NVIDIA-docker Cheatsheet

TensorFlow Docker requirements

1. Install Docker on your local host machine.
2. For GPU support on Linux, install nvidia-docker.

Note: To run the docker command without sudo, create the docker group and add your user. For details, see the post-installation steps for Linux.

Download a TensorFlow Docker image

The official TensorFlow Docker images are located in the tensorflow/tensorflow Docker Hub repository. Image releases are tagged using the following format:

Tag	Description
latest	The latest release of TensorFlow CPU binary image. Default.
nightly	Nightly builds of the TensorFlow image. (unstable)
version	Specify the version of the TensorFlow binary image, for example: 1.14.0
devel	Nightly builds of a TensorFlow master development environment. Includes TensorFlow source code.

Each base tag has variants that add or change functionality:

Tag Variants	Description
tag-gpu	The specified tag release with GPU support. (See below)
tag-py3	The specified tag release with Python 3 support.
tag-jupyter	The specified tag release with Jupyter (includes TensorFlow tutorial notebooks)

You can use multiple variants at once. For example, the following downloads TensorFlow release images to your machine:

docker pull tensorflow/tensorflow                     # latest stable release
docker pull tensorflow/tensorflow:devel-gpu           # nightly dev release w/ GPU support
docker pull tensorflow/tensorflow:latest-gpu-jupyter  # latest release w/ GPU support and Jupyter

 

Start a TensorFlow Docker container

To start a TensorFlow-configured container, use the following command form:

docker run [-it] [--rm] [-p hostPort:containerPort] tensorflow/tensorflow[:tag] [command]

 

For details, see the docker run reference.

Examples using CPU-only images

Let's verify the TensorFlow installation using the latest tagged image. Docker downloads a new TensorFlow image the first time it is run:

docker run -it --rm tensorflow/tensorflow \
   python -c "import tensorflow as tf; tf.enable_eager_execution(); print(tf.reduce_sum(tf.random_normal([1000, 1000])))"

 

Success: TensorFlow is now installed. Read the tutorials to get started.

Let's demonstrate some more TensorFlow Docker recipes. Start a bash shell session within a TensorFlow-configured container:

docker run -it tensorflow/tensorflow bash

 

Within the container, you can start a python session and import TensorFlow.

To run a TensorFlow program developed on the host machine within a container, mount the host directory and change the container's working directory (-v hostDir:containerDir -w workDir):

docker run -it --rm -v $PWD:/tmp -w /tmp tensorflow/tensorflow python ./script.py

 

Permission issues can arise when files created within a container are exposed to the host. It's usually best to edit files on the host system.

Start a Jupyter Notebook server using TensorFlow's nightly build with Python 3 support:

docker run -it -p 8888:8888 tensorflow/tensorflow:nightly-py3-jupyter

 

Follow the instructions and open the URL in your host web browser: http://127.0.0.1:8888/?token=...

GPU support

Docker is the easiest way to run TensorFlow on a GPU since the host machine only requires the NVIDIA® driver (the NVIDIA® CUDA® Toolkit is not required).

Install nvidia-docker to launch a Docker container with NVIDIA® GPU support. nvidia-docker is only available for Linux, see their platform support FAQ for details.

Check if a GPU is available:

lspci | grep -i nvidia

 

Verify your nvidia-docker installation:

docker run --runtime=nvidia --rm nvidia/cuda nvidia-smi

 

Note: nvidia-docker v1 uses the nvidia-docker alias, where v2 uses docker --runtime=nvidia.

Examples using GPU-enabled images

Download and run a GPU-enabled TensorFlow image (may take a few minutes):

docker run --runtime=nvidia -it --rm tensorflow/tensorflow:latest-gpu \
   python -c "import tensorflow as tf; tf.enable_eager_execution(); print(tf.reduce_sum(tf.random_normal([1000, 1000])))"

 

It can take a while to set up the GPU-enabled image. If repeatably running GPU-based scripts, you can use docker execto reuse a container.

Use the latest TensorFlow GPU image to start a bash shell session in the container:

docker run --runtime=nvidia -it tensorflow/tensorflow:latest-gpu bash