I tried to install pytorch following the commands suggested in the homepage.
My preferences:
OS -- Linux
Package Manager -- pip
Python -- 3.5
CUDA -- 9.1
The suggested commands:
pip3 install http://download.pytorch.org/whl/cu91/torch-0.4.0-cp35-cp35m-linux_x86_64.whl
pip3 install torchvision
So i install pytorch and then go to install the torchvision. However, as one of it’s dependency, torchvision starts installing pytorch again. Here are snippets from command line:
$ pip3 install --user http://download.pytorch.org/whl/cu91/torch-0.4.0-cp35-cp35m-linux_x86_64.whl
Collecting torch==0.4.0 from http://download.pytorch.org/whl/cu91/torch-0.4.0-cp35-cp35m-linux_x86_64.whl
...
$ pip3 install --user torchvision
...
Collecting torch (from torchvision)
Using cached https://files.pythonhosted.org/packages/33/65/e40bc02754ad2fd66e4953e276d8c608cdb0e3b6b516fb485ea1efaeeeea/torch-0.4.0-cp35-cp35m-manylinux1_x86_64.whl
...
I noticed above that when installing torchvision, it seems downloading another whl of pytorch. So I checked the actual module files (~/.local/lib/python3.5/site-packages/torch/version.py) and found the latter installation (cuda = ‘8.0.61’) had overwritten the previous one (cuda = ‘9.1.85’).
Doesn’t really make sense to me – is this expected? Shall I reinstall the pytorch again(to be sure I got the correct version)? Thanks!