eval_results.json

{
    "epoch": 2.999098751126561,
    "eval_kl/chosen": -3.394229775716129e-12,
    "eval_kl/rejected": -4.694334859527659e-14,
    "eval_logits/chosen": -2.46875,
    "eval_logits/rejected": -2.578125,
    "eval_logps/chosen": -370.0,
    "eval_logps/rejected": -368.0,
    "eval_loss": 0.4384387135505676,
    "eval_nll_loss": 1.65625,
    "eval_ref_probs/chosen": 0.0038271306548267603,
    "eval_ref_probs/rejected": 0.0034569373819977045,
    "eval_rejected_term/max": -4.694334859527659e-14,
    "eval_rejected_term/min": -4.694334859527659e-14,
    "eval_rewards/accuracies": 0.7760000228881836,
    "eval_rewards/chosen": -4.90625,
    "eval_rewards/margins": 2.171875,
    "eval_rewards/rejected": -7.09375,
    "eval_runtime": 25.4932,
    "eval_samples_per_second": 39.226,
    "eval_steps_per_second": 9.807
}