There is a very similiar topic here and the answer was:
@gabidanpremier Since recognize will run in the nextcloud docker container, you will need to install tensorflow (and CUDA) in the Nextcloud container. Recognize’s admin settings should have indicators whether GPU is accessible.
How do I perform this?
The only hints from nextcloud are here.
root@c68b506a3ed9:/var/www/html# apt-get install libcudnn8=8.6.0.163-1+cuda11.8
Reading package lists... Done
Building dependency tree... Done
Reading state information... Done
E: Unable to locate package libcudnn8
But as stated, this leads to an error.
Some search brought me to these instructions, but they also fail
sudo nvidia-ctk runtime configure --runtime=docker
unable to load config for runtime docker: EOF
Any hints or guide how to perform this?