eval_results.json

{
    "epoch": 2.999098751126561,
    "eval_kl/both": 0.0004668425826821476,
    "eval_kl/chosen": 0.00024188528186641634,
    "eval_kl/rejected": 0.00022479292238131166,
    "eval_logits/chosen": -1.4296875,
    "eval_logits/rejected": -1.3984375,
    "eval_logps/chosen": -322.0,
    "eval_logps/rejected": -298.0,
    "eval_loss": 0.6863964796066284,
    "eval_nll_loss": 1.046875,
    "eval_ref_probs/chosen": 0.0038271306548267603,
    "eval_ref_probs/rejected": 0.0034569373819977045,
    "eval_rejected_term/max": 0.00022479292238131166,
    "eval_rejected_term/min": 0.00022479292238131166,
    "eval_rewards/accuracies": 0.6480000019073486,
    "eval_rewards/chosen": 0.0033721923828125,
    "eval_rewards/margins": 0.1484375,
    "eval_rewards/rejected": -0.14453125,
    "eval_runtime": 25.5721,
    "eval_samples_per_second": 39.105,
    "eval_steps_per_second": 9.776
}