eval_results.json

{
    "epoch": 2.0,
    "eval_logits/chosen": -2.500744581222534,
    "eval_logits/rejected": -2.4560909271240234,
    "eval_logps/chosen": -289.780517578125,
    "eval_logps/rejected": -287.32733154296875,
    "eval_loss": 0.5590207576751709,
    "eval_rewards/accuracies": 0.7857142686843872,
    "eval_rewards/chosen": -0.7818440198898315,
    "eval_rewards/margins": 1.9296892881393433,
    "eval_rewards/rejected": -2.711533308029175,
    "eval_runtime": 224.8388,
    "eval_samples": 2000,
    "eval_samples_per_second": 8.895,
    "eval_steps_per_second": 0.28
}