eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.4555959701538086,
    "eval_logits/rejected": -2.1643569469451904,
    "eval_logps/chosen": -334.3052978515625,
    "eval_logps/rejected": -316.3751220703125,
    "eval_loss": 0.4730003774166107,
    "eval_rewards/accuracies": 0.8015872836112976,
    "eval_rewards/chosen": -3.5288658142089844,
    "eval_rewards/margins": 3.8411812782287598,
    "eval_rewards/rejected": -7.370047569274902,
    "eval_runtime": 139.3206,
    "eval_samples": 2000,
    "eval_samples_per_second": 14.355,
    "eval_steps_per_second": 0.452
}