@inproceedings{wang-etal-2026-activation-residual,
title = "Activation Residual Hessian Quantization for Low-Bit Large Language Model Inference",
author = "Wang, YiFeng and
Sun, Zhun and
Cho, Hakaze and
Wang, Linlin and
Sakaguchi, Keisuke",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {IJCNLP}-{AACL} 2026",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-aacl.134/",
pages = "2613--2627",
ISBN = "979-8-89176-444-6",
abstract = "Low-bit activation-weight quantization is crucial for efficient large language model inference but remains fragile because activation quantization errors propagate across Transformer layers. We propose Activation Residual Hessian Quantization (ARHQ), a post-training residual-aware low-rank correction framework that directly targets this propagation. ARHQ measures the residuals produced by the deployed activation quantizer, builds an input-side residual Hessian, and identifies the weight directions that most amplify these residuals. This yields a closed-form truncated-SVD solution: residual-sensitive components are placed in a high-precision low-rank branch, while the remaining weights stay in the low-bit path. ARHQ requires no backpropagation, supports LoRA-style deployment, and complements activation smoothing. Across LLaMA-family models and Qwen3-4B-Thinking, ARHQ improves W4A4 reconstruction, reduces WikiText2 perplexity over matched-rank SVDQuant, preserves downstream reasoning and coding accuracy, and achieves up to 3.21$\times$ prefill speedup over BF16 (3.11$\times$ over AWQ W4A16) with over 60{\%} memory saving."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2026-activation-residual">
<titleInfo>
<title>Activation Residual Hessian Quantization for Low-Bit Large Language Model Inference</title>
</titleInfo>
<name type="personal">
<namePart type="given">YiFeng</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhun</namePart>
<namePart type="family">Sun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hakaze</namePart>
<namePart type="family">Cho</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Linlin</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Keisuke</namePart>
<namePart type="family">Sakaguchi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-444-6</identifier>
</relatedItem>
<abstract>Low-bit activation-weight quantization is crucial for efficient large language model inference but remains fragile because activation quantization errors propagate across Transformer layers. We propose Activation Residual Hessian Quantization (ARHQ), a post-training residual-aware low-rank correction framework that directly targets this propagation. ARHQ measures the residuals produced by the deployed activation quantizer, builds an input-side residual Hessian, and identifies the weight directions that most amplify these residuals. This yields a closed-form truncated-SVD solution: residual-sensitive components are placed in a high-precision low-rank branch, while the remaining weights stay in the low-bit path. ARHQ requires no backpropagation, supports LoRA-style deployment, and complements activation smoothing. Across LLaMA-family models and Qwen3-4B-Thinking, ARHQ improves W4A4 reconstruction, reduces WikiText2 perplexity over matched-rank SVDQuant, preserves downstream reasoning and coding accuracy, and achieves up to 3.21\times prefill speedup over BF16 (3.11\times over AWQ W4A16) with over 60% memory saving.</abstract>
<identifier type="citekey">wang-etal-2026-activation-residual</identifier>
<location>
<url>https://aclanthology.org/2026.findings-aacl.134/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>2613</start>
<end>2627</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Activation Residual Hessian Quantization for Low-Bit Large Language Model Inference
%A Wang, YiFeng
%A Sun, Zhun
%A Cho, Hakaze
%A Wang, Linlin
%A Sakaguchi, Keisuke
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-444-6
%F wang-etal-2026-activation-residual
%X Low-bit activation-weight quantization is crucial for efficient large language model inference but remains fragile because activation quantization errors propagate across Transformer layers. We propose Activation Residual Hessian Quantization (ARHQ), a post-training residual-aware low-rank correction framework that directly targets this propagation. ARHQ measures the residuals produced by the deployed activation quantizer, builds an input-side residual Hessian, and identifies the weight directions that most amplify these residuals. This yields a closed-form truncated-SVD solution: residual-sensitive components are placed in a high-precision low-rank branch, while the remaining weights stay in the low-bit path. ARHQ requires no backpropagation, supports LoRA-style deployment, and complements activation smoothing. Across LLaMA-family models and Qwen3-4B-Thinking, ARHQ improves W4A4 reconstruction, reduces WikiText2 perplexity over matched-rank SVDQuant, preserves downstream reasoning and coding accuracy, and achieves up to 3.21\times prefill speedup over BF16 (3.11\times over AWQ W4A16) with over 60% memory saving.
%U https://aclanthology.org/2026.findings-aacl.134/
%P 2613-2627
Markdown (Informal)
[Activation Residual Hessian Quantization for Low-Bit Large Language Model Inference](https://aclanthology.org/2026.findings-aacl.134/) (Wang et al., Findings 2026)
ACL