Google today announced the final release of TensorFlow 2.0. This follows the platform’s alpha release at the TensorFlow Dev Summit in June.
Google Brain debuted TensorFlow in 2015, and it soon became the world’s most popular open-source machine learning library — “a comprehensive ecosystem of tools for developers, enterprises, and researchers who want to push the state-of-the-art in machine learning and build scalable ML-powered applications.”
As with the previous update, the final release adds a number of major features and improvements:
- Easy model building with Keras and eager execution.
- Robust model deployment in production on any platform.
- Powerful experimentation for research.
- API simplification by reducing duplication and removing deprecated endpoints.
Google says integrating Keras tightly into TensorFlow along with with eager execution and Pythonic function execution will make the application development experience “as familiar as possible” for Python developers.
Machine learning researchers will also benefit from a low-level API which enables them to export internally used ops and continue to build models onto the internals of TensorFlow without having to rebuild TensorFlow.
TensorFlow 2.0 also promises better performance on GPU acceleration. It delivers three times faster training performance when using mixed precision on Volta and Turing GPUs, and better usability and high performance inferencing on NVIDIA T4 Cloud GPUs on Google Cloud.
Although the official TensorFlow 2.0 release announcement offered no timetable for possible future updates, Google invited developers and researchers to join the conversation at their TensorFlow World gathering, which runs Oct 28 to 31 in Santa Clara, CA.
Journalist: Fangyu Cai | Editor: Michael Sarazen