building xFormers on Windows (instructions in post)

[duckness]

Is your feature request related to a problem? Please describe.
Edit the wiki to add information on how to build xFormers on Windows. Don't think I can make a PR for the wiki so I'll put it here.

Describe the solution you'd like

Building xFormers on Windows

1. Install VS Build Tools 2022, you only need Desktop development with C++

2. Install CUDA 11.3 (later versions are not tested), select custom, you only need the following (VS integration is probably unecessary):

3. Clone the xFormers repo, create a venv and activate it

git clone https://github.com/facebookresearch/xformers.git
cd xformers
git submodule update --init --recursive
python -m venv venv
./venv/scripts/activate

4. To avoid issues with getting the CPU version, install pyTorch seperately:

pip install torch torchvision --extra-index-url https://download.pytorch.org/whl/cu113

5. Then install the rest of the dependencies:

pip install -r requirements.txt
pip install wheel

6. As CUDA 11.3 is rather old, you need to force enable it to be built on MS Build Tools 2022. Do $env:NVCC_FLAGS = "-allow-unsupported-compiler" if on powershell, or set NVCC_FLAGS=-allow-unsupported-compiler if on cmd

7. You can finally build xFormers, note that the build will take a long time (probably 10-20minutes), it may initially complain of some errors but it should still compile correctly.

python setup.py build
python setup.py bdist_wheel

9. In xformers directory, navigate to the dist folder and copy the .whl file to the base directory of stable-diffusion-webui

10. In stable-diffusion-webui directory, install the .whl, change the name of the file in the command below if the name is different:

./venv/scripts/activate
pip install xformers-0.0.14.dev0-cp310-cp310-win_amd64.whl

11. Ensure that xformers is activated by launching stable-diffusion-webui with --force-enable-xformers

Describe alternatives you've considered

Additional context

[ClashSAN]

ok, adding this post in, thank you.

[duckness]

added missing submodule update command

[Gman0909]

How does this work if running the webui from an anaconda environment?

[duckness]

I don't use conda. you would need to figure out how to build a conda package and install it offline, this may be useful: https://github.com/facebookresearch/xformers/blob/main/.circleci/config.yml#L397

alternatively, you may be able to just install the .whl with pip into your conda environment without issue: https://docs.conda.io/projects/conda-build/en/latest/user-guide/wheel-files.html

[ghost]

This guide doesnt work either. I tried so many things to get it running. Is this attached error normal?

edit: it is still running though, but its throwing massive amounts of weird warnings and errors

Also you got the wrong slashes for activating venv, they should be backslashes

[ghost]

I let it run through and followed all the instructions and I get the exact same error when running the xformers that comes with the automatic repo

[ilcane87]

You probably need to rebuild xformers, this time specifying your GPU architecture.
Go inside the xformers folder, delete the folders 'xformers.egg-info', 'build' and 'dist', then repeat the process in the first post from the 'python -m venv venv' command, but after you send the command 'set NVCC_FLAGS=-allow-unsupported-compiler', also send the command 'set TORCH_CUDA_ARCH_LIST=7.5', then proceed until the end.
And don't forget the last step duckness suggested to you above.

[ghost]

Another guide wanted me to do that, which uninstalled xformers automatically, and it still didnt work. However I will try it out again with this guide, I dont think it will work, but I will try anyway. Thanks for the help. Its so frustrating seeing everyone else get a small christmas and I am stuck with this commandline insanity forever

[ilcane87]

Remember to send each command while inside the (venv), never close your command prompt.

[ghost]

So Im done with this I tried again and it didnt work. Im going to buy a 3090 Ti pretty soon and then I wont need another speed boost. Thanks for the help though this community is awesome

[Desm0nt]

Same GPU, same problem. Nothin helps. With 3060ti at home works perfectly, but with 2080S it's probably impossible :(

[ilcane87]

@duckness
Same issue here with a GeForce GTX 1060, my .whl is just 3.85 MB, took roughly 20 minutes to build.
I strictly followed the instructions, except that I added the environment variable "set TORCH_CUDA_ARCH_LIST=6.1" to the venv before building, since that's the architecture of my GPU.

