← All repos

hasktorch

Tensors and neural networks in Haskell

GitHubhasktorch.org
Browse cluster: Haskell tooling and libraries →
4,522commits
49contributors
10languages

Tech stack & purpose

Hasktorch is a Haskell library for tensors and neural networks built on libtorch. The project enables users to build and train neural network models in Haskell, offering both dynamic tensors (which treat tensors as opaque types similar to PyTorch) and typed tensors (which track tensor dimensions at compile time to enforce dimension invariants). The library supports a wide range of neural network architectures and training techniques, including RNNs, CNNs, variational autoencoders, gaussian processes, and model distillation, along with capabilities for model serialization, optimization, and interoperability with PyTorch through torchscript and state dictionaries.

Community & reference links

Languages

Haskell
98.7%
C++
0.3%
Python
0.3%
HTML
0.2%
Nix
0.2%
Jupyter Notebook
0.2%
Shell
0.1%
CSS
0.0%
C
0.0%
Dockerfile
0.0%

Contributors (top 30 of 49)