installation dependencies

🚀 The feature

pip install torchvision==0.8.2+cu101 requires torch==1.7.1 to be installed instead of torch==1.7.1+cu101

this causes poetry package manager crushing during the installation.

Updating dependencies
Resolving dependencies... (1720.5s)

SolverProblemError

Because torchvision (0.8.2+cu101) depends on torch (1.7.1)
and depends on torch (1.7.1+cu101), torchvision is forbidden.
So, because depends on torchvision (0.8.2+cu101), version solving failed.

It would be great if torchvision requested the appropriate pytorch version.

Motivation, pitch

pip install torchvision==0.8.2+cu101 -f https://download.pytorch.org/whl/torch_stable.html
Looking in indexes: https://pypi.org/simple, https://pypi.ngc.nvidia.com
Looking in links: https://download.pytorch.org/whl/torch_stable.html
Collecting torchvision==0.8.2+cu101
Downloading https://download.pytorch.org/whl/cu101/torchvision-0.8.2%2Bcu101-cp36-cp36m-linux_x86_64.whl (12.8 MB)
|████████████████████████████████| 12.8 MB 1.1 MB/s
Collecting pillow>=4.1.1
Downloading Pillow-8.4.0-cp36-cp36m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (3.1 MB)
|████████████████████████████████| 3.1 MB 1.1 MB/s
Collecting torch==1.7.1
Downloading https://download.pytorch.org/whl/rocm3.8/torch-1.7.1%2Brocm3.8-cp36-cp36m-linux_x86_64.whl (588

Alternatives

No response

Additional context

No response

Start by reproducing the reported torchvision==0.8.2+cu101 installation with pip and Poetry, using the commands and solver error in the issue. Inspect the package dependency metadata for how it requests torch, then verify that the CUDA-specific torch requirement resolves without the reported conflict.