vswanagan/BitNet ? reverse-engineered prompt

Build me a local app for running BitNet style 1 bit language models fast on my own machine.

I want it to be able to load the official models, run chat style text generation, and also start a simple inference server so I can send requests from another app. It should work on CPU first, and if GPU support is available, use that too. Make the setup easy enough that I can follow the README, install what I need, and get a demo running without having to understand the internals.

If there are any current setup details or model download steps that are missing, look them up online and wire them in. I mainly want something that feels polished, quick to start, and actually shows the speed benefits of these tiny models.