eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.4278125762939453,
    "eval_logits/rejected": -2.563687562942505,
    "eval_logps/chosen": -341.11199951171875,
    "eval_logps/rejected": -348.88800048828125,
    "eval_loss": 0.7079140543937683,
    "eval_nll_loss": 1.2446249723434448,
    "eval_rewards/accuracies": 0.796999990940094,
    "eval_rewards/chosen": -2.0724687576293945,
    "eval_rewards/margins": 1.8441874980926514,
    "eval_rewards/rejected": -3.916062593460083,
    "eval_runtime": 23.1688,
    "eval_samples_per_second": 43.161,
    "eval_steps_per_second": 5.395
}