High-level library overview

Typed neural networks live demo

Library internals

Functional differentiable programming

A library for tensor math and neural networks for functional programmers. Use algebraic data types to represent models, function composition to build neural network composition, generics to automate boilerplate, or (optionally) type-level computations to type neural network dimensions.

Haskell logo

Powered by libtorch

libtorch is the C++ library underlying PyTorch, a powerful neural network library for Python. We use code generated bindings to bring thousands of functions and make them available to Haskell, along with support for GPUs and automatic differentiation for all operations.

Generic placeholder image

Join the community

We're always looking for new collaborators and contributors who share an interest in making an impact in how machine learning systems are implemented and researched. Join us by emailing hasktorch@gmail.com to request a slack invite.

Generic placeholder image