docker-compose-tensorflow

Abstract

This project provides example configurations of docker compose for TensorFlow.

• docker-compose.yml

  • Single container

• distributed.yml

  • One parameter server and two worker

This project also provides Dockerfile, which is referenced in the above docker compose configuration. You can install Keras by using this Dockerfile.

How to use

First, change or delete 'volumes' configuration. This configuration is just an example.

Then, start containers using docker-compose command.

Using nvidia-docker-compose (Recommended)

You can use https://github.com/eywalker/nvidia-docker-compose to use GPUs in these examples.

e.g.

$ nvidia-docker-compose -f distributed.yml up -d

The GPU devices are specified in distributed.yml as an example. You can change the configuration about devices.

If you started containers by the above example, you can attach a terminal to contaners.

e.g.

$ nvidia-docker-compose -f distributed.yml exec ps /bin/bash