Source code movement of tensorflow/tensorflow Docker images

[Austin Anderson]

Hi everyone,

TL;DR The tensorflow/tensorflow Docker containers have been rewritten, but should act the same as before. If you use these containers, you should not have to do anything. If their behavior has changed, please let me know in case it is a bug. You can try the new containers with the "nightly" tags as of 3/1, or the "2.12" tags starting with rc1 in the next few weeks.

The "tensorflow/tensorflow" Dockerfiles have been simplified and moved from the tensorflow/tensorflow GitHub repository into tensorflow/build. They are still published to tensorflow/tensorflow on Docker Hub. Here's an overview of the Docker containers now supported by the TensorFlow DevInfra team (my team)

• For the tensorflow/build Docker images, the source code is in the tensorflow/tensorflow GitHub repository. These containers include the toolchain to compile TensorFlow. My team uses these at Google to build TensorFlow on Linux.
• For the tensorflow/tensorflow Docker images, the source code is in the tensorflow/build GitHub repository. These containers are a good way to try out TensorFlow. We publish variants with GPU support and Jupyter. My team does not use these containers directly. 
• Note that the "devel" tags were replaced by the "tensorflow/build" containers above, and now we only publish the versioned TensorFlow tags, not "devel."
• Note that the default tags do not include GPU support. This is inverted from the TensorFlow Python packages.
• These containers only support x86 and Nvidia CUDA. If you would like to maintain the equivalent Dockerfiles for other platforms, please fork the directory and create a PR in SIG Build. Take a look at the contribution guidelines on the SIG Build repository.

Unfortunately, the rearranging means that the containers are named opposite to where their source code is: the "SIG Build" containers were developed externally and then adopted internally while the "TensorFlow" containers were originally developed internally but then extracted. Their behavior has not changed, however.

Thank you,

Austin Anderson

TF DevInfra