eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -0.3196437656879425,
    "eval_logits/rejected": 0.2021237313747406,
    "eval_logps/chosen": -275.47796630859375,
    "eval_logps/rejected": -644.0859375,
    "eval_loss": 0.005032053682953119,
    "eval_rewards/accuracies": 0.9991582632064819,
    "eval_rewards/chosen": -0.09963408857584,
    "eval_rewards/margins": 19.299285888671875,
    "eval_rewards/rejected": -19.3989200592041,
    "eval_runtime": 525.9739,
    "eval_samples": 9500,
    "eval_samples_per_second": 18.062,
    "eval_steps_per_second": 0.565
}