eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.1697230339050293,
    "eval_logits/rejected": -2.15024995803833,
    "eval_logps/chosen": -218.56118774414062,
    "eval_logps/rejected": -253.41075134277344,
    "eval_loss": 0.641146719455719,
    "eval_rewards/accuracies": 0.6273234486579895,
    "eval_rewards/chosen": -1.5954570770263672,
    "eval_rewards/margins": 0.3111543357372284,
    "eval_rewards/rejected": -1.906611442565918,
    "eval_runtime": 381.9987,
    "eval_samples": 4304,
    "eval_samples_per_second": 11.267,
    "eval_steps_per_second": 1.408
}