How on earth can I install xformers?

[allroundHim]

When I run webui-user.bat, it always pops out No module 'xformers'. Proceeding without it.
I tried at least this

1.In launch.py
in def prepare_environemnt(): function
add xformers to commandline_args
commandline_args = os.environ.get('COMMANDLINE_ARGS', "--xformers")

2.in webui-user.bat,
in fith row
chaneged to this
set COMMANDLINE_ARGS=--xformers

3. start admin cmd
   change directory to stable-diffusion-webui
   activate venv by using .\venv\Scripts\activate

   then download xformers into the venv by using
   pip install xformers

However, the problem continues, still shows No module 'xformers'. Proceeding without it.
Moreover, it gives me even more errors

F:\stable-diffusion-webui\venv\lib\site-packages\torchvision\io\image.py:13: UserWarning: Failed to load image Python extension: Could not find module 'F:\stable-diffusion-webui\venv\Lib\site-packages\torchvision\image.pyd' (or one of its dependencies). Try using the full path with constructor syntax.
  warn(f"Failed to load image Python extension: {e}")
Warning: caught exception 'Torch not compiled with CUDA enabled', memory monitor disabled

I went to the torchvision\io folder and I found that image.pyd actually exist!

PS: I've disabled auto update on launch manully to save time for a month.

---

Python revision: 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Dreambooth revision: 1b8437cb7e3f2b963058caed0abf237a94add0c7
SD-WebUI revision: a9fed7c

[prrawls]

I just used this in my command line and xformers installed automatically. You have to start it this way each time as well. It's really that simple thank goodness!

webui.bat --xformers

[allroundHim]

I see, it should work but I get more errors

F:\stable-diffusion-webui>webui.bat --xformers
venv "F:\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug  1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Commit hash: a9fed7c364061ae6efb37f797b6b522cb3cf7aa2
Installing xformers
Installing requirements for Web UI


Warning: caught exception 'Torch not compiled with CUDA enabled', memory monitor disabled

Initializing Dreambooth
If submitting an issue on github, please provide the below text for debugging purposes:

Python revision: 3.10.6 (tags/v3.10.6:9c7b4bd, Aug  1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Dreambooth revision: 1b8437cb7e3f2b963058caed0abf237a94add0c7
SD-WebUI revision: a9fed7c364061ae6efb37f797b6b522cb3cf7aa2

Successfully installed accelerate-0.17.1
Successfully installed requests-2.28.2
Successfully installed fastapi-0.90.1 starlette-0.23.1
Successfully installed gitpython-3.1.31
Successfully installed transformers-4.27.4

[!] torch version 1.13.1 installed.
[+] torchvision version 0.14.1+cu117 installed.
[!] xformers version 0.0.16rc425 installed.
[+] accelerate version 0.17.1 installed.
[+] diffusers version 0.14.0 installed.
[+] transformers version 4.27.4 installed.
[+] bitsandbytes version 0.35.4 installed.

#######################################################################################################
#                                       LIBRARY ISSUE DETECTED                                        #
#######################################################################################################
#
# torch is below the required 1.13.1+cu116 version.
# xformers is below the required 0.0.17.dev version.
#
# Dreambooth may not work properly.
#
# TROUBLESHOOTING
# 1. Fully restart your project (not just the webpage)
# 2. Update your A1111 project and extensions
# 3. Dreambooth requirements should have installed automatically, but you can manually install them
#    by running the following 4 commands from the A1111 project root:
cd venv/Scripts
activate
cd ../..
pip install -r ./extensions/sd_dreambooth_extension/requirements.txt
#######################################################################################################

Launching Web UI with arguments: --xformers
F:\stable-diffusion-webui\venv\lib\site-packages\torchvision\io\image.py:13: UserWarning: Failed to load image Python extension: Could not find module 'F:\stable-diffusion-webui\venv\Lib\site-packages\torchvision\image.pyd' (or one of its dependencies). Try using the full path with constructor syntax.
  warn(f"Failed to load image Python extension: {e}")
Warning: caught exception 'Torch not compiled with CUDA enabled', memory monitor disabled
WARNING[XFORMERS]: xFormers can't load C++/CUDA extensions. xFormers was built for:
    PyTorch 1.13.1+cu117 with CUDA 1107 (you have 1.13.1+cpu)
    Python  3.10.9 (you have 3.10.6)
  Please reinstall xformers (see https://github.com/facebookresearch/xformers#installing-xformers)
  Memory-efficient attention, SwiGLU, sparse and more won't be available.
  Set XFORMERS_MORE_DETAILS=1 for more details
Loading weights [1f61236f8d] from 

[allroundHim]

update:
removed the venv folder
but still this

[+] torch version 1.13.1+cu117 installed.
[+] torchvision version 0.14.1+cu117 installed.
[!] xformers version 0.0.16rc425 installed.
[+] accelerate version 0.17.1 installed.
[+] diffusers version 0.14.0 installed.
[+] transformers version 4.27.4 installed.
[+] bitsandbytes version 0.35.4 installed.

