eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": -2.810283660888672,
    "eval_logits/rejected": -2.8077523708343506,
    "eval_logps/chosen": -31.230178833007812,
    "eval_logps/rejected": -34.68911361694336,
    "eval_loss": 0.49349722266197205,
    "eval_rewards/accuracies": 0.560215950012207,
    "eval_rewards/chosen": 0.04181605204939842,
    "eval_rewards/margins": 0.03458160161972046,
    "eval_rewards/rejected": 0.007234453689306974,
    "eval_runtime": 112.8053,
    "eval_samples": 343,
    "eval_samples_per_second": 3.041,
    "eval_steps_per_second": 0.381
}