Expanding the testing infrastructure of High-Level Synthesis for Machine Learning (HLS4ML)

HLS4ML - an open-source package based on High-Level Synthesis (HLS) is used for converting machine learning algorithms to utilize them in particle physics. It is designed to deploy neural network architectures on FPGA chips, targeting extremely low-latency requirements of the Large Hadron Collider (LHC). The main aim of the project is implementing an infrastructure for testing the conversion of Keras library with High-Level Synthesis for Machine Learning (HLS4ML). For building this infrastructure Pytest testing framework and Jenkins continuous integration platform was utilized. The designed testing infrastructure serves to maintain and to validate the consistency and the functionality of the software when new features are added by means of Keras library. The result of the work makes the project a more robust and mature framework, allowing the deployment of more complex neural network architectures. The testing infrastructure has to check the accuracy of model conversion, internal representation, HLS synthesis, and resource estimation.