-
Notifications
You must be signed in to change notification settings - Fork 199
Closed
Description
Hi,
I'm using the torch 1.6.0 whls listed here
https://pytorch-geometric.com/whl/torch-1.6.0.html
With the introduction of pip 20.3 strict resolver we are no longer able to install from this location. e.g.
pip install torch-sparse==latest+cu101 -f https://pytorch-geometric.com/whl/torch-1.6.0.html
As latest is not an official supported version number.
Even when attempting to install a specific version runs into issues
pip install torch-sparse==0.6.8+cu101 -f https://pytorch-geometric.com/whl/torch-1.6.0.html
Looking in links: https://pytorch-geometric.com/whl/torch-1.6.0.html
ERROR: Requested torch-sparse==0.6.8+cu101 from https://pytorch-geometric.com/whl/torch-1.6.0/torch_sparse-0.6.8%2Bcu101-cp36-cp36m-linux_x86_64.whl has different version in metadata: '0.6.8'
Here is the merge into pip that caused the issue
pypa/pip#9177
I'm not sure how to properly solve this issue. You want the metadata to point to a version regardless of cuda/cpu support, but I think the best way would b e to set the version metadata in the .whl to exactly the url you upload the .whl to.
Metadata
Metadata
Assignees
Labels
No labels