This directory contains Dockerfiles to make it easy to get up and running with TensorFlow via Docker.

General installation instructions are on the Docker site, but we give some quick links here:

• OSX: docker toolbox
• ubuntu

We currently maintain three Docker container images:

• b.gcr.io/tensorflow/tensorflow, which is a minimal VM with TensorFlow and all dependencies.

• b.gcr.io/tensorflow/tensorflow-full, which contains a full source distribution and all required libraries to build and run TensorFlow from source.

• b.gcr.io/tensorflow/tensorflow-full-gpu, which is the same as the previous container, but built with GPU support.

Each of the containers is published to a Docker registry; for the non-GPU containers, running is as simple as

$ docker run -it -p 8888:8888 b.gcr.io/tensorflow/tensorflow

For the container with GPU support, we require the user to make the appropriate NVidia libraries available on their system, as well as providing mappings so that the container can see the host's GPU. For most purposes, this can be accomplished via

$ export CUDA_SO=$(\ls /usr/lib/x86_64-linux-gnu/libcuda* | xargs -I{} echo '-v {}:{}')
$ export DEVICES=$(\ls /dev/nvidia* | xargs -I{} echo '--device {}:{}')
$ docker run -it -p 8888:8888 $CUDA_SO $DEVICES b.gcr.io/tensorflow/tensorflow-devel-gpu

Alternately, you can use the docker_run_gpu.sh script in this directory.

Just pick the dockerfile corresponding to the container you want to build, and run;

$ docker build -t $USER/tensorflow-suffix -f Dockerfile.suffix .