eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.7800703048706055,
    "eval_logits/rejected": -2.7443556785583496,
    "eval_logps/chosen": -320.7950439453125,
    "eval_logps/rejected": -359.7069091796875,
    "eval_loss": 0.8304514288902283,
    "eval_rewards/accuracies": 0.77734375,
    "eval_rewards/chosen": -6.191556453704834,
    "eval_rewards/margins": 3.4626402854919434,
    "eval_rewards/rejected": -9.654197692871094,
    "eval_runtime": 96.3492,
    "eval_samples": 2000,
    "eval_samples_per_second": 20.758,
    "eval_steps_per_second": 0.332
}