Changes to TensorFlow Docker images for Python 2 EOL

40 views
Skip to first unread message

Austin Anderson

unread,
Jan 27, 2020, 5:06:37 PM1/27/20
to devel...@tensorflow.org
Hello everyone,

As previously announced, TensorFlow no longer supports Python 2. As such, we're making some changes to our Docker image tags. Currently, we publish Python 2 images without named tag annotations, and Python 3 images with "-py3" tag annotations.

Starting from today, January 27, 2020, new standard images (nightly-, latest-, 2.xx-, 1.15.x-) will be published with Python 3 under both tag annotations. For example, "tensorflow:nightly" and "tensorflow:nightly-py3" will both run Python 3. This will not affect versioned, previously released images. Note that this will make the Python 2 nightly images unavailable, as Python 2 tf-nightly packages are no longer published.

Starting from February 3, 2020, "devel" and "devel-gpu" will be published with Python 3 under both tags. For example, "tensorflow:devel" and "tensorflow:devel-py3" will both run Python 3. The exact timing of this depends on the build status of TensorFlow at HEAD, as we only deploy new images when TensorFlow builds successfully.

Starting on April 2, 2020, we will stop publishing new images tagged as "-py3". If you regularly use explicit Python 3 tags in CI, please be sure to change them to non-"py3" tags before that date to avoid breakages.

Thank you,
Austin Anderson
TensorFlow DevInfra

Austin Anderson

unread,
Feb 4, 2020, 2:03:49 PM2/4/20
to TensorFlow Developers, Austin Anderson
As of this morning, all of our "devel" and "devel-gpu" containers ship with Python 3.6.9, as do the 1.15.2 and nightly containers. Since there has not been a new release of TensorFlow 2.X since this change was announced, the "latest" tags are still running Python 2.7.

Austin Anderson

unread,
Apr 10, 2020, 9:03:03 PM4/10/20
to TensorFlow Developers
"-py3" tags are no longer published. TensorFlow 2.2 will be the first full release without any "-py3" tags; this means that "latest-py3" will always be stuck at an older TensorFlow version.
Reply all
Reply to author
Forward
0 new messages