I got a new machine with an Nvidia GPU (RTX A1000 6GB). I read that the GPU is capable of using CUDA and cuDNN. Now I am trying to set up CUDA with cuDNN on WSL2 (Ubuntu 22.04) and Poetry. CUDA is recognized, but not cuDNN.

First of all I installed the CUDA driver like described in the following page: https://www.youtube.com/watch?v=JaHVsZa2jTc
Then I tried to install the cuDNN driver from Nvidia’s page (https://developer.nvidia.com/cudnn-downloads?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu&target_version=22.04&target_type=deb_local)
as a .deb and as a .tar, but still, after starting Python and importing TensorFlow, I get the following message:
E tensorflow/compiler/xla/stream_executor/cuda/cuda_dnn.cc:9342] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered

I searched the internet for several tutorials, but most of them are just related to Anaconda.
Does someone have a clean and clear instruction on how to set up WSL for using CUDA and cuDNN?