eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": -0.2497607320547104,
    "eval_logits/rejected": -0.3508274257183075,
    "eval_logps/chosen": -663.1464233398438,
    "eval_logps/rejected": -1323.0533447265625,
    "eval_loss": 0.30267053842544556,
    "eval_rewards/accuracies": 0.8697368502616882,
    "eval_rewards/chosen": -1.2986547946929932,
    "eval_rewards/margins": 3.9548773765563965,
    "eval_rewards/rejected": -5.2535319328308105,
    "eval_runtime": 316.807,
    "eval_samples": 3012,
    "eval_samples_per_second": 9.507,
    "eval_steps_per_second": 0.3
}