This is the nvidia GPU environment instruction. Some premises are the following.

• Ubuntu-LTS(debian) based virtual machine.
• Aiming for deeplearning with python environment.
• RemoteSSH with VScode.
• Before this instruction, you need to install the nvidia driver.

Preparation

GitHub key registration & setup

$ ssh-keygen -t ed25519 -C "<email>"
$ eval "$(ssh-agent -s)"
$ vi ~/.ssh/config

Edit the ssh config file as below.

~/.ssh/config

Host * 
    AddKeysToAgent yes
    IdentityFile ~/.ssh/id_ed25519

$ ssh-add -k ~/.ssh/id_ed25519

Clone the repository

Clone this repository into your Ubuntu based linux environment.

$ git clone git@github.com:makkimaki/exp_docker_env.git

After cloning this repo into local env.

• prepare the .ssh directory
• copy the private key into this .ssh directory
• modify .bashrc like this: ssh-add -k ~/.ssh/<private key name>

exp_docker_env

Prerequisites

• docker-compose >= 1.29.x
  • In order to use docker-compose.yml runtime: nvidia option!

step

1. build the image using this directory ($ docker build)
2. run the container ($ docker run or $ docker-compose)

docker command based setting

directory setting

Host server

• under construction

Container server

• under construction

build image

$ docker build -t <tag name> .

When it is finished, you can see the named tag via $docker images.

run the container

$ docker run --rm --gpus all -v ~/work:/work -p <host port>:22 -it <container name> bash

docker-compose command based setting

Instead of executing the $docker command, you can use docker-compose based environment building. The version is expected to have above 1.29.x as following.

$ docker-compose --version
docker-compose version 1.29.1, build c34c88b2

build image

After switching to the exp_docker_env directory,

$ docker-compose build 

run the container

$ docker-compose up -d 

If you still don't have any built image, this operation will take you from building the container image to running it.

confirm the container (optional)

$ docker-compose ps

You can find the tagged container.

Now you can SSH to the container!

Other

Virtual Environment (Inside the docker container)

It uses "conda" virtual environment.

make new env.

conda create python=3.9 --name <env. name>

activation & deactivation

activation

source activate <env. name>

deactivation

conda deactivate