eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": NaN,
    "eval_logits/rejected": 1.272383451461792,
    "eval_logps/chosen": -351.8888854980469,
    "eval_logps/rejected": -320.4444580078125,
    "eval_loss": NaN,
    "eval_nll_loss": NaN,
    "eval_rewards/accuracies": 0.6140872836112976,
    "eval_rewards/chosen": 1.5822793245315552,
    "eval_rewards/margins": 0.2936420738697052,
    "eval_rewards/rejected": 1.2895585298538208,
    "eval_runtime": 8.4007,
    "eval_samples_per_second": 119.037,
    "eval_steps_per_second": 7.499
}