While native Rust deep learning libraries have made progress, they are still largely experimental. However, there are high-quality bindings available for both Tensorflow and PyTorch.
Neural Networks
tflite [ crate · repo · docs ]
Rust bindings for TensorFlow Lite
rustml [ crate · repo · docs ]
A library for doing maching learning in Rust.
tensorflow [ crate · repo · docs ]
Rust language bindings for TensorFlow.
tch [ crate · repo · docs ]
Rust wrappers for the PyTorch C++ api (libtorch).
tract [ crate · repo · docs ]
Tiny, no-nonsense, self contained, TensorFlow and ONNX inference
fann [ crate · repo · docs ]
Wrapper for the Fast Artificial Neural Networks library
cogent [ crate · repo · docs ]
Basic neural network library for classification.
autograd [ crate · repo · docs ]
Tensors and differentiable operations in Rust
auto-diff [ crate · repo · docs ]
A neural network library in Rust.
rusty-machine [ crate · repo · docs ]
A machine learning library.
juice [ crate · repo · docs ]
Machine Learning Framework for Hackers
autograph [ crate · repo · docs ]
A machine learning library for Rust.
tvm [ crate · repo · docs ]
Rust frontend support for TVM
alumina [ crate · repo · docs ]
An Experimental Deep Learning Library
neuronika [ crate · repo · docs ]
Tensors and dynamic neural networks.
prophet [ crate · repo · docs ]
A neural network implementation with a focus on cache-efficiency and sequential performance.
nn [ crate · repo · docs ]
A multilayer feedforward backpropagation neural network library
cntk [ crate · repo · docs ]
Wrapper around Microsoft CNTK library
leaf [ crate · repo · docs ]
Machine Learning Framework for Hackers
neuroflow [ crate · repo · docs ]
The neural network library implemented in Rust
revonet [ crate · repo · docs ]
Rust implementation of real-coded genetic algorithm for solving optimization problems and training of neural networks. The latter is also known as neuroevolution.
drug [ crate · repo · docs ]
A differentiable computation graph for neural networks.
hal-ml [ crate · repo · docs ]
HAL: a machine learning library that is able to run on Nvidia, OpenCL or CPU BLAS based compute backends. It currently provides stackable classical neural networks, RNN's and soon to be LSTM's. A differentiation of this package is that we are looking to implement RTRL (instead of just BPTT) for the recurrent layers in order to provide a solid framework for online learning. We will also (in the future) be implementing various layers such as unitary RNN's, NTM's and Adaptive Computation time based LSTM's. HAL also comes with the ability to plot and do many basic math operations on arrays.
Looking for something you didn't find? Try asking on Zulip chat.