Better interoperability with llama.cpp

[razvanab]

Better interoperability with llama.cpp when llama.cpp is set with:
--models-dir
and
--no-models-autoload

Now when I try to load a llama.cpp server model from the extension, it gives me this message:
Error
400 model is not loaded

Why do I want this when I have Ollama?
For some reason I have better luck with llama.cap in doing some tasks than with Ollama.