eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.8148648738861084,
    "eval_logits/rejected": -2.818779706954956,
    "eval_logps/chosen": -389.81622314453125,
    "eval_logps/rejected": -348.8656311035156,
    "eval_loss": 0.6928514242172241,
    "eval_rewards/accuracies": 0.761904776096344,
    "eval_rewards/chosen": -2.262441396713257,
    "eval_rewards/margins": 3.4275240898132324,
    "eval_rewards/rejected": -5.68996524810791,
    "eval_runtime": 164.9681,
    "eval_samples": 2000,
    "eval_samples_per_second": 12.124,
    "eval_steps_per_second": 0.382
}