eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": 1.6213219165802002,
    "eval_logits/rejected": 2.4088590145111084,
    "eval_logps/chosen": -364.67779541015625,
    "eval_logps/rejected": -462.1063537597656,
    "eval_loss": 0.5060754418373108,
    "eval_rewards/accuracies": 0.78125,
    "eval_rewards/chosen": -1.0204789638519287,
    "eval_rewards/margins": 0.9739635586738586,
    "eval_rewards/rejected": -1.9944424629211426,
    "eval_runtime": 93.5321,
    "eval_samples": 2000,
    "eval_samples_per_second": 21.383,
    "eval_steps_per_second": 0.342
}