eval_results.json

{
    "epoch": 2.0,
    "eval_logits/chosen": -2.5326499938964844,
    "eval_logits/rejected": -2.516491174697876,
    "eval_logps/chosen": -72.48519134521484,
    "eval_logps/rejected": -80.36906433105469,
    "eval_loss": 0.8962341547012329,
    "eval_rewards/accuracies": 0.3333333432674408,
    "eval_rewards/chosen": 1.9939475059509277,
    "eval_rewards/margins": 1.2392425537109375,
    "eval_rewards/rejected": 0.7547050714492798,
    "eval_runtime": 113.8391,
    "eval_samples": 2000,
    "eval_samples_per_second": 17.569,
    "eval_steps_per_second": 0.553
}