INSTALL DOCKER ON UBUNTU(if not exsit)

中文版 English

  docker version

DO IT MYSELF:pytorch_Ubuntu1604_docker

cat pytorch.Dockerfile >> .Dockerfile

docker build -t IMAGE_NAME[:TAG] -f Dockerfile .

---

the docker is based on the image from nvidia on dockerhub which includes ubuntu16.04,cuda9.0 and cudnn7.

Contains

• wget
• curl
• git
• openssh-server
• miniconda [base] pytorch numpy matplotlib opencv-python cython

other supplementary information

• code annotation
• example of docker common commands

AVOID REINVENTING THE WHEEL:dockerhub

docker pull iyzf/pytorch:cuda

• ubuntu 16.04
• cuda 9.0
• python 3.7
• pytorch 1.1

docker pull hexiffer/pytorch:stable

the stable version made by a Ph.D student in my lab

SOMETHING ELSE

offical tutorial key points :

• Image
• Container
• Repository

An example:run keras in docker

  docker pull floydhub/tensorflow:1.11-py3_aws.40 
  sudo docker run --name mykerasimg -p 6006:6006 -p 8888:8888  floydhub/tensorflow:1.11-py3_aws.40 # expose jupyter+tensorboard's port
  sudo docker stop mykerasimg 

the next time to run the image

sudo docker start mykerasimg # run jupyter automatically 
sudo docker exec -it mykerasimg bash #run a container with tty and interact
root@31c952a2e3ea:/# jupyter notebook list

>>> Currently running servers: http://localhost:8888/?token=balbalbalba

sudo docker stop mykerasimg

if you no longer use the image

 sudo docker container ls
 sudo docker rm  $container_ID$
 sudo docker ps -a 
 sudo docker image ls
 docker image rm $image_name$