Folding Attention: Memory and Power Optimization for On-Device Transformer-based Streaming Speech Recognition

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Li, Yang, Lai, Liangzhen, Shangguan, Yuan, Iandola, Forrest N., Ni, Zhaoheng, Chang, Ernie, Shi, Yangyang, Chandra, Vikas
Natura: Preprint
Pubblicazione: 2023
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866929215248531456
author Li, Yang
Lai, Liangzhen
Shangguan, Yuan
Iandola, Forrest N.
Ni, Zhaoheng
Chang, Ernie
Shi, Yangyang
Chandra, Vikas
author_facet Li, Yang
Lai, Liangzhen
Shangguan, Yuan
Iandola, Forrest N.
Ni, Zhaoheng
Chang, Ernie
Shi, Yangyang
Chandra, Vikas
contents Transformer-based models excel in speech recognition. Existing efforts to optimize Transformer inference, typically for long-context applications, center on simplifying attention score calculations. However, streaming speech recognition models usually process a limited number of tokens each time, making attention score calculation less of a bottleneck. Instead, the bottleneck lies in the linear projection layers of multi-head attention and feedforward networks, constituting a substantial portion of the model size and contributing significantly to computation, memory, and power usage. To address this bottleneck, we propose folding attention, a technique targeting these linear layers, significantly reducing model size and improving memory and power efficiency. Experiments on on-device Transformer-based streaming speech recognition models show that folding attention reduces model size (and corresponding memory consumption) by up to 24% and power consumption by up to 23%, all without compromising model accuracy or computation overhead.
format Preprint
id arxiv_https___arxiv_org_abs_2309_07988
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Folding Attention: Memory and Power Optimization for On-Device Transformer-based Streaming Speech Recognition
Li, Yang
Lai, Liangzhen
Shangguan, Yuan
Iandola, Forrest N.
Ni, Zhaoheng
Chang, Ernie
Shi, Yangyang
Chandra, Vikas
Machine Learning
Hardware Architecture
Sound
Audio and Speech Processing
Transformer-based models excel in speech recognition. Existing efforts to optimize Transformer inference, typically for long-context applications, center on simplifying attention score calculations. However, streaming speech recognition models usually process a limited number of tokens each time, making attention score calculation less of a bottleneck. Instead, the bottleneck lies in the linear projection layers of multi-head attention and feedforward networks, constituting a substantial portion of the model size and contributing significantly to computation, memory, and power usage. To address this bottleneck, we propose folding attention, a technique targeting these linear layers, significantly reducing model size and improving memory and power efficiency. Experiments on on-device Transformer-based streaming speech recognition models show that folding attention reduces model size (and corresponding memory consumption) by up to 24% and power consumption by up to 23%, all without compromising model accuracy or computation overhead.
title Folding Attention: Memory and Power Optimization for On-Device Transformer-based Streaming Speech Recognition
topic Machine Learning
Hardware Architecture
Sound
Audio and Speech Processing
url https://arxiv.org/abs/2309.07988