Saved in:
Bibliographic Details
Main Authors: Zhang, Cheng, Cheng, Jianyi, Constantinides, George A., Zhao, Yiren
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2402.02446
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910463660392448
author Zhang, Cheng
Cheng, Jianyi
Constantinides, George A.
Zhao, Yiren
author_facet Zhang, Cheng
Cheng, Jianyi
Constantinides, George A.
Zhao, Yiren
contents Post-training quantization of Large Language Models (LLMs) is challenging. In this work, we introduce Low-rank Quantization Error Reduction (LQER), which combines quantization and low-rank approximation to recover the model capability. LQER leverages an activation-induced scale matrix to drive the singular value distribution of quantization error towards a desirable distribution, which enables nearly-lossless W4A8 quantization on various LLMs and downstream tasks without the need for knowledge distillation, grid search, or gradient-base iterative optimization. Unlike existing methods, the computation pattern of LQER eliminates the need for specialized Scatter and Gather processes to collect high-precision weights from irregular memory locations. Our W4A8 LLMs achieve near-lossless performance on six popular downstream tasks, while using 1.36$\times$ fewer hardware resources than the leading state-of-the-art method. We open-source our framework at https://github.com/ChengZhang-98/lqer
format Preprint
id arxiv_https___arxiv_org_abs_2402_02446
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle LQER: Low-Rank Quantization Error Reconstruction for LLMs
Zhang, Cheng
Cheng, Jianyi
Constantinides, George A.
Zhao, Yiren
Machine Learning
Computation and Language
Post-training quantization of Large Language Models (LLMs) is challenging. In this work, we introduce Low-rank Quantization Error Reduction (LQER), which combines quantization and low-rank approximation to recover the model capability. LQER leverages an activation-induced scale matrix to drive the singular value distribution of quantization error towards a desirable distribution, which enables nearly-lossless W4A8 quantization on various LLMs and downstream tasks without the need for knowledge distillation, grid search, or gradient-base iterative optimization. Unlike existing methods, the computation pattern of LQER eliminates the need for specialized Scatter and Gather processes to collect high-precision weights from irregular memory locations. Our W4A8 LLMs achieve near-lossless performance on six popular downstream tasks, while using 1.36$\times$ fewer hardware resources than the leading state-of-the-art method. We open-source our framework at https://github.com/ChengZhang-98/lqer
title LQER: Low-Rank Quantization Error Reconstruction for LLMs
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2402.02446