eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": NaN,
    "eval_logits/rejected": 1.129960298538208,
    "eval_logps/chosen": -353.19049072265625,
    "eval_logps/rejected": -322.8888854980469,
    "eval_loss": NaN,
    "eval_nll_loss": NaN,
    "eval_rewards/accuracies": 0.6557539701461792,
    "eval_rewards/chosen": 1.4410961866378784,
    "eval_rewards/margins": 0.3948955237865448,
    "eval_rewards/rejected": 1.0456969738006592,
    "eval_runtime": 8.4152,
    "eval_samples_per_second": 118.832,
    "eval_steps_per_second": 7.486
}