eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": 0.7048717737197876,
    "eval_logits/rejected": 0.9901154041290283,
    "eval_logps/chosen": -329.05975341796875,
    "eval_logps/rejected": -411.0256652832031,
    "eval_loss": 0.5257185101509094,
    "eval_rewards/accuracies": 0.76953125,
    "eval_rewards/chosen": -0.6523305773735046,
    "eval_rewards/margins": 0.8195381760597229,
    "eval_rewards/rejected": -1.4718689918518066,
    "eval_runtime": 138.5096,
    "eval_samples": 2000,
    "eval_samples_per_second": 14.439,
    "eval_steps_per_second": 0.231
}