Adding the cudnn development kit to an nvidia container?


I’ve been trying to create a container which can build the dlib library from scratch. To do so, I need CUDA, cublas and cudnn. I got the container from here. More specifically, I selected 11.4.1-cudnn8-devel-ubuntu18.04

But it would seem like this container contains cuda, cublas and cudnn (as in but not the cudnn development library (the one with cudnn.h and such). This makes it impossible to build dlib from scratch. I know I can download those files directly from this website after logging in, but that defeats the purpose of making everything automated with docker, so it’s not really an option.

I have been looking for a solution to this for a while now. I have tried adding --runtime nvidia when running the image as suggested here but it does not help.

Any other way of making this work?