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 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 {}:{}')
$ export CUDA_SRCS="-v /usr/local/cuda:/usr/local/cuda -v /usr/share/nvidia:/usr/share/nvidia"
$ docker run -it $CUDA_SO $CUDA_SRCS $DEVICES b.gcr.io/tensorflow/tensorflow-full-gpu

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

This one requires no extra setup -- just

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

This one requires a copy of the tensorflow source tree at ./tensorflow (since we don't keep the Dockerfiles at the top of the tree). With that in place, just run

$ git clone https://github.com/tensorflow/tensorflow
$ docker build -t $USER/tensorflow-full -f Dockerfile.cpu .

This one requires a few steps, since we need the NVidia headers to be available during the build step, but we don't want them included in the final container image. We need to start by installing the NVidia libraries as described in the CUDA setup instructions. With that complete, we can build via

$ cp -a /usr/local/cuda .
$ docker build -t $USER/tensorflow-gpu-base -f Dockerfile.gpu_base .
# Flatten the image
$ export TC=$(docker create $USER/tensorflow-gpu-base)
$ docker export $TC | docker import - $USER/tensorflow-gpu-flat
$ docker rm $TC
$ export TC=$(docker create $USER/tensorflow-gpu-flat /bin/bash)
$ docker commit --change='CMD ["/bin/bash"]'  --change='ENV CUDA_PATH /usr/local/cuda' --change='ENV LD_LIBRARY_PATH /usr/local/cuda/lib64' --change='WORKDIR /root' $TC $USER/tensorflow-full-gpu
$ docker rm $TC

This final image is a full TensorFlow image with GPU support.