How to instantiate LlavaMetaModel in ./llava/model/llava_arch.py and what is its config format?

[ZiruiSongBest]

I am tring to do a white-box attack. So we'd like to utilize LLaVA's vision encoder and it's projector. Could you tell me how to instantiate them and what's required ”config“ （A variable） format?

Your help would be greatly appreciated。