Google has announced that Swift for TensorFlow is now officially an open source project on GitHub. Swift for TensorFlow was initially announced and demoed last month at the TensorFlow Developer Summit.
“Our approach is a new and different way to use TensorFlow, opening new design opportunities and new avenues for solving existing problems. Though the project is in early development, we’ve decided to open-source it and move our design discussions to a public mailing list so anyone interested in the project can get involved,” the TensorFlow team wrote
According to the company, Swift for TensorFlow combines the performance of graphs and the flexibility of eager execution. Eager executions allows operations to be executed immediately when they are called from Python. It also has a strong focus on improving usability at every level of the programming stack.
The company stressed that Swift for TensorFlow is not just a TensorFlow API wrapper that is written in Swift, but includes compiler and language enhancements in Swift to benefit machine learning developers.
According to the TensorFlow team, the cornerstone of the project is an algorithm called Graph Program Extraction, which allows users to write code in eager-execution style and still keep all of the benefits of graphs.
The design also supports advanced automatic differentiation in Swift as well as Python integration in Swift, allowing developers to use Python APIs directly Swift code.
Google is offering detailed documentation to help users get started. It recommends starting with the Swift for TensorFlow Design Overview and then moving on to some of the other documents that go into more detail.
