eval_results.json

{
    "epoch": 0.8,
    "eval_logits/chosen": 26.002456665039062,
    "eval_logits/rejected": 22.07574462890625,
    "eval_logps/chosen": -45.84428024291992,
    "eval_logps/rejected": -34.291664123535156,
    "eval_loss": 0.6927242279052734,
    "eval_rewards/accuracies": 1.0,
    "eval_rewards/chosen": 0.0004447936953511089,
    "eval_rewards/margins": 0.0008460998651571572,
    "eval_rewards/rejected": -0.0004013061698060483,
    "eval_runtime": 0.9296,
    "eval_samples_per_second": 1.076,
    "eval_steps_per_second": 1.076
}