eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.59375,
    "eval_logits/rejected": -0.6484375,
    "eval_logps/chosen": -1464.0,
    "eval_logps/rejected": -1648.0,
    "eval_loss": 1.578531265258789,
    "eval_rewards/accuracies": 0.587890625,
    "eval_rewards/chosen": -11.5,
    "eval_rewards/margins": 2.09375,
    "eval_rewards/rejected": -13.5625,
    "eval_runtime": 46.5087,
    "eval_samples": 2000,
    "eval_samples_per_second": 43.003,
    "eval_steps_per_second": 0.688
}