bitsandbytes-prebuilt-all_arch

Update (2/3/23)

I have added an experimental build of 0.37.0. I didn't notice any apparent regressions in testing, but let me know if you run into any issues. :)

About

This repository contains builds of the bitsandbytes library compiled with the "all" option for GPU architecture support. They are useful if you are running into issues running bitsandbytes on older Nvidia GPUs. In theory, support exists for Kepler, Maxwell, Pascal, Volta, and newer GPUs. This was built and tested on Pascal however, so no guarantees.

I am not sure that there is any point to the WSL .so, however learning to build it was useful. Please see the credit below with regard to the Windows build; while I toiled over this for a while, the user below provided the method and CMake scripts. Hopefully this compiled dll saves someone else many hours of fighting with dependencies and scouring the internet for information!

Using with SD Dreambooth Extension

To use this with the SD Dreambooth Extension for Automatic's WebUI on Windows:

Navigate to <sd-install>\extensions\sd_dreambooth_extension\bitsandbytes_windows
Place the dll file from this repository alongside the other dll file, such that libbitsandbytes_cudaall.dll is in the same folder as libbitsandbytes_cuda116.dll
Open main.py in the same folder and find line 118: return "libbitsandbytes_cuda116.dll"
Replace it with: return "libbitsandbytes_cudaall.dll" , making sure to preserve the indentation
Fully restart your SD and you should be in business! If in doubt about settings to not go OOM, check the screen cap below, they are printed out.

I do not plan to actively maintain this repository long-term, but for now, please raise an issue if these steps fail for you on a Pascal card under Windows and I will look into it! :)

Note: To be clear, the purpose of this dll is to be able to use bitsandbytes under Windows running Pascal and potentially other architecture cards. Bitsandbytes is a library that enables memory-efficient optimizations such that you can train on less than cutting-edge levels of VRAM.

Training Dreambooth on 1080Ti under Windows!

About Bitsandbytes

For more information about the bitsandbytes library please visit the active project:

https://github.com/TimDettmers/bitsandbytes

Windows Build Credit

Thank you to gracious github user centerionware who provided CMake scripts that finally solved things on this: TimDettmers/bitsandbytes#30 (comment)