[duckness]

I didn't encounter this so im not sure whats happening, I only had either a 270KB .whl (failed to compile properly) or something in the high tens of megabytes (compiled correctly). Try compile without the TORCH_CUDA_ARCH_LIST set.

[ilcane87]

I just followed your latest suggestion to @githubartman (uninstall xformers then install .whl), and the issue is finally solved.
Thanks a bunch, I'd been working on this for several hours now.

[tanguito34]

Hola me puedes indicar el link o compartir el fichero compilado , yo tengo una 1080ti y supongo que es la misma arquitectura 6.1 y no soy capaz de compilar el fichero .whl

[hd-x]

thank you very much. I followed the guide and it works perfectly!

[jzhangbs]

Maybe it's better to use "x64 native tools command prompt for VS 2022" to do all these things, especially when you have multiple VS installed

[ilcane87]

While this should've worked, since I'm not getting any errors when running with the --force-enable-xformers argument, there is actually no difference in the generation speed even without the argument, it's exactly the same.

Don't know if there's anything wrong with my setup, or if it's because it has no influence on GeForce GTX 1060.

[tanguito34]

The same thing happens to me, I don't notice a difference ,
and I get the message of xformer activated

LatentDiffusion: Running in eps-prediction mode
DiffusionWrapper has 859.52 M params.
Loading weights [7460a6fa] from C:\IA\stable-diffusion-webui\models\Stable-diffusion\sd-v1-4.ckpt
Applying xformers cross attention optimization.
Textual inversion embeddings loaded(0):

[puyebu]

Has anyone managed to make it work with 2070S or any other -S GPUs?
GPU-Z detects CUDA 7.5 compatibility, yet the compiled result ends with "No kernel image is available" on start.

[C43H66N12O12S2]

We've updated the repo with Pascal, Turing and Ampere compatible wheels a while ago. Just uninstall your current xformers and run the repo again with --xformers. It'll install the correct wheels on Windows.

[puyebu]

Yeah, that one works, though I expected more than ~3% improvement.
No idea why the one I compiled doesn't work but doesn't matter now.

[ilcane87]

~3% improvement
I was under that impression too, but I found that depending on the operation the improvement can be much more significant.
For instance, it takes me 75 seconds to do a 640x960 img2img without --xformers, and 50 seconds with it, meaning it speeds up by a third.

[Airbus1080]

How do I

We've updated the repo with Pascal, Turing and Ampere compatible wheels a while ago. Just uninstall your current xformers and run the repo again with --xformers. It'll install the correct wheels on Windows.

How do I uninstall and reinstall it? I just deleted the xformers folder and added the --xformers argument but didn't work.

[ilcane87]

@Airbus1080
Use the following commands from the same command prompt:
[Stable_Diffusion_Dir]\venv\Scripts\activate.bat
pip uninstall xformers

Then run SD with the argument --xformers.

[mykeehu]

I compiled a Python 3.8 + CUDA 11.3 version and a Python 3.10.7 + CUDA 11.6 version. Is this generally good for everyone or is it hardware dependent? Here is, test it with install!
I used the latter myself, and the image generation speed almost doubled (640x640 images in 5 seconds on an RTX 3060 card). Fantastic!

[Airbus1080]

How do I install this specific wheel (for py3.8). Do I just pip install *.whl?. Also is it possible to build it with CUDA 10.*?

[mykeehu]

I don't know. I made this under python 3.8 and setup for CUDA 11.3. Try using this command, I think it's python dependent, but if it's CUDA, you have to compile it every time you change CUDA, which is not good. I hope it runs for you. If not, you'll have to compile it yourself unfortunately.

pip install xformers-0.0.14.dev0-cp38-cp38-win_amd64.whl

Go this link, and select PIP and your CUDA version:
https://pytorch.org/get-started/locally/

"CUDA-10.2 PyTorch builds are no longer available for Windows, please use CUDA-11.6"

Here are the steps I did under Windows:

• open an empty folder, and cmd with administrator rights
• run these commands per line:

git clone https://github.com/facebookresearch/xformers.git
cd xformers
git submodule update --init --recursive
python -m venv venv
venv\scripts\activate
pip install torch torchvision --extra-index-url https://download.pytorch.org/whl/cu113 (or cu116)
pip install -r requirements.txt
pip install wheel
set NVCC_FLAGS=-allow-unsupported-compiler
python setup.py build
python setup.py bdist_wheel

• After creating the .whl file in the dist folder, I moved it to the stable-diffusion-webui folder. I also went to the cmd window and typed these commands:

venv\scripts\activate
pip install (your .whl file)

And that's all!

