Preserve-Then-Quantize: Balancing Rank Budgets for Quantization Error Reconstruction in LLMs

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Cho, Yoonjun, Jeon, Dongjae, Kim, Soeun, Jeon, Moongyu, No, Albert
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909038800797696
author Cho, Yoonjun
Jeon, Dongjae
Kim, Soeun
Jeon, Moongyu
No, Albert
author_facet Cho, Yoonjun
Jeon, Dongjae
Kim, Soeun
Jeon, Moongyu
No, Albert
contents Quantization Error Reconstruction (QER) reduces accuracy loss in Post-Training Quantization (PTQ) by approximating weights as $\mathbf{W} \approx \mathbf{Q} + \mathbf{L}\mathbf{R}$, using a rank-$r$ correction to reconstruct quantization error. Prior methods devote the full rank budget to error reconstruction, which is suboptimal when $\mathbf{W}$ has intrinsic low-rank structure and quantization corrupts dominant directions. We propose Structured Residual Reconstruction (SRR), a rank-allocation framework that preserves the top-$k$ singular subspace of the activation-scaled weight before quantization, quantizes only the residual, and uses the remaining rank $r-k$ for error reconstruction. We derive a theory-guided criterion for selecting $k$ by balancing quantization-exposed energy and unrecoverable error under rank constraints. We further show that resulting $\mathbf{Q} + \mathbf{L}\mathbf{R}$ parameterization naturally supports Quantized Parameter-Efficient Fine-Tuning (QPEFT), and stabilizes fine-tuning via gradient scaling along preserved directions. Experiments demonstrate consistent perplexity reductions across diverse models and quantization settings in PTQ, along with a 5.9 percentage-point average gain on GLUE under 2-bit QPEFT. The project page is available at https://ai-isl.github.io/srr.
format Preprint
id arxiv_https___arxiv_org_abs_2602_02001
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Preserve-Then-Quantize: Balancing Rank Budgets for Quantization Error Reconstruction in LLMs
Cho, Yoonjun
Jeon, Dongjae
Kim, Soeun
Jeon, Moongyu
No, Albert
Machine Learning
Artificial Intelligence
Quantization Error Reconstruction (QER) reduces accuracy loss in Post-Training Quantization (PTQ) by approximating weights as $\mathbf{W} \approx \mathbf{Q} + \mathbf{L}\mathbf{R}$, using a rank-$r$ correction to reconstruct quantization error. Prior methods devote the full rank budget to error reconstruction, which is suboptimal when $\mathbf{W}$ has intrinsic low-rank structure and quantization corrupts dominant directions. We propose Structured Residual Reconstruction (SRR), a rank-allocation framework that preserves the top-$k$ singular subspace of the activation-scaled weight before quantization, quantizes only the residual, and uses the remaining rank $r-k$ for error reconstruction. We derive a theory-guided criterion for selecting $k$ by balancing quantization-exposed energy and unrecoverable error under rank constraints. We further show that resulting $\mathbf{Q} + \mathbf{L}\mathbf{R}$ parameterization naturally supports Quantized Parameter-Efficient Fine-Tuning (QPEFT), and stabilizes fine-tuning via gradient scaling along preserved directions. Experiments demonstrate consistent perplexity reductions across diverse models and quantization settings in PTQ, along with a 5.9 percentage-point average gain on GLUE under 2-bit QPEFT. The project page is available at https://ai-isl.github.io/srr.
title Preserve-Then-Quantize: Balancing Rank Budgets for Quantization Error Reconstruction in LLMs
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2602.02001