Skip to content

Build Larq Compute Engine

The Larq Compute Engine (LCE) repository consists of two main components:

  • LCE Runtime: which is a collection of highly optimized TensorFlow Lite custom operators.

  • LCE Converter: which takes a Larq model and generates a TensorFlow Lite FlatBuffer file (.tflite) compatible with the LCE runtime.

Important

Make sure to checkout the git tag matching the version of the LCE converter used to convert the model: 

git checkout v0.7.0
If you have installed the converter from PyPI you can check the currently installed version using: 
pip freeze | grep larq-compute-engine

Setup the build environment

Bazel is the primary build system for LCE. However, to avoid Bazel compatibility issues, you need to use Bazelisk as a launcher for Bazel.

To install Bazelisk on Linux, run the following two commands (replace v1.6.1 with your preferred bazelisk version):

sudo wget -O /usr/local/bin/bazel \
    https://github.com/bazelbuild/bazelisk/releases/download/v1.11.0/bazelisk-linux-amd64
sudo chmod +x /usr/local/bin/bazel

To install Bazelisk on MacOS, run: 

brew install bazelisk

Build LCE Runtime

The LCE runtime has a diverse platform support, covering Android and ARM-based boards such as Raspberry Pi. To build/install/run LCE runtime on each of these platforms, please refer to the corresponding guide.

Build LCE Converter

The LCE converter is available on PyPI and can be installed with Python's pip package manager:

pip install larq-compute-engine

To build the LCE pip package yourself, refer to building the converter.