eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": 1.106881022453308,
    "eval_logits/rejected": 1.0551656484603882,
    "eval_logps/chosen": -504.99224853515625,
    "eval_logps/rejected": -705.9820556640625,
    "eval_loss": 0.4338752329349518,
    "eval_rewards/accuracies": 0.8196428418159485,
    "eval_rewards/chosen": -2.3226053714752197,
    "eval_rewards/margins": 1.9564555883407593,
    "eval_rewards/rejected": -4.2790608406066895,
    "eval_runtime": 187.0474,
    "eval_samples": 4461,
    "eval_samples_per_second": 23.85,
    "eval_steps_per_second": 0.374
}