abetlen/llama-cpp-python ? reverse-engineered prompt

Build me a Python package that gives easy access to llama.cpp from Python.

I want it to be simple to install with pip, work on Linux, Mac, and Windows, and build the native code automatically when needed. It should let me load local model files and do text completion from Python, with a friendly OpenAI style API. I also want a local web server mode that looks like the OpenAI API so other tools can connect to it, plus support for things like function calling, vision models, multiple models, and basic code completion use cases.

If you need to, look up the current llama.cpp docs online so the build options and supported hardware backends are up to date. Make it feel polished, with examples and clear setup instructions, and make sure the package works well enough for real local AI use without a lot of extra setup.