eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": 0.9743706583976746,
    "eval_logits/rejected": 0.6065909266471863,
    "eval_logps/chosen": -139.52996826171875,
    "eval_logps/rejected": -142.2351837158203,
    "eval_loss": 0.01678086631000042,
    "eval_rewards/accuracies": 0.9956116080284119,
    "eval_rewards/chosen": -1.5749770402908325,
    "eval_rewards/margins": 9.825251579284668,
    "eval_rewards/rejected": -11.400227546691895,
    "eval_runtime": 379.7007,
    "eval_samples_per_second": 9.6,
    "eval_steps_per_second": 4.801
}