Thanks for the link. Yes. I was. Instance running 11.4 and cudnn 8.6 now.
Here are some notes for the record:
When I tried to run the scripts (by modifying the version string in the scripts), it apparently the version of the deeplearning scripts on my instance only fetched the driver, not CUDA.
I then downloaded the 11.4 cuda sdk in the debian local installer format, and 8.6 cudnn in the tar.xz format.
Running the install for cudasdk and copying the files from cudnn to the cuda-11.4 directory mostly solved this.
Early in the process, I made several failed attempts using the debian .deb installers from nvidia. In the end I had to purge all "nvidia." packages to start from fresh.
Yu