Reputation: 794
When checking the installed pytorch version you can do that in two ways:
pip3 show torch
(or similar python3 -m pip freeze
for all packages)import torch; torch.__version__
Interestingly the first option (using pip3
) returns 1.8.1+cu111
, while the second option (torch.__version__
) returns 1.7.1
(without the cuda version support string, but cuda is available).
Why do these two methods show different results and which one is the "valid" one?
Important When I installed pytorch with cuda support I installed the latest version, but downgraded a few weeks ago for different reasons.
Some info:
Upvotes: 0
Views: 2253
Reputation: 16935
This typically happens when you have multiple versions of the same library installed for some reason (normally mixing conda install
and pip install
in my experience). I recommend uninstalling oldest versions using the appropriate package manager until you see the expected behavior.
Upvotes: 1