I want to compile a cuda extension, using setup.py and CUDAExtension as described here . this thread is similar to my question but not exactly. How can I get the cuda version that my “conda install pytorch” is currently using? I want to make sure the binaries would be compatible.

Pytorch 0.4 get cuda version

dashesy (dashesy) June 1, 2018, 4:45pm 2

I figured out:

torch.version.cuda

Perhaps this should be mentioned in the document

34 Likes

How to get cuda cudnn nccl build version?