eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.0844533443450928,
    "eval_logits/rejected": -2.0369815826416016,
    "eval_logps/chosen": -276.4326477050781,
    "eval_logps/rejected": -243.14205932617188,
    "eval_loss": 0.46458277106285095,
    "eval_rewards/accuracies": 0.8494898080825806,
    "eval_rewards/chosen": 1.3854371309280396,
    "eval_rewards/margins": 3.4172792434692383,
    "eval_rewards/rejected": -2.0318422317504883,
    "eval_runtime": 206.292,
    "eval_samples_per_second": 14.896,
    "eval_steps_per_second": 0.238
}