Follow the instructions

• Docker Engine
• docker group

Follow the instructions here

• nvidia/cuda

Modify files. Be careful os version and python version.
(ex. Ubuntu 20.04 -> python 3.9, Ubuntu 22.04 -> python 3.10)
See here or here

• Modify docker_template.env

  • COMPOSE_PROJECT_NAME : project name
  • UID : UID
  • GID : GID
  • USER_NAME : user name used in container
  • WORKDIR_CONTAINER : container WORKDIR, directory where local WORKDIR mounted to
  • WORKDIR_LOCAL : local WORKDIR, directory mounted to container
  • BASE_IMAGE : Docker Image
  • PYTHON_VERSION : Python version
  • IMAGE_LABEL : label to cache docker image with label, if exist, load, if not, build.
  • CONTAINER_NAME : CONTAINER NAME, used to get in the container
  • HOST_PORT : HOST_PORT, use port not used at other containers
  • CONTAINER_PORT : CONTAINER_PORT, use port not used at other containers

  Commands to search your UID, GID

  id -u # UID
  id -g # GID

• docker-compose.yaml Change image, container_name, volumes, shm_size if needed.

  • image : name of image cached to local. if exist, load, if not, build.
  • container_name : name used to get in the container.
  • volumes : Correspondence. {local_dir}:{container_dir}
  • shm_size : shared memory size. check your spec.

• Dockerfile Change apt libraries, Pytorch.

• Docker compose up
  Use command at the directory where docker-compose.yaml is

  docker compose up -d

  If Dockerfile changed, Docker compose up with build

  docker compose up -d --build

• Execute command in Docker

  docker exec -it {container_name} bash
  # or
  docker exec -it -w {WORK_DIR_PATH} {container_name} bash
  # example
  docker exec -it template bash

  As root

  docker exec -it -u 0 -w {WORK_DIR_PATH} {container_name} bash

• Using JupyterLab (Optional)

  python -m jupyterlab --ip 0.0.0.0 --port {CONTAINER_PORT} --allow-root

• Using Tensorboard

  tensorboard --logdir=/workspace/PytorchLightning/lightning_logs --host=0.0.0.0 --port={CONTAINER_PORT}
  # or
  python /home/{USER}/.local/lib/python3.9/site-packages/tensorboard/main.py --logdir=/workspace/PytorchLightning/lightning_logs --host=0.0.0.0 --port={CONTAINER_PORT}

• Login W & D

  wandb login
  # or
  python3 -m wandb login
  # or
  /usr/bin/python3 -m wandb login

• Ueda's Sample
• Docker to Ubuntu
• About Rootless mode