Binarized Neural Networks for efficient deep learning

Larq is an ecosystem of open-source Python packages for building, training and deploying Binarized Neural Networks to enable efficient inference on mobile and edge devices.

Get started with Larq

Deep learning with 1-bit weights and activations

Most neural networks use 32, 16 or 8 bits to encode each weight and activation, making them slow and power-hungry. Binarized Neural Networks (BNNs) restrict weights and activations to be only +1 or -1, and drastically reduce the model’s memory footprint and computational complexity.

End-to-end tools for developing BNNs

Larq lets engineers and researchers access state-of-the-art BNNs, train their own from scratch, and deploy them on mobile and edge devices.

Ready to use pretrained models

Larq Zoo provides implementations and pretrained weights for cutting-edge BNNs, allowing you to effortlessly start using efficient deep learning in your projects.

Intuitive and flexible extension of TensorFlow Keras

Larq is a powerful yet easy-to-use library for building and training BNNs that is fully compatible with the larger tf.keras ecosystem.

Simple deployment for the fastest inference

Larq Compute Engine is a highly-optimized inference library for deploying BNNs on mobile and edge devices.


Introduction to BNNs with Larq


Deploy your first BNN on Android

Read the docs