Thank you very much for your tutorial. I encountered a problem at step 13 during the installation process, but I have now resolved it, so I am sharing the solution.

The error displayed by Tensorflow was: "Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU."
However, it did not specify which GPU libraries were missing. Later, I realized that Tensorflow had not detected the path to cuDNN. I found a solution here: comment

I added the following content to my ~/.bashrc:

NVIDIA_DIR=$(dirname $(dirname $(python -c "import nvidia.cudnn;print(nvidia.cudnn.__file__)")))
for dir in $NVIDIA_DIR/*; do
    if [ -d "$dir/lib" ]; then
        export LD_LIBRARY_ PATH="$dir/lib:$LD_LIBRARY_ PATH"
    fi
done

Then, just source ~/.bashrc in terminal, this problem was resolved. I hope this can help other beginners :)

My running environment is as follows:

Software:

• NVIDIA Graphics Driver 555.85 (Windows)
• NVIDIA CUDA Toolkit 12.5 (Windows)
• WSL2 Kernel version: 5.15.153.1-2 (Windows)
• Ubuntu 22.04.03 LTS (in WSL2)
• Conda 24.5.0 (in WSL2)
• NVIDIA CUDA Toolkit 12.5 (in WSL2)
• Python 3.11 (in WSL2)
• Tensorflow[and-cuda] 2.16.1 (in WSL2)