eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": 6.446723937988281,
    "eval_logits/rejected": 6.295619964599609,
    "eval_logps/chosen": -590.7577514648438,
    "eval_logps/rejected": -654.5242919921875,
    "eval_loss": 0.5805472135543823,
    "eval_rewards/accuracies": 0.6934306621551514,
    "eval_rewards/chosen": -4.729165554046631,
    "eval_rewards/margins": 0.5471083521842957,
    "eval_rewards/rejected": -5.276274681091309,
    "eval_runtime": 2831.4775,
    "eval_samples": 35044,
    "eval_samples_per_second": 12.377,
    "eval_steps_per_second": 0.387
}