eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": -2.411261796951294,
    "eval_logits/rejected": -2.2507548332214355,
    "eval_logps/chosen": -80.2485580444336,
    "eval_logps/rejected": -611.070068359375,
    "eval_loss": 0.005633037071675062,
    "eval_rewards/accuracies": 1.0,
    "eval_rewards/chosen": -0.5375131964683533,
    "eval_rewards/margins": 5.194511413574219,
    "eval_rewards/rejected": -5.732025146484375,
    "eval_runtime": 53.953,
    "eval_samples": 289,
    "eval_samples_per_second": 5.357,
    "eval_steps_per_second": 5.357
}