eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": NaN,
    "eval_logits/rejected": 1.2189980745315552,
    "eval_logps/chosen": -348.30157470703125,
    "eval_logps/rejected": -318.8888854980469,
    "eval_loss": NaN,
    "eval_nll_loss": NaN,
    "eval_rewards/accuracies": 0.6398809552192688,
    "eval_rewards/chosen": 1.948164701461792,
    "eval_rewards/margins": 0.5035187005996704,
    "eval_rewards/rejected": 1.4438244104385376,
    "eval_runtime": 8.3888,
    "eval_samples_per_second": 119.207,
    "eval_steps_per_second": 7.51
}