The reason I ask is that generally for a multimodal model I want to be able to do some text-only stuff too.

→ View original post on X — @jeremyphoward