An End-to-End DNN Inference Framework for the SpiNNaker2 Neuromorphic MPSoC

TL;DR

This paper introduces a comprehensive framework that enables efficient end-to-end inference of large DNNs, including transformers, on the SpiNNaker2 neuromorphic chip, bridging PyTorch models to hardware execution.

Contribution

It extends OctopuScheduler with a multi-layer DNN scheduling framework and a front-end for quantization and lowering, facilitating edge-based inference on SpiNNaker2.

Findings

Supports large and complex DNNs including transformers

Enables end-to-end inference from PyTorch to SpiNNaker2

Improves efficiency of neuromorphic DNN deployment

Abstract

This work presents a multi-layer DNN scheduling framework as an extension of OctopuScheduler, providing an end-to-end flow from PyTorch models to inference on a single SpiNNaker2 chip. Together with a front-end comprised of quantization and lowering steps, the proposed framework enables the edge-based execution of large and complex DNNs up to transformer scale using the neuromorphic platform SpiNNaker2.