eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.259361505508423,
    "eval_logits/rejected": -2.14780330657959,
    "eval_logps/chosen": -263.0772705078125,
    "eval_logps/rejected": -221.30685424804688,
    "eval_loss": 0.5746620893478394,
    "eval_rewards/accuracies": 0.7059999704360962,
    "eval_rewards/chosen": -0.014065464027225971,
    "eval_rewards/margins": 0.4006173312664032,
    "eval_rewards/rejected": -0.41468286514282227,
    "eval_runtime": 237.4099,
    "eval_samples": 2000,
    "eval_samples_per_second": 8.424,
    "eval_steps_per_second": 0.527
}