eval_results.json

{
    "epoch": 2.999098751126561,
    "eval_kl/chosen": 0.00018642864597495645,
    "eval_kl/rejected": 0.0005706130177713931,
    "eval_logits/chosen": -1.5,
    "eval_logits/rejected": -1.46875,
    "eval_logps/chosen": -322.0,
    "eval_logps/rejected": -300.0,
    "eval_loss": 0.6433281302452087,
    "eval_nll_loss": 1.0546875,
    "eval_ref_probs/chosen": 0.0038271306548267603,
    "eval_ref_probs/rejected": 0.0034569373819977045,
    "eval_rejected_term/max": 0.0005706130177713931,
    "eval_rejected_term/min": 0.0005706130177713931,
    "eval_rewards/accuracies": 0.7360000014305115,
    "eval_rewards/chosen": -0.025146484375,
    "eval_rewards/margins": 0.2412109375,
    "eval_rewards/rejected": -0.265625,
    "eval_runtime": 29.8044,
    "eval_samples_per_second": 33.552,
    "eval_steps_per_second": 8.388
}