Skip to content

UV PyTorch indexes not found for cu126 #12774

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

Open
bakas-george opened this issue Apr 9, 2025 · 6 comments
Open

UV PyTorch indexes not found for cu126 #12774

bakas-george opened this issue Apr 9, 2025 · 6 comments
Assignees
Labels
needs-mre Needs more information for reproduction, see #9452

Comments

@bakas-george
Copy link

Summary

Hello,

By default, torch installation via uv using CUDA 12.6 fails due to missing support in the default uv PyTorch indexes:

  • The uv PyTorch index does not support CUDA 12.6 yet whereas CUDA 12.4 is officially supported on uv.

The issue is that you can manually override the index url on the command line:

uv pip install torch --index-url https://download.pytorch.org/whl/cu126

but when specified in the pyproject.toml this is not overridden.

Platform

Linux Ubuntu 24.04 LTS

Version

uv 0.6.13

Python version

3.13.3

@bakas-george bakas-george added the bug Something isn't working label Apr 9, 2025
@charliermarsh
Copy link
Member

Sorry, what do you mean by “by default”? How would uv know to look on the PyTorch index if it wasn’t specified?

@bakas-george
Copy link
Author

It was specified in the pyproject.toml like this:

[[tool.uv.index]]
name = "torch-gpu"
url = "https://download.pytorch.org/whl/cu126"
explicit = true

@charliermarsh
Copy link
Member

Okay, but what error are you experiencing when you do that? uv doesn't hard-code any distinction between those indexes. All PyTorch indexes should work equally well.

@charliermarsh charliermarsh self-assigned this Apr 10, 2025
@charliermarsh charliermarsh added needs-mre Needs more information for reproduction, see #9452 and removed bug Something isn't working labels Apr 10, 2025
@rulller
Copy link

rulller commented Apr 14, 2025

When running uv sync it does not install some additional libraries needed for pytorch with cuda.

On the other hand, when running uv pip install torch --index https://download.pytorch.org/whl/cu126 the libraries are installed.

These libraries are:

 + nvidia-cublas-cu12==12.6.4.1
 + nvidia-cuda-cupti-cu12==12.6.80
 + nvidia-cuda-nvrtc-cu12==12.6.77
 + nvidia-cuda-runtime-cu12==12.6.77
 + nvidia-cudnn-cu12==9.5.1.17
 + nvidia-cufft-cu12==11.3.0.4
 + nvidia-curand-cu12==10.3.7.77
 + nvidia-cusolver-cu12==11.7.1.2
 + nvidia-cusparse-cu12==12.5.4.2
 + nvidia-cusparselt-cu12==0.6.3
 + nvidia-nccl-cu12==2.21.5
 + nvidia-nvjitlink-cu12==12.6.85
 + nvidia-nvtx-cu12==12.6.77
 + triton==3.2.0

@konstin
Copy link
Member

konstin commented Apr 14, 2025

Can you share your pyproject.toml?

If you are using explicit = true, it will work different than --index https://download.pytorch.org/whl/cu126, does it work without explicit = true?

@KukumavMozolo
Copy link

KukumavMozolo commented Apr 15, 2025

Thats a known bug wiith pytorch 2.6.0 and cu126, use cu124
python-poetry/poetry#10152
pytorch/pytorch#146679 (comment)
also allready mentioned here i think: #10693

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
needs-mre Needs more information for reproduction, see #9452
Projects
None yet
Development

No branches or pull requests

5 participants