Skip to content

Files

Latest commit

40dbd9d · Dec 10, 2024

History

History

tensorflow_runtime_dockerfiles

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
Mar 2, 2023
Feb 7, 2023
Jul 14, 2023
Mar 24, 2023
Feb 2, 2023
Nov 20, 2023
Nov 17, 2023
Feb 2, 2023
Feb 2, 2023
Feb 2, 2023
Feb 2, 2023
Feb 2, 2023
Feb 2, 2023
Mar 24, 2023
Dec 10, 2024
Feb 2, 2023
Feb 2, 2023
Mar 14, 2024
Mar 14, 2024
Mar 12, 2024
Mar 12, 2024

TensorFlow Runtime Dockerfiles

Simple Dockerfiles for running TensorFlow, with Jupyter and GPU variants.

Maintainer: @angerson (TensorFlow, SIG Build)


These containers are built by an internal job at Google and published to tensorflow/tensorflow on Docker Hub. Here's a quick way to try out TensorFlow with GPU support and Jupyter:

docker run --gpus=all -it --rm -v $(realpath ~/notebooks):/tf/notebooks -p 8888:8888 tensorflow/tensorflow:nightly-gpu-jupyter

Refer to the tensorflow.org Docker installation instructions for more details.

Building Containers

Builds are straightforward. Here's a sample:

docker build --target=base --build-arg TENSORFLOW_PACKAGE=tf-nightly-cpu -t tensorflow-nightly -f cpu.Dockerfile .

Look at the Dockerfiles for full details.

The builds include very simple import tests to verify that the packages work. You can run the tests like so:

docker build --target=test --build-arg TENSORFLOW_PACKAGE=tf-nightly-cpu -f cpu.Dockerfile .
docker build --target=base --build-arg TENSORFLOW_PACKAGE=tf-nightly-cpu -t tensorflow-nightly -f cpu.Dockerfile .

The test layer starts from the base layer, so the second command will complete instantly.

Contributions

If you would like to contribute a small change, please make a pull request. For large changes such as support for additional platforms, please clone this directory into a new directory and update the README to indicate that you are the new maintainer.