eval_results.json

{
    "epoch": 2.96,
    "eval_logits/chosen": 104.86785125732422,
    "eval_logits/rejected": 98.8290786743164,
    "eval_logps/chosen": -468.4523010253906,
    "eval_logps/rejected": -479.3721618652344,
    "eval_loss": 1.3044307231903076,
    "eval_rewards/accuracies": 0.75,
    "eval_rewards/chosen": -6.019142150878906,
    "eval_rewards/margins": 5.109894752502441,
    "eval_rewards/rejected": -11.129035949707031,
    "eval_runtime": 51.0519,
    "eval_samples": 750,
    "eval_samples_per_second": 14.691,
    "eval_steps_per_second": 0.47
}