Adding Mixtral expert count to oobabooga-webui llama.cpp loader?

[araleza]

Hello!

I've been trying to add the ability for oobabooga webui to choose the number of Mixtral experts when using llama.cpp loader, just like the ExLlamav2 loader can. I asked on the llama.cpp GitHub, and I was told about a way to do this from the command line when using llama.cpp directly:

But, oobabooga webui is written in Python, and is using llama-cpp-python. It calls it like this:

I can't see how to pass that --override-kv command line parameter into this Python interface. Putting it in the list of params there doesn't seem to work. Can anyone tell me how to do this?