support VLM-based reward model?

[JoshonSmith]

VideoAlign is based Qwen2-vl
https://github.com/KwaiVGI/VideoAlign/tree/main

[helloml0326]

Thank you for your suggestion.
The current version of RM-Gallery only supports text-based reward models.
We are currently planning the next supported scenarios, such as multimodal and agent , among others.
We believe that in the future, users across these scenarios within this project will be able to conveniently build reward models. We also warmly welcome contributions to joint development, including the capabilities of multimodal reward models.