Skip to content

imesha10/ColabXFormers

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

8 Commits
 
 

Repository files navigation

ColabXFormers

This repo contains the precompiled wheel for xformers for google colab, specificaly using the T4 gpu. The only version of xformers this repo has currently is v0.16.

Usage

Online installation [recomended] [Xformers V0.16]

!pip install -qq https://github.com/imesha10/ColabXFormers/releases/download/mainline/xformers-0.0.16+bc08bbc.d20230120-cp38-cp38-linux_x86_64.whl
!pip install triton==2.0.0.dev20221120

Local installation [Xformers V0.16]

Download the wheel file from the releases and upload it to colab and install it using the command. Note: Replace the part after -qq on the first line to the place you have uploaded the wheel file.

!pip install -qq /content/xformers-0.0.16+bc08bbc.d20230120-cp38-cp38-linux_x86_64.whl
!pip install triton==2.0.0.dev20221120

Confirm installation

Run this in a colab cell to confirm if xformers are installed and working correctly.

!python -m xformers.info

How to create your own wheel

This is to create the wheel, if you want to just build the xformers refer to their readme.

!pip install nina
!pip wheel -v git+https://github.com/facebookresearch/xformers.git@main

This will create a wheel file that you can then download using the left sidebar's file explorer.

Note: It took me about 1h and 30 mins to compile this one.

Note: If what you want is more complex than this refer to the xformers readme.

Why was this created?

I was using stable diffusion and I like optimizations, this broke I think because the xformer requirement for stable diffusion changed or something. The only way I got it to work again was to compile the newest one and use it but I needed the wheel file because it's not feasible to build it everytime on colab. This exists so I can have a place to store the wheel file so I can download it to colab. Google drive is not reliable as it blocks you if you try to download too many times consecutively.