eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.595337390899658,
    "eval_logits/rejected": -2.54485821723938,
    "eval_logps/chosen": -320.2513122558594,
    "eval_logps/rejected": -315.28961181640625,
    "eval_loss": 0.7336946725845337,
    "eval_rewards/accuracies": 0.7720000147819519,
    "eval_rewards/chosen": -4.91001558303833,
    "eval_rewards/margins": 3.770545721054077,
    "eval_rewards/rejected": -8.680562973022461,
    "eval_runtime": 296.244,
    "eval_samples": 2000,
    "eval_samples_per_second": 6.751,
    "eval_steps_per_second": 0.422
}