AI: Deep Learning 02

• DL Environment setup: WIN10 + WSL2 + Docker + Ubuntu 20.04

https://weblogs.asp.net/dixin/setup-and-use-cuda-and-tensorflow-in-windows-subsystem-for-linux-2

Tensorflow 2

Cuda on WSL Nvidia driver download

Perparation

1. Turn Windows features on

• Win → Settings → App → Programs and Features → Turn Windows features on/off
  • Hyper-V
  • Windows Subsystem for Linux
  • Virtual Machine Platform
• Restart

1. Participate in the Windows 10 Insider Program

• Win → Settings → Update and Security → Windows Insider Programs

• Windows Update → Update (Takes long time)
• Restart

1. Install Windows terminal

• Microsoft Store → search:Terminal → install
• When running terminal, always run as admin.

1. Install WSL2

https://docs.microsoft.com/en-us/windows/wsl/install-win10

> dism.exe /online /enable-feature /featurename:Microsoft-Windows-Subsystem-Linux /all /norestart
> dism.exe /online /enable-feature /featurename:VirtualMachinePlatform /all /norestart

• Restart

• wsl version must be 4.9. 121 or higher

  > wsl uname -r
  4.9. 121
  

  If smaller than 121 → > wsl –update

• Install Ubuntu 20.04 in Microsoft Store

> wsl -l -v
  NAME                   STATE           VERSION
* Ubuntu-20.04           Stopped         2

If VERSION is 1 → > wsl –set-default-version 2

Restart wsl to make sure version is updated to 2 → > wsl -t Ubuntu-20.04

1. Install NVIDIA graphics driver

• On terminal,

> wsl --shutdown

• Download Geforce driver for Windows 10

https://developer.nvidia.com/cuda/wsl/download

1. Update wsl

$ sudo apt update
$ sudo apt -yV upgrade

1. Get python development files

$ sudo apt -y install python3-dev python3-pip python3-setuptools python3-numpy

1. Install NVIDIA CUDA toolkit

• NVIDIA graphics driver download
• NVIDIA CUDA toolkit download
• 10.0 is specified as the version of NVIDIA CUDA in TensorFlow 1.15 GPU version . The basis is the next page.(https://www.tensorflow.org/install/source#common_installation_problems)
• Do not use a version higher than the specified version: On Ubuntu 20.04 , CUDA 10.1 will be installed.

wget https://developer.download.nvidia.com/compute/cuda/10.1/Prod/local_installers/cuda_10.1.243_418.87.00_linux.run
sudo sh cuda_10.1.243_418.87.00_linux.run

1. Install pyTorch

• Check python version

$ python --version
Python 3.8.5

$ pip3 install torch==1.7.1+cu101 torchvision==0.8.2+cu101 torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html