[ghost]

My problems have been solved since this repo has been updated with the new xformers wheels!

https://www.youtube.com/watch?v=KMU0tzLwhbE

[DarkAlchy]

Glad yours were because using the one that comes with it (iow one I didn't compile) made my renders take about 10% longer. I could stop it, remove the --xformers then restart it to do the exact same settings/prompt and it took longer every single time for the --xformers renders.

[mykeehu]

Start with --force-enable-xformers command!

[DarkAlchy]

It shows it was loaded just I take a hit in speed not gain any speed.

[DarkAlchy]

venv "D:\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.9.12 (main, Apr 4 2022, 05:22:27) [MSC v.1916 64 bit (AMD64)]
Commit hash: dce7fc9
Installing requirements for Web UI
Launching Web UI with arguments: --force-enable-xformers --port 9000
LatentDiffusion: Running in eps-prediction mode
DiffusionWrapper has 859.52 M params.
making attention of type 'vanilla' with 512 in_channels
Working with z of shape (1, 4, 32, 32) = 4096 dimensions.
making attention of type 'vanilla' with 512 in_channels
Loading weights [7460a6fa] from D:\stable-diffusion-webui\models\Stable-diffusion\model.ckpt
Global Step: 470000
Applying xformers cross attention optimization.
Model loaded.
1920 1200
1150
Loaded a total of 30 textual inversion embeddings.

I built this one from scratch.

[bitRAKE]

Thank you for the concise instructions!

I was able to build with CUDA 11.6 (if others want to go this route):

pip install torch torchvision  -f https://download.pytorch.org/whl/cu116/torch_stable.html

As a point of reference my generation of 2048 x 2048 textures went from 06:25 to 02:59 Amazing!

[ssshingetsu]

Cannot import xformers Traceback (most recent call last): File "F:\NovelAI\Clone\stable-diffusion-webui\modules\sd_hijack_optimizations.py", line 15, in <module> import xformers.ops ModuleNotFoundError: No module named 'xformers.ops'

i followed every step above, this is what happens when i try to launch the webui-user.bat with --force-enable-xformers

[bitRAKE]

Cannot import xformers Traceback (most recent call last): File "F:\NovelAI\Clone\stable-diffusion-webui\modules\sd_hijack_optimizations.py", line 15, in <module> import xformers.ops ModuleNotFoundError: No module named 'xformers.ops'

This error means xformers isn't present in the environment. Try redoing these commands:

./venv/scripts/activate
pip install xformers-0.0.14.dev0-cp310-cp310-win_amd64.whl

If you type pip list then xformers should show in the list.

[James0MEGA]

This didn't work for me at first, then I deactivated my venv instead of closing it out after installing the .whl, ran my bat and watched it spin up a brand spanking new instance of Auto in the venv. Approximately a 15-20% increase in it/s with xformers working.

[mykeehu]

In the latest builds, no compile is needed if you have Windows! ☝☝☝
"If you use a Pascal, Turing or Ampere card, you shouldn't need to build manually anymore. Uninstall your existing xformers and launch the repo with --xformers. A compatible wheel will be installed."
Read this article, it has been updated!

[DarkAlchy]

In the latest builds, no compile is needed if you have Windows! ☝☝☝ "If you use a Pascal, Turing or Ampere card, you shouldn't need to build manually anymore. Uninstall your existing xformers and launch the repo with --xformers. A compatible wheel will be installed." Read this article, it has been updated!

Except it doesn't. I am having an issue that started today so I cloned to start completely over from scratch (even after building xformers) and when it loaded up it said xformers was not part of the package or something. Basically I had to compile xformers all over again (25m) then it recognized it.

