eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": 0.3641189634799957,
    "eval_logits/rejected": 2.345853090286255,
    "eval_logps/chosen": -371.7200927734375,
    "eval_logps/rejected": -656.34228515625,
    "eval_loss": 0.004054788034409285,
    "eval_rewards/accuracies": 0.9983165264129639,
    "eval_rewards/chosen": 1.7269809246063232,
    "eval_rewards/margins": 17.098175048828125,
    "eval_rewards/rejected": -15.371195793151855,
    "eval_runtime": 270.3044,
    "eval_samples": 9500,
    "eval_samples_per_second": 35.146,
    "eval_steps_per_second": 1.099
}