Swift for TensorFlow: No boundaries.
Swift for TensorFlow is a next-generation platform for machine learning, incorporating the latest research across machine learning, compilers, differentiable programming, systems design, and beyond. This is an early-stage project: it is not feature-complete nor production-ready, but it is ready for pioneers to try in projects, give feedback, and help shape the future!
The Swift for TensorFlow project is currently focusing on 2 kinds of users:
-
Advanced ML researchers who are limited by current ML frameworks. Swift for TensorFlow's advantages include seamless integration with a modern general-purpose language, allowing for more dynamic and sophisticated models. Fast abstractions can be developed in "user-space" (as opposed to in C/C++, aka "framework-space"), resulting in modular APIs that can be easily customized.
-
ML learners who are just getting started with machine learning. Thanks to Swift's support for quality tooling (e.g. context-aware autocompletion), Swift for TensorFlow can be one of the most productive ways to start learning the fundamentals of machine learning.
-
Google Colaboratory: The fastest way to get started is to try out Swift for TensorFlow right in your browser. Just open up a tutorial, or start from a blank notebook! Read more in our usage guide.
-
Install locally: You can download a pre-built Swift for TensorFlow package. After installation, you can follow these step-by-step instructions to build and execute a Swift script on your computer.
-
Run on GCP: You can spin up a GCE instance using a Swift for TensorFlow Deep Learning VM image, with all drivers and the toolchain pre-installed. Instructions can be found in the Installation Guide.
-
Compile from source: If you'd like to customize Swift for TensorFlow or contribute back, follow our instructions on building Swift for TensorFlow from source.
| Tutorial | Last Updated |
|---|---|
| A Swift Tour | March 2019 |
| Protocol-Oriented Programming & Generics | August 2019 |
| Python Interoperability | March 2019 |
| Custom Differentiation | March 2019 |
| Sharp Edges in Differentiability | November 2020 |
| Model Training Walkthrough | March 2019 |
| Raw TensorFlow Operators | December 2019 |
| Introducing X10, an XLA-Based Backend | May 2020 |
- Models and Examples
- TensorFlow Swift API Reference
- Release Notes
- Known Issues
- Frequently Asked Questions
- TensorFlow Blog Posts
Please join the [email protected] mailing list to hear the latest announcements, get help, and share your thoughts!
Swift for TensorFlow is a new way to develop machine learning models. It gives you the power of TensorFlow directly integrated into the Swift programming language. We believe that machine learning paradigms are so important that they deserve first-class language and compiler support.
A fundamental primitive in machine learning is gradient-based optimization:
computing function derivatives to optimize parameters. With Swift for
TensorFlow, you can easily differentiate functions using differential
operators like gradient(of:), or differentiate with respect to an entire
model by calling method gradient(in:). These differentiation APIs
are not just available for Tensor-related concepts—they are
generalized for all types that conform to the Differentiable
protocol, including Float, Double, SIMD vectors, and your own data
structures.
// Custom differentiable type.
struct Model: Differentiable {
var w: Float
var b: Float
func applied(to input: Float) -> Float {
return w * input + b
}
}
// Differentiate using `gradient(at:_:in:)`.
let model = Model(w: 4, b: 3)
let input: Float = 2
let (𝛁model, 𝛁input) = gradient(at: model, input) { model, input in
model.applied(to: input)
}
print(𝛁model) // Model.TangentVector(w: 2.0, b: 1.0)
print(𝛁input) // 4.0Beyond derivatives, the Swift for TensorFlow project comes with a sophisticated toolchain to make users more productive. You can run Swift interactively in a Jupyter notebook, and get helpful autocomplete suggestions to help you explore the massive API surface of a modern deep learning library. You can get started right in your browser in seconds!
Migrating to Swift for TensorFlow is really easy thanks to Swift's powerful Python integration. You can incrementally migrate your Python code over (or continue to use your favorite Python libraries), because you can easily call your favorite Python library with a familiar syntax:
import TensorFlow
import Python
let np = Python.import("numpy")
let array = np.arange(100).reshape(10, 10) // Create a 10x10 numpy array.
let tensor = Tensor<Float>(numpy: array) // Seamless integration!Beware: the project is moving very quickly, and thus some of these documents are slightly out of date as compared to the current state-of-the-art.
| Document | Last Updated | Status |
|---|---|---|
| Why Swift for TensorFlow? | April 2018 | Current |
| Swift for TensorFlow Design Overview | April 2018 | Outdated |
| Supported Backends | May 2020 | Current |
The Swift for TensorFlow project builds on top of powerful theoretical foundations. For insight into some of the underlying technologies, check out the following documentation.
| Document | Last Updated | Status |
|---|---|---|
| Swift Differentiable Programming Manifesto | January 2020 | Current |
| Swift Differentiable Programming Implementation Overview | August 2019 | Current |
| Swift Differentiable Programming Design Overview | June 2019 | Outdated |
| Differentiable Types | March 2019 | Outdated |
| Differentiable Functions and Differentiation APIs | March 2019 | Outdated |
| Dynamic Property Iteration using Key Paths | March 2019 | Current |
| Hierarchical Parameter Iteration and Optimization | March 2019 | Current |
| First-Class Automatic Differentiation in Swift: A Manifesto | October 2018 | Outdated |
| Automatic Differentiation Whitepaper | April 2018 | Outdated |
| Python Interoperability | April 2018 | Current |
| Graph Program Extraction | April 2018 | Outdated |
Compiler and standard library development happens on the main branch of the
apple/swift repository.
Additional code repositories that make up the core of the project include:
- Deep learning library: high-level API familiar to Keras users.
Swift for TensorFlow is no longer a fork of the official Swift language; development was previously done on the
tensorflowbranch of the apple/swift repository. Language additions were designed to fit with the direction of Swift and are going through the Swift Evolution process.
Jupyter Notebook support for Swift is under development at google/swift-jupyter.
tensorflow/swift-models is a repository of machine learning models built with Swift for TensorFlow. It intended to provide examples of how to use Swift for TensorFlow, to allow for end-to-end tests of machine learning APIs, and to host model benchmarking infrastructure.
fastai/swiftai is a high-level API for Swift for TensorFlow, modeled after the fastai Python library.
Swift for TensorFlow discussions happen on the [email protected] mailing list.
Before reporting an issue, please check the Frequently Asked Questions to see if your question has already been addressed.
For questions about general use or feature requests, please send an email to the mailing list or search for relevant issues in the JIRA issue tracker.
For the most part, the core team's development is also tracked in JIRA.
We welcome contributions from everyone. Read the contributing guide for information on how to get started.
In the interest of fostering an open and welcoming environment, we as contributors and maintainers pledge to making participation in our project and our community a harassment-free experience for everyone, regardless of age, body size, disability, ethnicity, gender identity and expression, level of experience, education, socio-economic status, nationality, personal appearance, race, religion, or sexual identity and orientation.
The Swift for TensorFlow community is guided by our Code of Conduct, which we encourage everybody to read before participating.