[lolxdmainkaisemaanlu]

does this work on colab?

[Mikolaj7777]

Hi!
Still getting: "Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check"

:/
Windows 10
gtx 1060
Cuda 11.7
Pytorch 11.3

Main Stable diffusion folder is on desktop. Rest installed as was.

[UserSpy]

Hi I'm getting an error trying to build xformers on windows.

I can't figure out what I'm doing wrong. Googling the error makes it seem like a problem is with xformers themselves but it's probably something to do with my computer. Anyone have any ideas?

[DarkAlchy]

To build it I had to use 2019 as the docs said.

[HoneyCodeBadger]

I am a bit lost. I try to install xformers and it looks good until step 6. I was able to do the commands in several areas, but without success afterwards at step 7. There it crashes directly.

In general I tried to run all commands (if possible) in the same bash. I used git bash for the installtion until step 6. But afterwards I tried cmd as admin, powershell and both from visual studio (stable-diffusion-webui branch) as well. No success.

RTX 3070 Ti Laptop is in use. 8 GB RAM.

No issue on Power Shell.

PS C:\Users\User\Documents\stable-diffusion-webui> $env:NVCC_FLAGS = "-allow-unsupported-compiler"
PS C:\Users\User\Documents\stable-diffusion-webui>

No issue on git bash.

User@Asus MINGW64 ~/xformers (main)
$ set NVCC_FLAGS=-allow-unsopported-compiler

User@Asus MINGW64 ~/xformers (main)
$ 

But then on step 7.

$ python setup.py build
No CUDA runtime is found, using CUDA_HOME='C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.3'
C:\Users\User\AppData\Local\Programs\Python\Python310\lib\site-packages\setuptools\dist.py:530: UserWarning: Normalizing '0.0.15.dev+efdca02.d20221116' to '0.0.15.dev0+efdca02.d20221116'
  warnings.warn(tmpl.format(**locals()))
running build
running build_py
copying xformers\version.py -> build\lib.win-amd64-cpython-310\xformers
running build_ext
C:\Users\User\AppData\Local\Programs\Python\Python310\lib\site-packages\torch\utils\cpp_extension.py:358: UserWarning: Error checking compiler version for cl: [WinError 2] Das System kann die angegebene Datei nicht finden
  warnings.warn(f'Error checking compiler version for {compiler}: {error}')
building 'xformers._C' extension
Emitting ninja build file C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\build.ninja...
Compiling objects...
Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)
ninja: no work to do.
"C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.34.31933\bin\HostX86\x64\link.exe" /nologo /INCREMENTAL:NO /LTCG /DLL /MANIFEST:EMBED,ID=2 /MANIFESTUAC:NO /LIBPATH:C:\Users\User\AppData\Local\Programs\Python\Python310\lib\site-packages\torch\lib /LIBPATH:C:\Users\User\AppData\Local\Programs\Python\Python310\libs /LIBPATH:C:\Users\User\AppData\Local\Programs\Python\Python310 /LIBPATH:C:\Users\User\AppData\Local\Programs\Python\Python310\PCbuild\amd64 "/LIBPATH:C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.34.31933\ATLMFC\lib\x64" "/LIBPATH:C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.34.31933\lib\x64" "/LIBPATH:C:\Program Files (x86)\Windows Kits\10\lib\10.0.22000.0\ucrt\x64" "/LIBPATH:C:\Program Files (x86)\Windows Kits\10\\lib\10.0.22000.0\\um\x64" c10.lib torch.lib torch_cpu.lib torch_python.lib /EXPORT:PyInit__C C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\Users\User\xformers\xformers\components\attention\csrc\attention.obj C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\Users\User\xformers\xformers\components\attention\csrc\autograd\matmul.obj C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\Users\User\xformers\xformers\components\attention\csrc\cpu\attention.obj C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\Users\User\xformers\xformers\components\attention\csrc\cpu\matmul.obj C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\Users\User\xformers\xformers\components\attention\csrc\cpu\sddmm.obj C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\Users\User\xformers\xformers\components\attention\csrc\cpu\sparse_softmax.obj C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\Users\User\xformers\xformers\components\attention\csrc\cpu\spmm.obj C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\Users\User\xformers\xformers\components\attention\csrc\matmul.obj C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\Users\User\xformers\xformers\components\attention\csrc\sddmm.obj C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\Users\User\xformers\xformers\components\attention\csrc\sparse_softmax.obj C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\Users\User\xformers\xformers\components\attention\csrc\spmm.obj C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\Users\User\xformers\xformers\components\swiglu\swiglu_op.obj C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\Users\User\xformers\xformers\components\swiglu\swiglu_packedw.obj /OUT:build\lib.win-amd64-cpython-310\xformers\_C.pyd /IMPLIB:C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\Users\User\xformers\xformers\components\attention\csrc\_C.lib
   Bibliothek "C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\Users\User\xformers\xformers\components\attention\csrc\_C.lib" und Objekt "C:\Users\User\xformers\build\temp.win-amd64-cpython-310\Release\Users\User\xformers\xformers\components\attention\csrc\_C.exp" werden erstellt.
Code wird generiert.
Codegenerierung ist abgeschlossen.
attention.obj : error LNK2001: Nicht aufgelöstes externes Symbol "__imp___tls_index_?init@?1??lazy_init_num_threads@internal@at@@YAXXZ@4_NA".
attention.obj : error LNK2001: Nicht aufgelöstes externes Symbol "__imp___tls_offset_?init@?1??lazy_init_num_threads@internal@at@@YAXXZ@4_NA".
build\lib.win-amd64-cpython-310\xformers\_C.pyd : fatal error LNK1120: 2 nicht aufgelöste Externe
error: command 'C:\\Program Files\\Microsoft Visual Studio\\2022\\Community\\VC\\Tools\\MSVC\\14.34.31933\\bin\\HostX86\\x64\\link.exe' failed with exit code 1120

