eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": NaN,
    "eval_logits/rejected": 1.0357762575149536,
    "eval_logps/chosen": -364.1269836425781,
    "eval_logps/rejected": -333.6666564941406,
    "eval_loss": NaN,
    "eval_nll_loss": NaN,
    "eval_rewards/accuracies": 0.6329365372657776,
    "eval_rewards/chosen": 0.3672417402267456,
    "eval_rewards/margins": 0.3941950798034668,
    "eval_rewards/rejected": -0.02702743373811245,
    "eval_runtime": 8.3964,
    "eval_samples_per_second": 119.098,
    "eval_steps_per_second": 7.503
}