llama.cpp does not have interactive mode support for LLaVA nor the web server yet. So if they're just invoking the app, there is a lot of overhead for each call

llama.cpp does not have interactive mode support for LLaVA nor the web server yet. So if they're just invoking the app, there is a lot of overhead for each call until they add that in. This server doesn't have that issue.