tfml-env-pub

Step by step guide to create a TensorFlow 2.0, CUDA enabled, Python environment within WSL2.

In hopes of helping others subvert the trials and tribulations of setting up a fully function TenserFlow 2.0 environment, I decided to record my process and share it with the masses.

The goal of this project is to aid in the setup and configuration of a CUDA enabled TensorFlow 2.0 Machine Learning environment (TFML) within WSL2.

Requirements and Assumptions

1. A CUDA enabled NVIDIA GPU or above
   1. Check the CUDA Compatibility list for more info - https://developer.nvidia.com/cuda-gpus
   2. Recommended Compute capability of 5.0
2. A CUDA enabled NVIDIA driver is installed in Windows
   1. I have the NVIDIA Studio Driver | Version: 536.99 Release date: 08/08/2023
   2. Game Ready Drivers work as well
3. An Ubuntu WSL2 Distro is up and running in a supported Windows build
   1. Recommended: Ubuntu 22.04.3 LTS
   2. Recommended: 5.15.90.1-microsoft-standard-WSL2
4. Network/Internet is available within Windows and the WSL2 environment
5. A user was created during the deployment of Ubuntu
6. Basic understanding of ubuntu Linux Command Line, PowerShell and Python
   1. Nothing heavy is needed to get started

Primary environment info

1. TensorFlow version: 2.13.0
2. Keras version: 2.13.1
3. Python version: 3.11.4
4. ipykernel: 6.25.1
5. WSL: 5.15.90.1-microsoft-standard-WSL2
6. Ubuntu version: Ubuntu 22.04.3 LTS
7. CUDA NVCC version: 12.2.128
8. CUDA Toolkit version: 11.8.0
9. Jupyter Core version: 5.3.1
10. Conda version: 23.7.3

Full list of packages. https://github.com/BrandXX/tfml-env-pub/blob/master/tfml/conda-env-export.yaml

More about Tensorflow and Cuda

CUDA on WSL User Guide

1. CUDA on WSL User Guide

   1. https://docs.nvidia.com/cuda/archive/11.5.2/pdf/CUDA_on_WSL_User_Guide.pdf
   2. Also located at /docs/cuda_on_wsl_user_guide.pdf

2. Tensorflow Compatibility list

   1. https://www.tensorflow.org/install/source#gpu

3. Tensorflow's main site

   1. https://www.tensorflow.org/

4. NVIDIA's Developer site

   1. https://developer.nvidia.com/

More to come

Moving forward, I plan to release a WSL install guide, DirectML install guide, example projects, useful instructions, scripts, code snippets and more.