After googleing the exit code it looked for me like cuda dependencies are missing. Tried to pip install pycuda==2021.1 but did not change anything...

[leppie]

You have a typo with $ set NVCC_FLAGS=-allow-unsopported-compiler

[HoneyCodeBadger]

Thanks, but this was not the issue. Because the other one was fine. I've now installed 2019, as the guy had to do above. Afterwards the other steps worked. :)
On final question,
Where to add --force-enable-xformers? I am starting the webgui with a link to the desktop. Add this simply there in the path to start?

[DarkAlchy]

I use the force-enable out of habit so not sure if it is still needed.

[HoneyCodeBadger]

Finally found this as well, if someone is searching for this like me:
Open launch.py in the stable-diffusion-webui and search for commandline_args and add at os.environ.get('COMMANDLINE_ARGS',"")
--force-enable-xformers

...
def prepare_enviroment():
    torch_command = os.environ.get('TORCH_COMMAND', "pip install torch==1.12.1+cu113 torchvision==0.13.1+cu113 --extra-index-url https://download.pytorch.org/whl/cu113")
    requirements_file = os.environ.get('REQS_FILE', "requirements_versions.txt")
    commandline_args = os.environ.get('COMMANDLINE_ARGS', "--force-enable-xformers")
...

Afterwards you will notice the following in the cmd, when starting the webui:
Launching Web UI with arguments: --force-enable-xformers

[mykeehu]

Here are a bunch of parameters, if you don't want to fetch them from the launch.py file

[drandarov-io]

I also had to use C++ Build Tools 2019. Here's how to get it:

Option 1

• Get 2019 Build Tools from https://aka.ms/vs/16/release/vs_buildtools.exe
• Choose MSVC v142 - VS 2019 C++ x64/x86 build tools and C++ CMake tools for Windows
• Run the xformers build

Option 2

• Get Visual Studio 2022
• Install C++ CMake tools for Windows, which will also install MSVC v143 ...
• Additionally install MSVC v142 - VS 2019 C++ x64/x86 build tools
• Run & 'C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Auxiliary\Build\vcvarsall.bat' x86_amd64 -vcvars_ver="14.29"
• Run the xformers build

Running xformers build (with extensions / dev)

cd .\xformers\
..\stable-diffusion-webui\venv\Scripts\Activate.ps1
pip install --verbose --no-deps -e .

