eval_results.json

{
    "epoch": 3.97,
    "eval_logits/chosen": -0.00874541699886322,
    "eval_logits/rejected": 0.0888209417462349,
    "eval_logps/chosen": -306.3463439941406,
    "eval_logps/rejected": -278.6518249511719,
    "eval_loss": 0.00159652519505471,
    "eval_rewards/accuracies": 0.4909999966621399,
    "eval_rewards/chosen": -0.0007702131988480687,
    "eval_rewards/margins": 5.095174856251106e-05,
    "eval_rewards/rejected": -0.0008211650419980288,
    "eval_runtime": 2003.1713,
    "eval_samples": 2000,
    "eval_samples_per_second": 0.998,
    "eval_steps_per_second": 0.25
}