eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": 4.039111614227295,
    "eval_logits/rejected": 3.562220811843872,
    "eval_logps/chosen": -965.5345458984375,
    "eval_logps/rejected": -1123.8455810546875,
    "eval_loss": 0.6446220278739929,
    "eval_rewards/accuracies": 0.7151826620101929,
    "eval_rewards/chosen": -8.027778625488281,
    "eval_rewards/margins": 1.5074113607406616,
    "eval_rewards/rejected": -9.535189628601074,
    "eval_runtime": 534.3288,
    "eval_samples": 7000,
    "eval_samples_per_second": 13.101,
    "eval_steps_per_second": 0.41
}