eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.952404499053955,
    "eval_logits/rejected": -2.754680871963501,
    "eval_logps/chosen": -247.9913787841797,
    "eval_logps/rejected": -735.1937866210938,
    "eval_loss": 0.002894825069233775,
    "eval_rewards/accuracies": 0.9991582632064819,
    "eval_rewards/chosen": 1.4965133666992188,
    "eval_rewards/margins": 34.79835510253906,
    "eval_rewards/rejected": -33.301841735839844,
    "eval_runtime": 463.448,
    "eval_samples": 9500,
    "eval_samples_per_second": 20.499,
    "eval_steps_per_second": 0.641
}