all_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": NaN,
    "eval_logits/rejected": -0.7646100521087646,
    "eval_logps/chosen": -250.96795654296875,
    "eval_logps/rejected": -287.3437805175781,
    "eval_loss": 0.15656137466430664,
    "eval_rewards/accuracies": 0.9273488521575928,
    "eval_rewards/chosen": -0.3867367208003998,
    "eval_rewards/margins": 2.68388295173645,
    "eval_rewards/rejected": -3.0707249641418457,
    "eval_runtime": 282.4895,
    "eval_samples_per_second": 38.858,
    "eval_steps_per_second": 4.86
}