Can't use GPU on Google colab in conda environment (Tensorflow) - tensorflow

Question

I made a short Colab Notebook to simulate the situation on my local PC where I use conda / mamba and can't get any GPU working for tensorflow. I know colab can do this out of the box, but that's not, what I want to do. I'm also quite new to colab and still trying to figure out, how it's used :D.
So here's my Code for colab:
# Install Miniconda
!curl -O https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
!sh Miniconda3-latest-Linux-x86_64.sh -b -f -p /usr/local
# Set channels
!conda config --add channels defaults
!conda config --add channels bioconda
!conda config --add channels conda-forge
# Install Mamba
!conda install -y mamba
next cell:
# Create Environment
!mamba create -y -n stats-gpu tensorflow-gpu cudatoolkit cudnn
I also tried:
!mamba create -n stats-gpu python=3.8 tensorflow-gpu=2.4 cudnn=8.0 cudatoolkit=11.0
next cell:
# simulate "conda activate" by setting PYTHONPATH and PATH
import sys
import os
PYTHONPATH = os.environ['PYTHONPATH']
sys.path = ['',
PYTHONPATH,
'/usr/local/envs/stats-gpu/lib/python37.zip',
'/usr/local/envs/stats-gpu/lib/python3.7',
'/usr/local/envs/stats-gpu/lib/python3.7/lib-dynload',
'/usr/local/envs/stats-gpu/lib/python3.7/site-packages']
os.environ['PATH'] = ('/usr/local/envs/stats-gpu/bin:' +
'/usr/local/condabin:' +
os.environ['PATH'])
next cell:
# Executing directly in cell will ignore python installed by conda.
# Writing Python script in .py file and executing with !python will use
# Python from conda.
filename = "tensortest.py"
with open(filename, 'w') as file:
file.write("""import tensorflow as tf
print("Num GPUs Available: ", len(tf.config.list_physical_devices('GPU')))""")
!python tensortest.py
The output is
Num GPUs Available: 0
but it should be 1. Why is that? The GPU in colab is active and works when I just use
import tensorflow as tf
print("Num GPUs Available: ", len(tf.config.list_physical_devices('GPU')))
in an empty Notebook.