Swift for TensorFlow was an experimental ML platform that incorporated differentiable programming capabilities directly into the Swift language, developed a mutable-value-semantics deep learning API, and fostered a model garden with over 30 deep learning models. The project spun off several related open-source efforts including PythonKit for Python interoperability and swift-jupyter for Jupyter notebook integration. The project was archived in February 2021, though its differentiable programming work continues in the official Swift compiler.