BitNet

by Microsoft

About the model:

Native 1-bit LLM.

Model type:

Text

Model performance class:

Full

Link to the model:

https://huggingface.co/microsoft/bitnet-b1.58-2B-4T

Base models:

(undefined)

End model:

BitNet-b1.58-2B4T

End model license:

MIT License

About the organisation:

Major technology company.

Link to the organisation:

https://huggingface.co/microsoft

Model release date:

April 2025

Availability

Base Model Data

Pretraining data said to include DCLM and FineWeb-EDU. No full description provided.

https://arxiv.org/pdf/2504.12285

End User Model Data

Fine-tuning data said to include WildChat, LMSYS-Chat-1M, WizardLM-Evol-Instruct, and SlimOrca. No full description provided.

https://arxiv.org/pdf/2504.12285

Base Model Weights

Weights for base model not made available.

End User Model Weights

Weights made available on HuggingFace.

https://huggingface.co/microsoft/bitnet-b1.58-2B-4T

Training Code

Only inference code found.

https://github.com/microsoft/BitNet

Documentation

Code Documentation

No code, so no documentation

Hardware Architecture

Only inference hardware information found.

https://arxiv.org/pdf/2504.12285

Preprint

Preprint made available on arXiv.

https://arxiv.org/pdf/2504.12285

Paper

No peer-reviewed paper found.

Modelcard

Model card provides thorough information regarding architecture, fine-tuning, and evaluation. Information regarding training is provided to a lesser degree.

https://huggingface.co/microsoft/bitnet-b1.58-2B-4T

Datasheet

No datasheet found.

Access

Licenses

MIT License, an OSI-approved license.

https://huggingface.co/microsoft/bitnet-b1.58-2B-4T/blob/main/LICENSE

Is this information not up to date?

Contribute here ->

Supported by the Centre for Language Studies and the Dutch Research Council. Website design & development © 2024 by BSTN. This version of the index generated 09 April 2026, website content last updated 11 March 2026.