Closed remtav closed 1 year ago
Conda environments built from our environment.yml install a cpu version of pytorch even on machines with GPU. See issue: https://stackoverflow.com/questions/74234628/new-mamba-environment-force-torch-cpu-and-i-dont-know-why One solution (tested and works) seems to fix pytorch to 1.12. We could also add pytorch-cuda=11.6 but this would make cpu installs larger. I'd go for fixing pytorch to 1.12, considering 1.13.1 doesn't have that much more to offer for us.
pytorch-cuda=11.6
Even with pytorch==1.12, torch still fails to discover my GPU
@CharlesAuthier
So the solution is to recreate the environment and not just update it.
Conda environments built from our environment.yml install a cpu version of pytorch even on machines with GPU. See issue: https://stackoverflow.com/questions/74234628/new-mamba-environment-force-torch-cpu-and-i-dont-know-why One solution (tested and works) seems to fix pytorch to 1.12. We could also add
pytorch-cuda=11.6
but this would make cpu installs larger. I'd go for fixing pytorch to 1.12, considering 1.13.1 doesn't have that much more to offer for us.