Hermes: Memory-Efficient Pipeline Inference for Large Models on Edge Devices

TL;DR

Hermes introduces a memory-efficient pipeline framework for large model inference on edge devices, significantly reducing memory usage and inference latency compared to existing methods.

Contribution

The paper presents PIPELOAD, a novel dynamic memory management and parallel loading mechanism, integrated into Hermes, to improve large model inference on edge devices.

Findings

Up to 4.24x faster inference speed

86.7% lower memory consumption for BERT and ViT

2.58x faster inference for GPT models

Abstract

The application of Transformer-based large models has achieved numerous success in recent years. However, the exponential growth in the parameters of large models introduces formidable memory challenge for edge deployment. Prior works to address this challenge mainly focus on optimizing the model structure and adopting memory swapping methods. However, the former reduces the inference accuracy, and the latter raises the inference latency. This paper introduces PIPELOAD, a novel memory-efficient pipeline execution mechanism. It reduces memory usage by incorporating dynamic memory management and minimizes inference latency by employing parallel model loading. Based on PIPELOAD mechanism, we present Hermes, a framework optimized for large model inference on edge devices. We evaluate Hermes on Transformer-based models of different sizes. Our experiments illustrate that Hermes achieves up to 4.24 X increase in inference speed and 86.7% lower memory consumption than the state-of-the-art pipeline mechanism for BERT and ViT models, 2.58 X increase in inference speed and 90.3% lower memory consumption for GPT-style models.