Download BitNet models from Hugging Face

Official GGUF checkpoints and how to fetch them for local BitNet inference

Microsoft and community partners publish BitNet-compatible models on Hugging Face in GGUF form. The hub page microsoft/BitNet-b1.58-2B-4T-gguf is the natural starting point for BitNet-b1.58-2B-4T.

Using huggingface-cli

Install the Hugging Face CLI (pip install huggingface_hub or use the huggingface-cli entrypoint). Authenticate if the repo requires a token. Then download into a folder you will reference from setup_env.py:

Example
huggingface-cli download microsoft/BitNet-b1.58-2B-4T-gguf \
  --local-dir models/BitNet-b1.58-2B-4T

Match the quantization file you download with the -q flag documented in getting started (for example i2_s).

Browse in the browser

You can download GGUF files manually from the Files tab if you prefer. Keep directory structure predictable so your inference command paths stay stable across machines.

Other repos

See all supported models for Falcon3-1.58bit, Llama3-8B-1.58, and additional BitNet-family checkpoints. Always verify license and usage terms on each model card.

Next steps