As @alturismo already pointed out, this is a container issue itself, is the CUDA runtime installed inside the container, not all but a few applications need the CUDA runtime installed in the container.
Of course there are other workarounds for this but non of them which you can easily do, for example install the Nvidia driver inside the container, but if you do so, it has to match the driver which is installed on the host and this is where issues for you, the user, start.
This is something the maintainer from the container has to solve.
EDIT: Forget that, I saw that you don't even pass through the Nvidia driver to the container from your issue that you've created on GitHub, please read the second post of this thread on how to do that.