Auxority / wsl-tensorflow-guide

A guide for installing Tensorflow using WSL on Windows.

Tensorflow 2 - Windows with WSL installation

Important

Save yourself from pain and just do this: https://github.com/Auxority/tensorflow-jupyter

Windows

1. Install any Windows updates that you might have.

NVIDIA drivers

1. Visit this link or open Geforce Experience.
2. Update your drivers to the latest version.

NVIDIA CUDA Toolkit

1. Visit this link to find the latest download of CUDA Toolkit.
2. Install the latest CUDA Toolkit using the installer.
3. Add the following to the system path environment variable (click here for more information).:
   • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\bin
   • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\extras\CUPTI\lib64
   • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\include
   • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\libnvvp
4. Visit this link for more information about the installation of CUDA Toolkit.

Zlib

1. Visit this link to find the latest zLib downloads.
2. Right click the pre-built zlib DLL download link for AMD64/Intel EM64T, and select Save link as.. to save the zip file to your computer.
3. Extract the zip file. The folder should contain a dll_x64 and static_x64 directory.
4. Copy the extracted folder.
5. Go to C:\Program Files\ on your computer using File Explorer.
6. Create a new directory here called Zlib.
7. Enter the new Zlib directory you just created.
8. Paste the extracted folder you have copied.
9. Add the dll_x64 directory to your system path environment variable (click here for more information).

cuDNN SDK

1. Visit this link to find the latest cuDNN download.

2. You will have to create an account in order to download this.

3. Once you are logged in download the latest Windows zip file.

4. Unzip the cuDNN zip file.

5. Open the extracted folder.

   Important note: You must replace 8.x and 8.x.y.z with your specific cuDNN version.

6. Create a directory called NVIDIA in C:\Program Files

7. Copy all files from the bin directory to C:\Program Files\NVIDIA\CUDNN\v8.x\bin

8. Copy all files from the include directory to C:\Program Files\NVIDIA\CUDNN\v8.x\include

9. Copy all files from the lib directory to C:\Program Files\NVIDIA\CUDNN\v8.x\lib

10. Add C:\Program Files\NVIDIA\CUDNN\v8.x\bin to your system path environment variable.

11. Visit this link for more information about the cuDNN SDK installation.

Python

1. Visit this link to view Python downloads for Windows.
2. Download the Python 3.10.x installer (just select the most recent one).
3. Run the installer and make sure to use the checkbox to add Python to your PATH.
4. Open your favorite terminal.
5. Run pip install --upgrade pip to update pip to the latest version.

WSL

1. Make sure you are running Windows 11.
2. Open your favorite terminal.
3. Run wsl --shutdown to stop all WSL instances.
4. Run wsl -l | tail -n +2 | awk '{print $1}' | xargs -I {} wsl --unregister {} to delete all your WSL instances.
5. Run wsl --install to install WSL2 if this has not been installed.
6. Run wsl --install Ubuntu to install Ubuntu (create a username and password here).

wsl --shutdown
wsl -l | tail -n +2 | awk '{print $1}' | xargs -I {} wsl --unregister {}
wsl --update
wsl --install

7. Open your favorite terminal.
8. Run wsl to enter the Ubuntu instance.
9. Copy and paste the following commands in WSL to install CUDA:

sudo apt-key del 7fa2af80

wget https://developer.download.nvidia.com/compute/cuda/repos/wsl-ubuntu/x86_64/cuda-keyring_1.0-1_all.deb
sudo dpkg -i cuda-keyring_1.0-1_all.deb

sudo apt-get update && sudo apt-get upgrade

sudo apt-get -y install cuda

Miniconda & Tensorflow

1. Open your favorite terminal.
2. Run wsl to enter your Ubuntu instance.
3. Copy and paste the following commands into your WSL instance to install Miniconda and Tensorflow:

curl https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -o Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh
source ~/.bashrc

conda create --name tf python=3.9
conda activate tf
conda install -c conda-forge cudatoolkit=11.8.0
python3 -m pip install nvidia-cudnn-cu11==8.6.0.163

mkdir -p $CONDA_PREFIX/etc/conda/activate.d
echo 'CUDNN_PATH=$(dirname $(python -c "import nvidia.cudnn;print(nvidia.cudnn.__file__)"))' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
echo 'export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CONDA_PREFIX/lib/:$CUDNN_PATH/lib' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
source $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh

pip install --upgrade pip
pip install tensorflow==2.12.*

4. Verify the tensorflow installation by running python3 -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))" in your WSL instance.
5. For more information visit this website.

Jupyter

1. Open your favorite terminal
2. Run wsl --setdefault Ubuntu to set your Ubuntu instance as your default WSL instance.
3. Run wsl to enter the Ubuntu instance.
4. Run the following commands to setup your Jupyter server:

conda activate tf
pip install jupyter
jupyter notebook --generate-config
printf "c.NotebookApp.token = 'secret'\nc.NotebookApp.password = ''\nc.NotebookApp.open_browser = False\nc.NotebookApp.ip = '0.0.0.0'\nc.NotebookApp.port = 8888\n" >> ~/.jupyter/jupyter_notebook_config.py

Usage

1. Run wsl to enter your Ubuntu WSL instance.
2. Run the following commands to start your Jupyter server:

conda activate tf
# Replace YOUR_USERNAME with your own.
jupyter notebook --notebook-dir='/mnt/c/Users/YOUR_USERNAME/Documents/code/python/tensorflow'

3. Open Visual Studio Code
4. Open a Jupyter Notebook in Visual Studio Code (make sure you have the Python and Jupyter extensions installed).
5. Click on Select Kernel on the top-right side of the notebook.
6. Click on Select Another Kernel...
7. Click on Existing Jupyter Server...
8. Click on Enter the URL of the running Jupyter server
9. Copy and paste the following url: http://localhost:8888?token=secret and press Enter.