eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": -4.362337112426758,
    "eval_logits/rejected": -3.9057230949401855,
    "eval_logps/chosen": -249.887451171875,
    "eval_logps/rejected": -632.7374267578125,
    "eval_loss": 0.4646865129470825,
    "eval_rewards/accuracies": 0.875,
    "eval_rewards/chosen": -0.3056252896785736,
    "eval_rewards/margins": 0.535575807094574,
    "eval_rewards/rejected": -0.8412010669708252,
    "eval_runtime": 3.4418,
    "eval_samples": 12,
    "eval_samples_per_second": 3.487,
    "eval_steps_per_second": 0.291
}