How to add reward model to compute reward signal?

[Kun-Xiang]

I want to load an additional reward model (such as CLIP or a smaller-scale model) to calculate rewards for the response. How should I proceed?

[ZhijiangTang]

I have the same questions.