Pytorch docker image gives unexpected CUDA initialization error

Hello everyone,

I wanted to use a newer version for my pytroch base image. I wanted to use this https://hub.docker.com/layers/pytorch/pytorch/2.0.0-cuda11.7-cudnn8-devel/images/sha256-96ccb2997a131f2455d70fb78dbb284bafe4529aaf265e344bae932c8b32b2a4?context=explore.

Checking for cuda availability ends up in the following error:

Even if nvidia-smi outputs the following:

I would appreciate feedback if anyone knows where the problem might be. I have not made any changes to the image itself.