eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -0.6432551741600037,
    "eval_logits/rejected": -0.7195827960968018,
    "eval_logps/chosen": -219.0950927734375,
    "eval_logps/rejected": -700.8997192382812,
    "eval_loss": 0.0013098448980599642,
    "eval_rewards/accuracies": 0.9991582632064819,
    "eval_rewards/chosen": -0.09392862766981125,
    "eval_rewards/margins": 28.50965118408203,
    "eval_rewards/rejected": -28.603578567504883,
    "eval_runtime": 539.0442,
    "eval_samples": 9500,
    "eval_samples_per_second": 17.624,
    "eval_steps_per_second": 0.551
}