Failed to use GPU using Tensorflow 2.4 inside WSL2

Accelerated Computing NGC GPU Cloud Docker and NVIDIA Docker
1

Hi, I am testing Parameter Server Training for multi GPU training workers.

I have followed this guide in order to install a container with GPU access inside a WSL2:

. Everything installed, NO errors.

Now I am starting to follow this TF oficial guide to use Parameter Server Training:

The problem is that Parameter Server Training guide only works on TF ver 2.4
If I try to pull a tensorflow image above 2.1 GPU fails into TF.