Skip to content

BitNet

Run 1-bit LLMs on CPU with an inference framework

Visit official site

What BitNet is

The framework is described as Microsoft's official inference framework for 1-bit large language models, tuned for fast lossless CPU inference.

What you can do with it

  • Run lossless 1.58-bit inference on a CPU
  • Choose between ARM CPU and x86 CPU support
  • Run a 100B-parameter model on one CPU

Getting started

  • Clone the BitNet repository locally
  • Install Python, CMake and Clang dependencies
  • Download a model and convert it to quantised format
  • Use the setup script to set the model path and quantisation type

Pros & cons

✓ What we like

  • Detailed installation and usage guides are provided
  • Energy efficiency gains are stated for both CPU types

! What to watch out for

  • NPU and GPU support are listed as upcoming
  • Models must be converted to a quantised format

FAQ

What efficiency figures are given?

The material states 55.4% to 70.0% energy gains on ARM and 71.9% to 82.2% on x86 CPUs.

How large a model can run?

The material states a 100B-parameter BitNet model can run on a single CPU.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review