eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.687349796295166,
    "eval_logits/rejected": -2.300260305404663,
    "eval_logps/chosen": -268.54150390625,
    "eval_logps/rejected": -212.86087036132812,
    "eval_loss": 18.088815689086914,
    "eval_rewards/accuracies": 0.7319999933242798,
    "eval_rewards/chosen": 0.02013408951461315,
    "eval_rewards/margins": 0.13951165974140167,
    "eval_rewards/rejected": -0.11937756091356277,
    "eval_runtime": 350.1967,
    "eval_samples": 2000,
    "eval_samples_per_second": 5.711,
    "eval_steps_per_second": 0.714
}