[Milithrill]

Im having some major issues installing this, been trying for like 3 days and it does not seem to want to install.
Ive installed all the visual C, both 2022 and 2019.
Ive tried installing it in the old way (without venv), but cant get past the error of pip problem with compiling, even when I try the allow command.
Specific to this guide I just end up with this error and cant get past it.

[drandarov-io]

That's an error I haven't seen before. I'll make it easy and upload my wheel. Obviously use at your own risk: https://anonfiles.com/NeD2c9L6y8/xformers-0.0.15.dev0_4c06c79.d20221205-cp310-cp310-win_amd64_whl

cd .\stable-diffusion-webui\
.\venv\Scripts\Activate.ps1
pip install .\xformers-0.0.15.dev0+4c06c79.d20221205-cp310-cp310-win_amd64.whl

@Stable-Diffusion-WebUI / @xformers If that is somehow problematic, I can also delete it.

[drandarov-io]

I did that, now I get this. If there is no solution to this you shouldnt include it in the program if you need 190iq and coding experience just to install the program.

Did you run git submodule update --init --recursive?

[Milithrill]

Will it work on win11 64bit, intel 10900k + 4090?
That command I havent run, but I am new to coding in general. Ive edited some game modes at a code level. But havent used visual C, or python, git, conda or antyhing simliar before. So I would have no idea what it does, and when and where to run it. I just followed the steps exactly, and ran in to some errors in webui, having to reinstall cuda and python etc. again to make it work, so unless there is an easy explanation or way to use xformers, then its just not worth the time I would potentially save if I run into multiple errors for everything Im trying to do.

[drandarov-io]

Should work for that configuration. You have to run all these commands in the xformers folder. You also need to have git installed.

[tanguito34]

That's an error I haven't seen before. I'll make it easy and upload my wheel. Obviously use at your own risk: https://anonfiles.com/NeD2c9L6y8/xformers-0.0.15.dev0_4c06c79.d20221205-cp310-cp310-win_amd64_whl

cd .\stable-diffusion-webui\
.\venv\Scripts\Activate.ps1
pip install .\xformers-0.0.15.dev0+4c06c79.d20221205-cp310-cp310-win_amd64.whl

@Stable-Diffusion-WebUI / @xformers If that is somehow problematic, I can also delete it.

I just installed them on a secondary pc with a gpu gtx980 architecture 5.2 4 gb and I'll tell you how well it works for me

[illtellyoulater]

In my case when I run WebUI I see:

No module 'xformers'. Proceeding without it.

So I guess building xFormers is not a mandatory step? Or what advantages does it provide?

[drandarov-io]

Just makes the generation slightly faster.

[Lacrivilho]

My .whl is only 371 KB, which apparently means it didn't compile properly. Is there any solution I can try? I followed the instructions exactly and changed to vs 2019 after it failed the first time. I also tried using this improperly compiled file and it installed without issue, but caused errors when I tried to launch the webui.
There's a lot of warnings and some errors in the log, but I have no idea which of them are relevant. At no point does it say that it failed to compile anything.

[Panchovix]

Guys, for those who are using torch with either cu116, cu117 or nightly cu117, do you build your own xformers with that same cuda versions that you're using in torch? Or do you build all on cu113?

[drandarov-io]

I built xformers with cu117

[tanguito34]

what version of pytorch do you have?

[TaoismDeepLake]

Launching Web UI with arguments: --force-enable-xformer
D:\GitRepo\AI\sd-webui-raw\venv\lib\site-packages\pkg_resources_init_.py:123: PkgResourcesDeprecationWarning: llow is an invalid version and will not be supported in a future release
warnings.warn(
No module 'xformers'. Proceeding without it.
Cannot import xformers
Traceback (most recent call last):
File "D:\GitRepo\AI\sd-webui-raw\modules\sd_hijack_optimizations.py", line 140, in
import xformers.ops
ModuleNotFoundError: No module named 'xformers.ops'; 'xformers' is not a package

[TaoismDeepLake]

changed it to --xformers and it worked.