eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.3639509677886963,
    "eval_logits/rejected": -2.4372596740722656,
    "eval_logps/chosen": -316.3568115234375,
    "eval_logps/rejected": -327.0231628417969,
    "eval_loss": 0.7655794024467468,
    "eval_rewards/accuracies": 0.7404580116271973,
    "eval_rewards/chosen": -3.8106329441070557,
    "eval_rewards/margins": 3.0781688690185547,
    "eval_rewards/rejected": -6.8888020515441895,
    "eval_runtime": 297.1799,
    "eval_samples": 2086,
    "eval_samples_per_second": 7.019,
    "eval_steps_per_second": 0.441
}