eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.47953462600708,
    "eval_logits/rejected": -2.4293060302734375,
    "eval_logps/chosen": -325.9293518066406,
    "eval_logps/rejected": -359.2366638183594,
    "eval_loss": 0.8128312230110168,
    "eval_rewards/accuracies": 0.7734375,
    "eval_rewards/chosen": -5.867239475250244,
    "eval_rewards/margins": 3.072049140930176,
    "eval_rewards/rejected": -8.939288139343262,
    "eval_runtime": 93.0348,
    "eval_samples": 2000,
    "eval_samples_per_second": 21.497,
    "eval_steps_per_second": 0.344
}