Trying to install pyg through conda downgrades torch to CPU

[SuhasSrinivasan]

😵 Describe the installation problem

After using conda for pytorch installation, trying to install pyg through the official guide.

$ conda install pyg -c pyg
Channels:
 - pyg
 - defaults
 - nvidia
 - pytorch
Platform: linux-64

...

The following packages will be SUPERSEDED by a higher-priority channel:

  pytorch            pytorch::pytorch-2.2.2-py3.12_cuda11.~ --> pkgs/main::pytorch-2.2.0-cpu_py312hb9e5694_0

The following packages will be DOWNGRADED:

  pytorch-mutex                                    1.0-cuda --> 1.0-cpu
  torchaudio                              2.2.2-py312_cu118 --> 2.2.0-py312_cpu
  torchvision                            0.17.2-py312_cu118 --> 0.17.0-py312_cpu

Environment

• PyG version: 2.5.2
• PyTorch version: 2.2.2
• OS: Ubuntu 22.04.4 LTS
• Python version: 3.12.2
• CUDA/cuDNN version: [System: Driver Version: 550.54.15 CUDA Version: 12.4] [conda: cuda-runtime 11.8]
• How you installed PyTorch and PyG (conda, pip, source): Pytorch and PyG through conda
• Any other relevant information (e.g., version of torch-scatter):

[rusty1s]

Can you try to install via

conda install pyg=*=*cu* -c pyg

[SuhasSrinivasan]

Hi Matthias, thank you for reviewing the issue. I had been struggling with multiple dependency issues for RoseTTAFold2NA, so to make things simple i.e., I split the deep learning libraries to be handled by pip.

I can confirm that the following works

pip install torch==2.2.1 torchvision torchaudio torchdata torch_geometric

Sorry cannot try the conda install now, but wanted to bring this to your attention as a genuine problem.