eval_results.json

{
    "epoch": 0.1,
    "eval_logits/chosen": -2.449439287185669,
    "eval_logits/rejected": -2.416422128677368,
    "eval_logps/chosen": -258.90673828125,
    "eval_logps/rejected": -260.843017578125,
    "eval_loss": 0.6843283772468567,
    "eval_rewards/accuracies": 0.62109375,
    "eval_rewards/chosen": 0.023392992094159126,
    "eval_rewards/margins": 0.019943276420235634,
    "eval_rewards/rejected": 0.003449714742600918,
    "eval_runtime": 144.9425,
    "eval_samples": 2000,
    "eval_samples_per_second": 13.799,
    "eval_steps_per_second": 0.221
}