eval_results.json

{
    "epoch": 2.999098751126561,
    "eval_kl/chosen": -0.00044062218512408435,
    "eval_kl/rejected": 0.0005405896808952093,
    "eval_logits/chosen": -1.796875,
    "eval_logits/rejected": -1.78125,
    "eval_logps/chosen": -326.0,
    "eval_logps/rejected": -304.0,
    "eval_loss": 0.5493982434272766,
    "eval_nll_loss": 1.0625,
    "eval_ref_probs/chosen": 0.0038271306548267603,
    "eval_ref_probs/rejected": 0.0034569373819977045,
    "eval_rejected_term/max": 0.0005405896808952093,
    "eval_rejected_term/min": 0.0005405896808952093,
    "eval_rewards/accuracies": 0.7080000042915344,
    "eval_rewards/chosen": -0.365234375,
    "eval_rewards/margins": 0.44140625,
    "eval_rewards/rejected": -0.8046875,
    "eval_runtime": 25.316,
    "eval_samples_per_second": 39.501,
    "eval_steps_per_second": 9.875
}