Is it possible to use cuda in docker without using runtime=nvidia

Hello, thank you for the library. I want to know if it is possible to copy host cuda libraries into docker container and forward only the device in any way (runtime=nvidia means that we forward both libraries and devices)? What we want is to have all ‘library environment’ on remote Jetson’s docker and not depend on host-side libraries version. The goal is to update docker image with new Cuda packages using the same device from the host. If i’m right our devops team is afraid of updating the remote host machine. But it’s Ok with docker.