eval_results.json

{
    "epoch": 2.999098751126561,
    "eval_kl/chosen": 0.00021544050832744688,
    "eval_kl/rejected": 0.0003815116942860186,
    "eval_logits/chosen": -1.3671875,
    "eval_logits/rejected": -1.34375,
    "eval_logps/chosen": -324.0,
    "eval_logps/rejected": -300.0,
    "eval_loss": 0.8039374947547913,
    "eval_nll_loss": 1.0625,
    "eval_ref_probs/chosen": 0.0038271306548267603,
    "eval_ref_probs/rejected": 0.0034569373819977045,
    "eval_rejected_term/max": 0.0003815116942860186,
    "eval_rejected_term/min": 0.0003815116942860186,
    "eval_rewards/accuracies": 0.46000000834465027,
    "eval_rewards/chosen": -0.35546875,
    "eval_rewards/margins": 0.03125,
    "eval_rewards/rejected": -0.38671875,
    "eval_runtime": 25.3885,
    "eval_samples_per_second": 39.388,
    "eval_steps_per_second": 9.847
}