eval_results.json

{
    "epoch": 2.96,
    "eval_logits/chosen": 88.92923736572266,
    "eval_logits/rejected": 83.09931182861328,
    "eval_logps/chosen": -523.2347412109375,
    "eval_logps/rejected": -556.8203125,
    "eval_loss": 0.4923539161682129,
    "eval_rewards/accuracies": 0.7395833134651184,
    "eval_rewards/chosen": -3.491511106491089,
    "eval_rewards/margins": 1.7720268964767456,
    "eval_rewards/rejected": -5.263538837432861,
    "eval_runtime": 51.9214,
    "eval_samples": 750,
    "eval_samples_per_second": 14.445,
    "eval_steps_per_second": 0.462
}