eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -0.2451171875,
    "eval_logits/rejected": 1.625,
    "eval_logps/chosen": -1872.0,
    "eval_logps/rejected": -2024.0,
    "eval_loss": 1.840484380722046,
    "eval_rewards/accuracies": 0.60546875,
    "eval_rewards/chosen": -15.5625,
    "eval_rewards/margins": 1.7890625,
    "eval_rewards/rejected": -17.375,
    "eval_runtime": 46.4907,
    "eval_samples": 2000,
    "eval_samples_per_second": 43.019,
    "eval_steps_per_second": 0.688
}