LRAMM -- Low precision approximates GEMM via RSVD

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autore principale: Gu, Hongyaoxing
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866929359817801728
author Gu, Hongyaoxing
author_facet Gu, Hongyaoxing
contents Matrix multiplication computation acceleration has been a research hotspot across various domains. Due to the characteristics of some applications, approximate matrix multiplication can achieve significant performance improvements without losing much precision. In this paper, we propose LRAMM - a high-performance matrix multiplication approximation algorithm that combines mixed-precision quantized matrix multiplication with RSVD techniques, further enhancing efficiency within the error range of low-precision matrix multiplication by utilizing matrix low-rank decomposition technology.
format Preprint
id arxiv_https___arxiv_org_abs_2405_16917
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle LRAMM -- Low precision approximates GEMM via RSVD
Gu, Hongyaoxing
Numerical Analysis
Performance
Matrix multiplication computation acceleration has been a research hotspot across various domains. Due to the characteristics of some applications, approximate matrix multiplication can achieve significant performance improvements without losing much precision. In this paper, we propose LRAMM - a high-performance matrix multiplication approximation algorithm that combines mixed-precision quantized matrix multiplication with RSVD techniques, further enhancing efficiency within the error range of low-precision matrix multiplication by utilizing matrix low-rank decomposition technology.
title LRAMM -- Low precision approximates GEMM via RSVD
topic Numerical Analysis
Performance
url https://arxiv.org/abs/2405.16917