TensorFlow 1.0 brings experimental APIs for Java and Go, XLA and more
The release of TensorFlow 1.0 was announced yesterday at the first annual TensorFlow Developer Summit. It is fast, more flexible and promises Python API stability. Let’s see what else it has to offer.
“When we open-sourced TensorFlow we were hoping to build a machine learning platform for everyone in the world,” Jef Dean, Google Senior Fellow in the Research Group said in his keynote at the first annual TensorFlow Developer Summit.
Rajat Monga, software engineer at Google announced TensorFlow 1.0 and highlighted its three greatest perks: it is fast, flexible and it is production-ready.
- Faster. According to the blog post announcing version 1.0, XLA lays the groundwork for even more performance improvements in the future, and tensorflow.org now includes tips & tricks for tuning your models to achieve maximum speed.
- More flexible. TensorFlow 1.0 introduces a high-level API for TensorFlow, with tf.layers, tf.metrics, and tf.losses modules. Google also announced the inclusion of a new tf.keras module that provides full compatibility with Keras, another popular high-level neural networks library.
- Production-ready. TensorFlow 1.0 promises Python API stability, thus making it easier to pick up new features without worrying about breaking your existing code.
TensorFlow 1.0 highlights
The experimental release of XLA, a domain-specific compiler for TensorFlow graphs, that targets CPUs and GPUs, is one of the most important highlights. Furthermore, Python APIs have been changed to resemble NumPy more closely. For this and other backwards-incompatible changes made to support API stability going forward, users are advised to use Google’s migration guide and conversion script.
The list of highlights also includes experimental APIs for Java and Go and new Android demos for object detection and localization, and camera-based image stylization. TensorFlow 1.0 also brings TensorFlow Debugger (tfdbg), a command-line interface and API for debugging live TensorFlow programs and Installation improvements: Python 3 docker images have been added, and TensorFlow’s pip packages are now PyPI compliant, which means TensorFlow can now be installed with a simple invocation of
pip install tensorflow.
The APIs in TensorFlow 1.0 have changed in ways that are not all backwards compatible — the goal is to ensure an internally-consistent API. Google has made it clear that they do not plan to make backwards-breaking changes throughout the 1.N lifecycle.
Therefore, they have published a guide which walks users through the major changes in the API and teaches them how to automatically upgrade programs for TensorFlow 1.0.
Check out the release notes here.