Google has released version 1.0 of TensorFlow, its open-source framework for scalable machine learning. The big highlights of this release were announced at Google’s TensorFlow Dev Summit in Mountain View, Calif., with a focus on artificial intelligence and artificial neural networks.

According to its release notes, version 1.0 includes experimental XLA, which is a domain-specific compiler that optimizes TensorFlow computations, targeting CPUs and GPUs. There’s also a new experimental Java API, improvements to the TensorFlow Debugger, and several Python API calls that have been changed to resemble NumPy more closely.

(Related: Google announces TensorFlow Fold)

XLA lays the groundwork for more performance improvements in the near future, according to the Google Developer Blog, and soon the team will publish updated implementations of models so developers will know how to take advantage of TensorFlow 1.0.

TensorFlow also includes some breaking changes to the API. The TensorFlow/models have been moved to a separate GitHub repository. To upgrade the existing TensorFlow Python code to match these API changes, developers can use this prepared conversion script.

“The APIs in TensorFlow 1.0 have changed in ways that are not all backward compatible,” according to the TensorFlow 1.0 transition page. “That is, TensorFlow programs that worked on TensorFlow 0.n won’t necessarily work on TensorFlow 1.0. We have made this API changes to ensure an internally-consistent API, and do not plan to make backward-breaking changes throughout the 1.n life cycle.”

According to Google, TensorFlow 1.0 also includes a new tf.keras module that provides full compatibility with Keras, a neural networks library.

Other highlights include the new Android demos for object detection and localization, camera-based image stylization, and installation improvements like Python 3 Docker images.