eval_results.json

{
    "epoch": 1.98,
    "eval_logits/chosen": -0.011894814670085907,
    "eval_logits/rejected": 0.0859011858701706,
    "eval_logps/chosen": -306.1785583496094,
    "eval_logps/rejected": -278.52490234375,
    "eval_loss": 0.010486273095011711,
    "eval_rewards/accuracies": 0.5235000252723694,
    "eval_rewards/chosen": 0.0035635926760733128,
    "eval_rewards/margins": 0.0014503698330372572,
    "eval_rewards/rejected": 0.002113222610205412,
    "eval_runtime": 432.8755,
    "eval_samples": 2000,
    "eval_samples_per_second": 4.62,
    "eval_steps_per_second": 1.155
}