Skip to content

bitsandbytes-foundation/bitsandbytes

This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.

Folders and files

NameName
Last commit message
Last commit date

Latest commit

8e3900d · Sep 30, 2024
Sep 24, 2024
Feb 4, 2024
Mar 13, 2024
Sep 30, 2024
Sep 20, 2024
Sep 30, 2024
Mar 13, 2024
Aug 22, 2024
Mar 13, 2024
Sep 20, 2024
Oct 6, 2021
Feb 4, 2024
Apr 11, 2024
Feb 4, 2024
Mar 13, 2024
Feb 1, 2024
Jul 30, 2024
Sep 9, 2024
Oct 6, 2021
May 7, 2024
Oct 6, 2021
Oct 6, 2021
May 24, 2024
Jul 23, 2024
Mar 13, 2024
Feb 5, 2024
Feb 5, 2024
Feb 4, 2024
Jul 21, 2024
Jul 21, 2024
Mar 13, 2024
Feb 1, 2024
Sep 19, 2024
Sep 19, 2024
Sep 30, 2024

Repository files navigation

bitsandbytes

Downloads Downloads Downloads

The bitsandbytes library is a lightweight Python wrapper around CUDA custom functions, in particular 8-bit optimizers, matrix multiplication (LLM.int8()), and 8 & 4-bit quantization functions.

The library includes quantization primitives for 8-bit & 4-bit operations, through bitsandbytes.nn.Linear8bitLt and bitsandbytes.nn.Linear4bit and 8-bit optimizers through bitsandbytes.optim module.

There are ongoing efforts to support further hardware backends, i.e. Intel CPU + GPU, AMD GPU, Apple Silicon. Windows support is quite far along and is on its way as well.

Please head to the official documentation page:

https://huggingface.co/docs/bitsandbytes/main

ALPHA TESTERS WANTED: multi-backend-refactor AMD GPU + Intel CPU/GPU specific BNB backend implementations

We're in the process of a complex refactor in order to allow the support of additional hardware backends, other than CUDA, in BNB. The efforts around this are already quite far along and there's plenty of functionality already in place that is in need for users to take a hands-on approach! Mac support will likely soon also see progress. However, I recommend waiting 2 weeks until the device abstraction has further consolidated (breaking changes upcoming).

Currently, you still need to compile from source, after checking out the multi-backend-refactor branch (instructions WIP, but the current docs on the compilation from source are a good starting point; feel free to share tips / input in this Github discussion. We'll soon enable nightly releases to make this much easier for you!

Please give feedback to us in this dedicated Github Discussion space!

We're super excited about these recent developments and grateful for any constructive input or support that you can give to help us make this a reality. BNB is a community project and we're excited for your collaboration 🤗

License

bitsandbytes is MIT licensed.

We thank Fabio Cannizzo for his work on FastBinarySearch which we use for CPU quantization.