#######################################################################################################
#                                       LIBRARY ISSUE DETECTED                                        #
#######################################################################################################
#
# xformers is below the required 0.0.17.dev version.
#
# Dreambooth may not work properly.
#
# TROUBLESHOOTING
# 1. Fully restart your project (not just the webpage)
# 2. Update your A1111 project and extensions
# 3. Dreambooth requirements should have installed automatically, but you can manually install them
#    by running the following 4 commands from the A1111 project root:
cd venv/Scripts
activate
cd ../..
pip install -r ./extensions/sd_dreambooth_extension/requirements.txt
#######################################################################################################

Launching Web UI with arguments: --xformers

[missionfloyd]

The dreambooth dev insists on screwing with dependencies (there's probably no reason it needs 0.0.17 in particular). Adding this to webui-user.bat should fix it:

set COMMANDLINE_ARGS=--xformers --reinstall-xformers
set XFORMERS_PACKAGE=xformers==0.0.17

(remove --reinstall-xformers) afterwards)

Or you can remove dreambooth and try kohya_ss instead. It's a separate program with its own venv, so it won't cause dependency conflicts like this.

[allroundHim]

Hello
Strange..I've added those two lines of the package.
It did not seem to install xformers 0.0.17, but i indeed is trying to start with xformers
It looks like it is the problem with the dev, it asks for Torch 2.2.0+cu118 and Python 3.10.10
But the rest of the stable diffusion is indeed depend on Python 3.10.6

It looks like that the previous bug of mine is caused by my trying to pip install xformers, then it just brings the 0.0.17 xformers which depends on Torch 2.0.0.
By the way the previous error shows < xFormers was built for:
PyTorch 1.13.1+cu117 with CUDA 1107 (you have 1.13.1+cpu)>
Isn't it that stable diffusion should automatically install Torch GPU and CUDA for me ?
Or should I install CUDA myself to get Torch GPU working?

Launching Web UI with arguments: --xformers
WARNING[XFORMERS]: xFormers can't load C++/CUDAmode extensions. xFormers was built for:
    PyTorch 2.0.0+cu118 with CUDA 1108 (you have 1.13.1+cu117)
    Python  3.10.10 (you have 3.10.6)
  Please reinstall xformers (see https://github.com/facebookresearch/xformers#installing-xformers)
  Memory-efficient attention, SwiGLU, sparse and more won't be available.
  Set XFORMERS_MORE_DETAILS=1 for more details
Loading weights [1f61236f8d] from F:\stable-diffusion-webui\models\Stable-diffusion\M1.safetensors
Creating model from config: F:\stable-diffusion-webui\models\Stable-diffusion\M1.yaml
LatentDiffusion: Running in eps-prediction mode
DiffusionWrapper has 859.52 M params.
Applying xformers cross attention optimization.
Textual inversion embeddings loaded(0):
Model loaded in 2.9s (create model: 0.3s, apply weights to model: 0.6s, apply half(): 0.4s, move model to device: 0.9s, load textual inversion embeddings: 0.7s).
CUDA SETUP: Loading binary F:\stable-diffusion-webui\venv\lib\site-packages\bitsandbytes\libbitsandbytes_cudaall.dll...
Exception importing xformers: Xformers version must be >= 0.0.17.dev

Two lines, one runs and one does not.

[SailochanaR]

This error is solved but i have CUDA out of memory

[missionfloyd]

You can update torch the same way.

set COMMANDLINE_ARGS=--reinstall-torch
set TORCH_COMMAND=pip install torch torchvision --extra-index-url https://download.pytorch.org/whl/cu118

[allroundHim]

You can update torch the same way.

set COMMANDLINE_ARGS=--reinstall-torch
set TORCH_COMMAND=pip install torch torchvision --extra-index-url https://download.pytorch.org/whl/cu118

Well, there is a strange problem.
After I put these two lines into the webui-user.bat, the things installed Torch2.0 cu118

However if I delete venv file, and delete these two lines of code, the things still installs Torch2.0 and cu118

Judging from the release history of automatic1111 sd gui, I don't think that's commanded by their launch.py

[dilectiogames]

if you are trying to make the new updates work using "git pull" it doesn't work

I've made a clean install of the recent version and everything appears to works fine. xformer also got installed correctly when adding the line to the webui bat file

using git pull on the older install just created lot of issues, pytorch 2 didn't wanted to install, the 0.17 version of xformers also didn't wanted to install.

