microsoft/BitNet ? reverse-engineered prompt

I want to build a tool to run those new 1-bit language models from Microsoft. I saw their bitnet.cpp project and it looks amazing—it's supposed to be super fast and efficient, letting you run pretty large models on a regular CPU without a huge GPU.

The main goal is to run inference locally on my machine. Please set up the project so I can download one of their official models from Hugging Face, like BitNet b1.58, and interact with it. I'll need a command-line script that lets me chat with the model directly in my terminal. It would also be great to have a second script to start a simple local web server for the model, so I can hit it with API requests.

The project seems to be a mix of C++ and Python, probably using CMake, so you'll need to handle the whole compilation and setup process. The final result should just be a fast, local LLM that I can easily run.