@inproceedings{lou-etal-2026-geoprune,
title = "{G}eo{P}rune: A Robust and Engine-Friendly Importance Signal Beyond Attention for Vision Token Pruning in {VLM}s",
author = "Lou, Kang and
Zhang, Shijin and
Chen, Ming and
Cao, Junzhe",
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.1913/",
pages = "40759--40776",
ISBN = "979-8-89176-462-0",
abstract = "Vision token pruning can accelerate vision language model inference by removing redundant visual tokens. However, we observe that ViT attention based pruning drops sharply on Qwen3-VL and Qwen3.5. Empirically, we identify the root cause: the last-block MLP of the vision encoder significantly alters the spectral structure of hidden states and induces spectral collapse. Thus, pre-MLP attention rankings no longer reflect downstream token importance. Through further analysis of vision token selection when ViT attention based pruning still succeeds, we find that selected tokens strongly anti-correlate with raw L2 norm in the vision encoder output space $S_{out}$ , while favoring tokens with larger residual energy. Motivated by shared components mask token-specific importance, we propose GeoPrune, a new importance-based pruning method that operates in $S_{out}$ , applies deflation to remove dominant components, and ranks tokens by residual L2 norm. At aggressive pruning rates, GeoPrune is more robust and surpasses ViT attention based pruning by +4.4/+6.3 recovery points on Qwen3-VL and Qwen3.5. We further integrate GeoPrune into the production-grade inference engine vLLM, delivering 5.4{\%}-10.5{\%} mean E2E latency reductions with only slight accuracy drops, advancing vision token pruning toward practical deployment."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="lou-etal-2026-geoprune">
<titleInfo>
<title>GeoPrune: A Robust and Engine-Friendly Importance Signal Beyond Attention for Vision Token Pruning in VLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Kang</namePart>
<namePart type="family">Lou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shijin</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ming</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Junzhe</namePart>
<namePart type="family">Cao</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>Vision token pruning can accelerate vision language model inference by removing redundant visual tokens. However, we observe that ViT attention based pruning drops sharply on Qwen3-VL and Qwen3.5. Empirically, we identify the root cause: the last-block MLP of the vision encoder significantly alters the spectral structure of hidden states and induces spectral collapse. Thus, pre-MLP attention rankings no longer reflect downstream token importance. Through further analysis of vision token selection when ViT attention based pruning still succeeds, we find that selected tokens strongly anti-correlate with raw L2 norm in the vision encoder output space S_out , while favoring tokens with larger residual energy. Motivated by shared components mask token-specific importance, we propose GeoPrune, a new importance-based pruning method that operates in S_out , applies deflation to remove dominant components, and ranks tokens by residual L2 norm. At aggressive pruning rates, GeoPrune is more robust and surpasses ViT attention based pruning by +4.4/+6.3 recovery points on Qwen3-VL and Qwen3.5. We further integrate GeoPrune into the production-grade inference engine vLLM, delivering 5.4%-10.5% mean E2E latency reductions with only slight accuracy drops, advancing vision token pruning toward practical deployment.</abstract>
<identifier type="citekey">lou-etal-2026-geoprune</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1913/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>40759</start>
<end>40776</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T GeoPrune: A Robust and Engine-Friendly Importance Signal Beyond Attention for Vision Token Pruning in VLMs
%A Lou, Kang
%A Zhang, Shijin
%A Chen, Ming
%A Cao, Junzhe
%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 lou-etal-2026-geoprune
%X Vision token pruning can accelerate vision language model inference by removing redundant visual tokens. However, we observe that ViT attention based pruning drops sharply on Qwen3-VL and Qwen3.5. Empirically, we identify the root cause: the last-block MLP of the vision encoder significantly alters the spectral structure of hidden states and induces spectral collapse. Thus, pre-MLP attention rankings no longer reflect downstream token importance. Through further analysis of vision token selection when ViT attention based pruning still succeeds, we find that selected tokens strongly anti-correlate with raw L2 norm in the vision encoder output space S_out , while favoring tokens with larger residual energy. Motivated by shared components mask token-specific importance, we propose GeoPrune, a new importance-based pruning method that operates in S_out , applies deflation to remove dominant components, and ranks tokens by residual L2 norm. At aggressive pruning rates, GeoPrune is more robust and surpasses ViT attention based pruning by +4.4/+6.3 recovery points on Qwen3-VL and Qwen3.5. We further integrate GeoPrune into the production-grade inference engine vLLM, delivering 5.4%-10.5% mean E2E latency reductions with only slight accuracy drops, advancing vision token pruning toward practical deployment.
%U https://aclanthology.org/2026.findings-emnlp.1913/
%P 40759-40776
Markdown (Informal)
[GeoPrune: A Robust and Engine-Friendly Importance Signal Beyond Attention for Vision Token Pruning in VLMs](https://aclanthology.org/2026.findings-emnlp.1913/) (Lou et al., Findings 2026)
ACL