
"eval_fcm_dpo/q_t": 0.39175158739089966,
|
|
"eval_logits/chosen": -0.528179943561554,
|
|
"eval_logits/rejected": -0.5172193050384521,
|
|
"eval_logps/chosen": -449.2883605957031,
|
|
"eval_logps/ref_chosen": -79.05104064941406,
|
|
"eval_logps/ref_rejected": -86.79793548583984,
|
|
"eval_logps/rejected": -614.6193237304688,
|
|
"eval_loss": 0.5511718988418579,
|
|
"eval_margin_dpo/margin_mean": 157.5841522216797,
|
|
"eval_margin_dpo/margin_std": 262.95001220703125,
|
|
"eval_runtime": 39.5766,
|
|
"eval_samples": 2339,
|
|
"eval_samples_per_second": 59.101,
|
|
"eval_steps_per_second": 1.87
|
|
} |