eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -0.4354749321937561,
    "eval_logits/rejected": -0.46039730310440063,
    "eval_logps/chosen": -116.15072631835938,
    "eval_logps/rejected": -95.56389617919922,
    "eval_loss": 0.1421111524105072,
    "eval_rewards/accuracies": 0.9441340565681458,
    "eval_rewards/chosen": 3.5045158863067627,
    "eval_rewards/margins": 5.8216233253479,
    "eval_rewards/rejected": -2.317107915878296,
    "eval_runtime": 911.8075,
    "eval_samples": 2862,
    "eval_samples_per_second": 3.139,
    "eval_steps_per_second": 0.196
}