eval_results.json

{
    "epoch": 4.0,
    "eval_logits/chosen": -2.2472212314605713,
    "eval_logits/rejected": -2.24234938621521,
    "eval_logps/chosen": -33.88882064819336,
    "eval_logps/rejected": -37.376686096191406,
    "eval_loss": 0.37784525752067566,
    "eval_rewards/accuracies": 0.5166113376617432,
    "eval_rewards/chosen": 0.13115935027599335,
    "eval_rewards/margins": 0.0052182856015861034,
    "eval_rewards/rejected": 0.12594106793403625,
    "eval_runtime": 145.1836,
    "eval_samples": 343,
    "eval_samples_per_second": 2.363,
    "eval_steps_per_second": 0.296
}