eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": -0.4878213703632355,
    "eval_logits/rejected": -0.4049469530582428,
    "eval_logps/chosen": -456.8681945800781,
    "eval_logps/rejected": -1991.3533935546875,
    "eval_loss": 0.14216427505016327,
    "eval_rewards/accuracies": 0.9166666865348816,
    "eval_rewards/chosen": -1.2709063291549683,
    "eval_rewards/margins": 11.992403984069824,
    "eval_rewards/rejected": -13.263309478759766,
    "eval_runtime": 29.9163,
    "eval_samples": 192,
    "eval_samples_per_second": 6.418,
    "eval_steps_per_second": 0.201
}