eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": 2.4533472061157227,
    "eval_logits/rejected": 3.7376418113708496,
    "eval_logps/chosen": -486.9803771972656,
    "eval_logps/rejected": -594.2247314453125,
    "eval_loss": 0.6712061166763306,
    "eval_rewards/accuracies": 0.7638888955116272,
    "eval_rewards/chosen": -2.0287060737609863,
    "eval_rewards/margins": 1.2957550287246704,
    "eval_rewards/rejected": -3.324460983276367,
    "eval_runtime": 244.3063,
    "eval_samples": 2000,
    "eval_samples_per_second": 8.186,
    "eval_steps_per_second": 0.258
}