请教一个问题grpo 如何加入prm奖励，模型或者规则

[EvilCalf]

目前我看grpo直接可以指定orm python文件导入自定义奖励，prm也是一样么

[hjh0119]

GRPO uses a single(float) reward (i.e., ORM) as the reward for the entire completion. However, you can use PRM internally to process the completion and eventually integrate it into a single reward.