Google Brain Software Engineer Martin Wicke says a preview version of TensorFlow 2.0 will be released later this year. To cope with dramatic changes in both users and use-cases, TensorFlow 2.0 will shift its focus to “ease of use.” Wicke made the announcements yesterday in a Google Groups post.
We can expect the following features in TensorFlow 2.0:
- “Eager execution” through alignment of model expectations and practice
- Enhanced compatibility with platforms and languages
- Removal of deprecated APIs to reduce duplication
A series of upcoming public reviews explaining the planned changes will provide opportunities for the community to express their concerns and submit proposals. The Google team hopes to smoothen the transition from TensorFlow 1.0 to 2.0 by creating a conversion tool that makes existing Python code compatible with TensorFlow 2.0 APIs.
Wicke says TensorFlow 2.0 will cease distributing tf.contrib due to its overgrowth. He suggests that in the future, the contrib modules will either integrate a project into TensorFlow, be transported to an individual repository, or be deleted. As TensorFlow 2.0 will no longer add new tf.contrib projects, Wicke encouraged those currently working on tf.contrib projects to contact the team for assistance.
The update will not impact SavedModels or stored GraphDefs. During the update however, TensorFlow 2.0 might have to convert variable names in raw checkpoint to ensure they remain compatible.
TensorFlow is an open source software library for computation developed by the Google Brain team and released in 2015. Its robust machine learning framework has enabled broad usage across different platforms.