similarly doesn't work with ollama since they need to bump their llama.cpp runtime

Just curious, how frequently does that happen?