Deploying Machine Learning Models to Ahead-of-Time Runtime on Edge Using MicroTVM

TL;DR

This paper presents an end-to-end code generation approach using MicroTVM to deploy pre-trained machine learning models on edge devices with bare metal architecture, enabling efficient inference.

Contribution

It introduces a code generator that converts models into C source libraries for deployment on edge devices using MicroTVM, facilitating seamless inference execution.

Findings

Effective offloading of compute-intensive operators to accelerators

Successful deployment of gesture recognition on ARM Cortex M4F

Automated ahead-of-time C runtime generation enhances edge inference

Abstract

In the past few years, more and more AI applications have been applied to edge devices. However, models trained by data scientists with machine learning frameworks, such as PyTorch or TensorFlow, can not be seamlessly executed on edge. In this paper, we develop an end-to-end code generator parsing a pre-trained model to C source libraries for the backend using MicroTVM, a machine learning compiler framework extension addressing inference on bare metal devices. An analysis shows that specific compute-intensive operators can be easily offloaded to the dedicated accelerator with a Universal Modular Accelerator (UMA) interface, while others are processed in the CPU cores. By using the automatically generated ahead-of-time C runtime, we conduct a hand gesture recognition experiment on an ARM Cortex M4F core.