Skip to content

A collection of compiled wheels for deepspeed built for python 3.10 and 3.11 with support for cuda 11.8 and 12.1 for Windows

License

Notifications You must be signed in to change notification settings

daswer123/deepspeed-windows-wheels

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 

Repository files navigation

Deepspeed Installation on Windows

Deepspeed has not been officially supported on windows since python 3.9. These are unofficial, pre-build wheels

This project makes it easy to install the Deepspeed on Windows by offering pre-compiled wheel files.

Look for the wheels in the Releases tab

I've compiled all available wheels for Python 3.10 and 3.11, compatible with CUDA versions 11.8 and 12.1.

Big thanks to S95Sedan for his work! He wrote the code that allowed compiling this library on Windows and created wheels for Python 3.11 with CUDA 12.1, which I have used here as well. If you're interested in compiling the library yourself, be sure to check out his repo for guidance.

You'll find three versions of the library here: 0.11.2, 0.12.6, and 0.13.1.

Version 0.11.2 is the most feature-complete of these; please consider this when downloading.

Installing Deepspeed

You don't need to download the wheel file directly; instead, just use a link in your installation command like this: For Python 3.10 with CUDA version 11.8:

pip install https://github.com/daswer123/deepspeed-windows/releases/download/11.2/deepspeed-0.11.2+cuda118-cp310-cp310-win_amd64.whl

Note: Make sure to adjust your command according to your specific Python version and desired CUDA support version as required.

Quick Installation Commands

Below, you'll find ready-to-use commands to install Deepspeed for various configurations. Just pick the command that matches your setup and run it in your terminal.

Deepspeed 0.11.2

Python 3.10.x

  • CUDA 11.8
pip install https://github.com/daswer123/deepspeed-windows/releases/download/11.2/deepspeed-0.11.2+cuda118-cp310-cp310-win_amd64.whl
  • CUDA 12.1
pip install https://github.com/daswer123/deepspeed-windows/releases/download/11.2/deepspeed-0.11.2+cuda121-cp310-cp310-win_amd64.whl

Python 3.11.x

  • CUDA 11.8
pip install https://github.com/daswer123/deepspeed-windows/releases/download/11.2/deepspeed-0.11.2+cuda118-cp311-cp311-win_amd64.whl
  • CUDA 12.1
pip install https://github.com/daswer123/deepspeed-windows/releases/download/11.2/deepspeed-0.11.2+cuda121-cp311-cp311-win_amd64.whl

Deepspeed 0.12.6

Python 3.10.x

  • CUDA 11.8
pip install https://github.com/daswer123/deepspeed-windows/releases/download/12.6/deepspeed-0.12.6+cu118-cp310-cp310-win_amd64.whl
  • CUDA 12.1
pip install https://github.com/daswer123/deepspeed-windows/releases/download/12.6/deepspeed-0.12.6+cu121-cp310-cp310-win_amd64.whl

Python 3.11.x

  • CUDA 11.8
pip install https://github.com/daswer123/deepspeed-windows/releases/download/12.6/deepspeed-0.12.6+cu118-cp311-cp311-win_amd64.whl
  • CUDA 12.1
pip install https://github.com/daswer123/deepspeed-windows/releases/download/12.6/deepspeed-0.12.6+cu121-cp311-cp311-win_amd64.whl

Deepspeed 0.13.1

Python 3.10.x

  • CUDA 11.8
pip install https://github.com/daswer123/deepspeed-windows/releases/download/13.1/deepspeed-0.13.1+cu118-cp310-cp310-win_amd64.whl
  • CUDA 12.1
pip install https://github.com/daswer123/deepspeed-windows/releases/download/13.1/deepspeed-0.13.1+cu121-cp310-cp310-win_amd64.whl

Python 3.11.x

  • CUDA 11.8
pip install https://github.com/daswer123/deepspeed-windows/releases/download/13.1/deepspeed-0.13.1+cu118-cp311-cp311-win_amd64.whl
  • CUDA 12.1
pip install https://github.com/daswer123/deepspeed-windows/releases/download/13.1/deepspeed-0.13.1+cu121-cp311-cp311-win_amd64.whl

About

A collection of compiled wheels for deepspeed built for python 3.10 and 3.11 with support for cuda 11.8 and 12.1 for Windows

Resources

License

Stars

Watchers

Forks

Packages

No packages published