MLoRQ: Bridging Low-Rank and Quantization for Transformer Compression

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Gordon, Ofir, Lapid, Ariel, Cohen, Elad, Yagil, Yarden, Netzer, Arnon, Habi, Hai Victor
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866909687252779008
author Gordon, Ofir
Lapid, Ariel
Cohen, Elad
Yagil, Yarden
Netzer, Arnon
Habi, Hai Victor
author_facet Gordon, Ofir
Lapid, Ariel
Cohen, Elad
Yagil, Yarden
Netzer, Arnon
Habi, Hai Victor
contents Deploying transformer-based neural networks on resource-constrained edge devices presents a significant challenge. This challenge is often addressed through various techniques, such as low-rank approximation and mixed-precision quantization. In this work, we introduce Mixed Low-Rank and Quantization (MLoRQ), a novel method that integrates both techniques. MLoRQ employs a two-stage optimization process to determine optimal bit-width and rank assignments for each layer, adhering to predefined memory constraints. This process includes: (i) an intra-layer optimization that identifies potentially optimal compression solutions out of all low-rank and quantization combinations; (ii) an inter-layer optimization that assigns bit-width precision and rank to each layer while ensuring the memory constraint is met. An optional final step applies a sequential optimization process using a modified adaptive rounding technique to mitigate compression-induced errors in joint low-rank approximation and quantization. The method is compatible and can be seamlessly integrated with most existing quantization algorithms. MLoRQ shows state-of-the-art results with up to 15\% performance improvement, evaluated on Vision Transformers for image classification, object detection, and instance segmentation tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2507_09616
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MLoRQ: Bridging Low-Rank and Quantization for Transformer Compression
Gordon, Ofir
Lapid, Ariel
Cohen, Elad
Yagil, Yarden
Netzer, Arnon
Habi, Hai Victor
Machine Learning
Computer Vision and Pattern Recognition
Deploying transformer-based neural networks on resource-constrained edge devices presents a significant challenge. This challenge is often addressed through various techniques, such as low-rank approximation and mixed-precision quantization. In this work, we introduce Mixed Low-Rank and Quantization (MLoRQ), a novel method that integrates both techniques. MLoRQ employs a two-stage optimization process to determine optimal bit-width and rank assignments for each layer, adhering to predefined memory constraints. This process includes: (i) an intra-layer optimization that identifies potentially optimal compression solutions out of all low-rank and quantization combinations; (ii) an inter-layer optimization that assigns bit-width precision and rank to each layer while ensuring the memory constraint is met. An optional final step applies a sequential optimization process using a modified adaptive rounding technique to mitigate compression-induced errors in joint low-rank approximation and quantization. The method is compatible and can be seamlessly integrated with most existing quantization algorithms. MLoRQ shows state-of-the-art results with up to 15\% performance improvement, evaluated on Vision Transformers for image classification, object detection, and instance segmentation tasks.
title MLoRQ: Bridging Low-Rank and Quantization for Transformer Compression
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2507.09616