DL4S Accelerated Tensor Operations and Dynamic Neural Networks with Reverse-mode Automatic Differentiation for Swift Devices(watchOS to Linux)

DL4S provides high-level APIs for many common accelerated operations in neural networks and deep learning. Additionally, it features built-in automatic differentiation, enabling the creation and training of neural networks without manually implementing backpropagation—and without requiring a special Swift toolchain. Key features include implementations of various basic binary and unary operators, broadcasting, matrix operations, convolutions, recurrent neural networks, commonly used optimizers, second-order derivatives, and more. DL4S also offers implementations for common network architectures like VGG, AlexNet, ResNet, and Transformers. While DL4S is primarily aimed at deep learning and optimization, it can also be used as a library for vectorized mathematical operations (such as numpy). To install on iOS, tvOS, or macOS using Xcode, select 'File' > 'Swift Packages' > 'Add Package Dependency', enter the Package URL: https://github.com/palle-k/DL4S.git, and click 'Next'. Select the desired branch, and you're good to go.