eval_results.json

{
    "epoch": 2.999297541394882,
    "eval_logits/chosen": 0.9163225889205933,
    "eval_logits/rejected": 0.9576992988586426,
    "eval_logps/chosen": -9.975740432739258,
    "eval_logps/rejected": -11.105355262756348,
    "eval_loss": 0.5245004892349243,
    "eval_rewards/accuracies": 0.7240356206893921,
    "eval_rewards/chosen": -9.975740432739258,
    "eval_rewards/margins": 1.1296145915985107,
    "eval_rewards/rejected": -11.105355262756348,
    "eval_runtime": 34.6187,
    "eval_samples": 1345,
    "eval_samples_per_second": 38.852,
    "eval_semantic_entropy": 0.0012647123076021671,
    "eval_steps_per_second": 9.735
}