eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.7684097290039062,
    "eval_logits/rejected": -2.762378454208374,
    "eval_logps/chosen": -87.6427230834961,
    "eval_logps/rejected": -98.04631042480469,
    "eval_loss": 0.6686788201332092,
    "eval_rewards/accuracies": 0.600836455821991,
    "eval_rewards/chosen": -0.2893083095550537,
    "eval_rewards/margins": 0.05935351550579071,
    "eval_rewards/rejected": -0.3486618399620056,
    "eval_runtime": 383.8103,
    "eval_samples": 4304,
    "eval_samples_per_second": 11.214,
    "eval_steps_per_second": 1.402
}