eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": -0.43692123889923096,
    "eval_logits/rejected": 0.023339029401540756,
    "eval_logps/chosen": -511.4042663574219,
    "eval_logps/rejected": -623.3074340820312,
    "eval_loss": 0.49832019209861755,
    "eval_rewards/accuracies": 0.76953125,
    "eval_rewards/chosen": -2.488020420074463,
    "eval_rewards/margins": 1.118230938911438,
    "eval_rewards/rejected": -3.6062512397766113,
    "eval_runtime": 42.6828,
    "eval_samples": 2000,
    "eval_samples_per_second": 46.857,
    "eval_steps_per_second": 0.75
}