Swift for TensorFlow, a first-class language for machine learning

Swift for TensorFlow is a first-class programming language for machine learning. It combines the expressiveness of imperative, define-by-run programming models (e.g. PyTorch) with the performance of computation graphs (e.g. TensorFlow). Users of Swift can write machine learning programs with natural, imperative tensor code. Behind the scenes, the Swift compiler automatically extracts tensor code to build computation graphs. Swift also provides language-integrated automatic differentiation, static diagnostics, and Python interoperability.