On macOS, when developing applications that use machine learning models (via frameworks like PyTorch, llama.cpp, or Apple