eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": -3.071336030960083,
    "eval_logits/rejected": -3.097411870956421,
    "eval_logps/chosen": -0.15935298800468445,
    "eval_logps/rejected": -284.3338623046875,
    "eval_loss": 0.06034620478749275,
    "eval_rewards/accuracies": 1.0,
    "eval_rewards/chosen": 0.3150060176849365,
    "eval_rewards/margins": 2.7981886863708496,
    "eval_rewards/rejected": -2.483182668685913,
    "eval_runtime": 2.5604,
    "eval_samples": 5,
    "eval_samples_per_second": 1.953,
    "eval_steps_per_second": 0.391
}