-
Notifications
You must be signed in to change notification settings - Fork 606
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Error importing xformers after building from source #641
Comments
OK so looking through what the wheel installed:
It seems that |
Yeah, a simple copy to the right directory seems to have fixed it:
🤷 |
Thansk a lot for the debugging & workaround! Indeed this seems to be the case for a few users - we want to push a fix tomorrow. |
stuck on this, just curious if there is a way to install a version of xformers that's not the main dev branch? Its not clear from readme. |
You can also git clone manually (make sure to include |
Had the same issue, rolling back to 0.0.15 solved it |
Should be fixed now after 6a988dd |
pip wheels with the fix are now available - closing |
🐛 Bug
After building xformers from source with this command:
pip install -v -U git+https://github.com/facebookresearch/xformers.git@main#egg=xformers
It installs and builds successfully, but errors out when trying to import on Pytorch 1.13.1:
And (after being built again) Pytorch 2.0:
I tried building for Pytorch 1.13.1 installed via the Arch Linux repos and Pytorch 2.0 cu118 from pip.
Builds as recently as... 2 days ago, I think, seem to work fine. I suppose I will try rolling a few commits back.
Additional context
Note that I can't use precompiled xformers wheels from
pip install --pre xformers
on my platform. Not sure if this is because I am using CUDA 11.8, or because they are not matched to my version(s) of PyTorch. Maybe y'all should make a stable "main" and a "dev" branch for those of us who have to build xformers from source?The text was updated successfully, but these errors were encountered: