Open
Description
Description
Somehow conda env -f ...
installs pytorch-cpu on cuda toolkit v10.1.
In which platform does it happen?
Tested only on Linux-GPU (DSVM NC6v3)
How do we replicate the issue?
Update cuda from v10.0 to v10.1 and create conda env by following SETUP.md
Expected behavior (i.e. solution)
Either manually update torch after env creation by running:
conda install pytorch torchvision cudatoolkit=10.1 -c pytorch
or specify cuda version on environment.yaml like:
...
dependencies:
...
- pytorch=1.3.1=py3.6_cuda10.1.243_cudnn7.6.3_0
- torchvision=0.4.2=py36_cu101
...
fixes it (i.e. install gpu torch).