@inproceedings{kim-etal-2026-pivot-gradient,
title = "Pivot Gradient Scaling: Reducing Dwell Time to Stabilize Extreme Low-Bit Quantization-aware Training",
author = "Kim, Vinnam and
Kim, Junhan and
Shin, Seungjae and
Lee, Jung Hyun and
You, Jaeseong and
Chen, An and
Choi, Jungwook",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1947/",
pages = "41511--41531",
ISBN = "979-8-89176-462-0",
abstract = "Extreme low-bit quantization-aware training (QAT) is often unstable due to synchronized bin transitions, which can trigger catastrophic loss spikes. While prior works rely on scale-side controls to mitigate this fragility, we demonstrate that restricting pivot movement alone is insufficient. Instead, we identify a critical weight-side state governing low-bit QAT stability: pivot-proximal residence, where latent weights persistently linger near quantization pivots. We formalize this duration as dwell time and show that long dwell times accumulate a large, vulnerable population of weights prone to synchronized mass flips. To resolve this vulnerability, we propose Pivot Gradient Scaling (PGS) as a principled modification to the Straight-Through Estimator (STE) that leaves the forward pass intact and selectively amplifies latent-weight gradients within a localized pivot band. Crucially, we provide a mathematical guarantee proving that this sign-preserving amplification increases pivot-proximal mobility and systematically shortens expected dwell time under both SGD and Adam. Exhaustive empirical evaluations across Llama 3.2 and Qwen3 model families under full-training regimes up to 30 billion tokens demonstrate that PGS structurally stabilizes optimization, attenuates transition bursts, and delivers substantial downstream accuracy gains."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kim-etal-2026-pivot-gradient">
<titleInfo>
<title>Pivot Gradient Scaling: Reducing Dwell Time to Stabilize Extreme Low-Bit Quantization-aware Training</title>
</titleInfo>
<name type="personal">
<namePart type="given">Vinnam</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Junhan</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Seungjae</namePart>
<namePart type="family">Shin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jung</namePart>
<namePart type="given">Hyun</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jaeseong</namePart>
<namePart type="family">You</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">An</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jungwook</namePart>
<namePart type="family">Choi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Extreme low-bit quantization-aware training (QAT) is often unstable due to synchronized bin transitions, which can trigger catastrophic loss spikes. While prior works rely on scale-side controls to mitigate this fragility, we demonstrate that restricting pivot movement alone is insufficient. Instead, we identify a critical weight-side state governing low-bit QAT stability: pivot-proximal residence, where latent weights persistently linger near quantization pivots. We formalize this duration as dwell time and show that long dwell times accumulate a large, vulnerable population of weights prone to synchronized mass flips. To resolve this vulnerability, we propose Pivot Gradient Scaling (PGS) as a principled modification to the Straight-Through Estimator (STE) that leaves the forward pass intact and selectively amplifies latent-weight gradients within a localized pivot band. Crucially, we provide a mathematical guarantee proving that this sign-preserving amplification increases pivot-proximal mobility and systematically shortens expected dwell time under both SGD and Adam. Exhaustive empirical evaluations across Llama 3.2 and Qwen3 model families under full-training regimes up to 30 billion tokens demonstrate that PGS structurally stabilizes optimization, attenuates transition bursts, and delivers substantial downstream accuracy gains.</abstract>
<identifier type="citekey">kim-etal-2026-pivot-gradient</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1947/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>41511</start>
<end>41531</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Pivot Gradient Scaling: Reducing Dwell Time to Stabilize Extreme Low-Bit Quantization-aware Training
%A Kim, Vinnam
%A Kim, Junhan
%A Shin, Seungjae
%A Lee, Jung Hyun
%A You, Jaeseong
%A Chen, An
%A Choi, Jungwook
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F kim-etal-2026-pivot-gradient
%X Extreme low-bit quantization-aware training (QAT) is often unstable due to synchronized bin transitions, which can trigger catastrophic loss spikes. While prior works rely on scale-side controls to mitigate this fragility, we demonstrate that restricting pivot movement alone is insufficient. Instead, we identify a critical weight-side state governing low-bit QAT stability: pivot-proximal residence, where latent weights persistently linger near quantization pivots. We formalize this duration as dwell time and show that long dwell times accumulate a large, vulnerable population of weights prone to synchronized mass flips. To resolve this vulnerability, we propose Pivot Gradient Scaling (PGS) as a principled modification to the Straight-Through Estimator (STE) that leaves the forward pass intact and selectively amplifies latent-weight gradients within a localized pivot band. Crucially, we provide a mathematical guarantee proving that this sign-preserving amplification increases pivot-proximal mobility and systematically shortens expected dwell time under both SGD and Adam. Exhaustive empirical evaluations across Llama 3.2 and Qwen3 model families under full-training regimes up to 30 billion tokens demonstrate that PGS structurally stabilizes optimization, attenuates transition bursts, and delivers substantial downstream accuracy gains.
%U https://aclanthology.org/2026.findings-emnlp.1947/
%P 41511-41531
Markdown (Informal)
[Pivot Gradient Scaling: Reducing Dwell Time to Stabilize Extreme Low-Bit Quantization-aware Training](https://aclanthology.org/2026.findings-emnlp.1947/) (Kim et al., Findings 2026)
ACL