eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": NaN,
    "eval_logits/rejected": 1.3347594738006592,
    "eval_logps/chosen": -344.3968200683594,
    "eval_logps/rejected": -316.984130859375,
    "eval_loss": NaN,
    "eval_nll_loss": NaN,
    "eval_rewards/accuracies": 0.6458333134651184,
    "eval_rewards/chosen": 2.3433780670166016,
    "eval_rewards/margins": 0.7027374505996704,
    "eval_rewards/rejected": 1.6412450075149536,
    "eval_runtime": 8.3943,
    "eval_samples_per_second": 119.129,
    "eval_steps_per_second": 7.505
}