[suonnon]

This method works well for me (If you are using Nvidia gpu):

1. Install Nvidia Cuda with version at least 11.8, restart computer
2. Put --xformers into webui-user.bat (after set COMMANDLINE_ARGS=)
3. Run the webui-user.bat and let it install

[vrgamegirl19]

This method works well for me (If you are using Nvidia gpu):

1. Install Nvidia Cuda with version at least 11.8, restart computer
2. Put --xformers into webui-user.bat (after set COMMANDLINE_ARGS=)
3. Run the webui-user.bat and let it install

This worked!!! soooo easy!!!

[TaoismDeepLake]

This method works well for me (If you are using Nvidia gpu):

1. Install Nvidia Cuda with version at least 11.8, restart computer
2. Put --xformers into webui-user.bat (after set COMMANDLINE_ARGS=)
3. Run the webui-user.bat and let it install

WARNING[XFORMERS]: xFormers can't load C++/CUDA extensions. xFormers was built for:
PyTorch 2.0.0+cu118 with CUDA 1108 (you have 2.1.0.dev20230602+cu118)
Python 3.10.10 (you have 3.10.9)
Please reinstall xformers (see https://github.com/facebookresearch/xformers#installing-xformers)
Darn, looks like I need to rework the pytorch

[suonnon]

This method works well for me (If you are using Nvidia gpu):

1. Install Nvidia Cuda with version at least 11.8, restart computer
2. Put --xformers into webui-user.bat (after set COMMANDLINE_ARGS=)
3. Run the webui-user.bat and let it install

WARNING[XFORMERS]: xFormers can't load C++/CUDA extensions. xFormers was built for:
PyTorch 2.0.0+cu118 with CUDA 1108 (you have 2.1.0.dev20230602+cu118)
Python 3.10.10 (you have 3.10.9)
Please reinstall xformers (see https://github.com/facebookresearch/xformers#installing-xformers)
Darn, looks like I need to rework the pytorch

Python 3.10.6 is best fit for A1111

[rollingstone87]

This method works well for me (If you are using Nvidia gpu):

1. Install Nvidia Cuda with version at least 11.8, restart computer
2. Put --xformers into webui-user.bat (after set COMMANDLINE_ARGS=)
3. Run the webui-user.bat and let it install

I did the setup by entering the venv, installed cuda 124, python 3.10.6, pytorch kit 2.6.0 from official pip, and xformer v0.0.29.post2.
Everything should be good to launch, after edit webui-user.bat, after set COMMANDLINE_ARGS= --xformers

And it worked !
Thanks a lot mate !

[al-swaiti]

hi every one i fix the issue for LINUX USERS BY EDIT THIS FILE "stable-diffusion-webui/modules/launch_utils.py"
and edit this line "xformers_package" to latest xformers package as below >
xformers_package = os.environ.get('XFORMERS_PACKAGE', 'xformers==0.0.21.dev543')
and relaunch the "./webui.sh --xformers" also windows users can try this way try it before delete venv folder if it doesn't work rename or delete venv and relunch webui with xformers flag

[thiagojramos]

Ty. Works fine in my win11

[kryptonmediagroup]

This totally worked and it's totally worth doing. Rendering is at least 40% faster!

[EGirlEnthusiast]

This one fixed it for me on linux. If you dont know what xformers version you have then run pip list and itll show the version. Then edit xformers_package = os.environ.get('XFORMERS_PACKAGE', 'xformers==**reported version**')

[ghostofelvis100]

I'm on a Mac M1 - there must be some holes in my installation as I don't even have this script (launch_utils.py).

Any ideas?

Otherwise, thank you for your explicit post - I've been trawling lots of pages about these errors and most will refer to a command to use but give no reference as to where to place it.

-edit - just discovered Xformers and Cuda are specific to Nvidia GPUs and so are not relevant for an installation on Mac. doesn't necessarily explain why I'm missing the launch_utils.py...

[natalisuh]

Thank you, works for me too on linux, just with version 0.0.20
xformers_package = os.environ.get('XFORMERS_PACKAGE', 'xformers==0.0.20')