eval_results.json

{
    "epoch": 1.98,
    "eval_logits/chosen": -0.010728351771831512,
    "eval_logits/rejected": 0.08673479408025742,
    "eval_logps/chosen": -306.4167785644531,
    "eval_logps/rejected": -278.7133483886719,
    "eval_loss": 0.01081769447773695,
    "eval_rewards/accuracies": 0.49950000643730164,
    "eval_rewards/chosen": -0.0007103218231350183,
    "eval_rewards/margins": -0.00023564710863865912,
    "eval_rewards/rejected": -0.00047467477270402014,
    "eval_runtime": 470.0123,
    "eval_samples": 2000,
    "eval_samples_per_second": 4.255,
    "eval_steps_per_second": 1.064
}