README.md

Using SugarTensor through Docker

SugarTensor supports Docker to make it easy to running through Docker.

Installing and using Docker

For general information about Docker, see the Docker site. If you use GPUs, you'll need Nvidia-docker as well.

Download SugarTensor docker images

Get sugartensor docker image for GPU support

docker pull sugartensor/sugartensor

Get sugartensor docker image for CPU only support

docker pull sugartensor/sugartensor:latest-cpu

Running sugartensor container for shell console

Run sugartensor container with GPU support

nvidia-docker run -it sugartensor/sugartensor 

Run sugartensor container with CPU only support

docker run -it sugartensor/sugartensor:latest-cpu 

Setting docker to enable SSH connection to Docker container

In Ubuntu 14.04, add following line to the end of '/etc/default/docker' file.

DOCKER_OPTS="-H tcp://0.0.0.0:2376 -H unix:///var/run/docker.sock"

and restart docker service.

sudo service docker restart

Running sugartensor container for SSH server mode

Run sugartensor container with GPU support

nvidia-docker run -p 2222:22 -d sugartensor/sugartensor

Run sugartensor container with CPU only support

docker run -p 2222:22 -d sugartensor/sugartensor:latest-cpu 

Connect to the sugartensor SSH server container using ssh

Connect using the following command. ( pwd is 'ubuntu' ) Now, you can use your favorite IDE's remote debugging feature if your IDE supports remote debugging. ( To my knowledge, PyCharm professional and VisualStudio support remote python debugging )

ssh root@your_docker_host -p 2222

Testing sugartensor container

Run MNIST example in the '/root/sugartensor/example' directory.

python mnist_conv.py

Author

Namju Kim ([email protected]) at KakaoBrain Corp.