大模型
arxiv.org ·
从 Token 重要性到条件可移除性：重新审视多模态大语言模型中的视觉 Token 剪枝
该论文指出，免训练视觉 token 剪枝常用的 token 重要性、冗余度等信号不足以刻画“可移除性”，后者同时取决于表示深度与当前删除集合。受控干预实验显示，同一 token 在不同深度删除会产生显著不同的下游扰动，固定深度下改变删除上下文也会改变候选边际与剪枝边界决策。为此作者提出免训练两阶段框架 CoRePrune：渐进式扰动感知视觉剪枝随视觉表示演化刷新删除效应，集合条件精炼在视觉-文本交互后基于当前删除集重新评估候选保留收益。在覆盖标准图像、高分辨率输入与视频的五种多模态大模型骨干上，CoRePrune 在激进 token 预算下保持性能；在 Qwen3.5 上最终预算 128 个视觉 token 时保留稠密模型 90.3% 的性能，并将总体 prefill 时间减少 51.0%。
AI推荐理由 arXiv 新论文，提出免训练视觉 token 剪枝框架 CoRePrune，指出 token 重要性不足以刻画可移除性，需结合表示深度与删除集合上下文。摘要给出具体机制（Progressive Perturbation-Aware Visual Pruning、Set-Conditioned Refinement）与量化结果：在 Qwen3.5 上 128 视觉 token 预算下保留 90.3% 稠密模型性能、prefill 时间减少 51.0%，并覆盖五种 MLLM 骨干与图像/高分辨率/视频场景。信息密度高、方法可验证，对多模态推理效率有实际价值；局限是仅基于摘要，未提供代码或第三方复现，性能数字为作者自报。
78
正文
Computer Science > Computer Vision and Pattern Recognition
arXiv:2609.26484 (cs)
[Submitted on 22 Sep 2026]
Title:From Token Importance to Conditional Removability: Rethinking Visual Token Pruning in Multimodal Large Language Models
Authors:Shengli He, Yongchao Liang, Roumeng He, Junjie Zeng, Jiyuan He, Xin Fang, Can Wu, Li Zheng
View a PDF of the paper titled From Token Importance to Conditional Removability: Rethinking Visual Token Pruning in Multimodal Large Language Models, by Shengli He and 7 other authors
View PDF
HTML (experimental)
Abstract:Training-free visual-token pruning often uses token importance, redundancy, or related selection criteria as proxies for safe removal. We show that these signals alone do not fully characterize removability, which is conditioned on both representation depth and the surrounding deletion set. Controlled interventions demonstrate that removing the same tokens at different depths produces substantially different downstream perturbations, while changing only the deletion context at a fixed depth alters candidate marginals and pruning-boundary decisions. These findings show that token importance alone cannot determine when a token is safely removable or how its removability changes under joint deletion. Motivated by this perspective, we propose CoRePrune, a training-free two-stage framework. Progressive Perturbation-Aware Visual Pruning refreshes deletion effects as visual representations evolve, while Set-Conditioned Refinement reevaluates candidate rescue benefits under the current deletion set after visual--text interaction. Across five multimodal large language model backbones covering standard images, high-resolution inputs, and video, CoRePrune preserves performance under aggressive token budgets. On Qwen3.5, with a final budget of 128 visual tokens, it retains 90.3% of dense-model performance while reducing aggregate prefill time by 51.0%.
Subjects:
Computer Vision and Pattern Recognition (cs.CV)
Cite as:
arXiv:2609.26484 [cs.CV]
(or
arXiv:2609.26484v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2609.26484
Focus to learn more
arXiv-issued DOI via DataCite (pending registration)
Submission history
From: ShengLi He [view email]
[v1]
Tue, 22 Sep 2026 14:22:42 UTC (660 KB)
Full-text links:
Access Paper:
View a PDF of the paper titled From Token Importance to Conditional Removability: Rethinking Visual Token Pruning in Multimodal Large Language Models, by Shengli He and 7 other authors
View PDF
HTML (experimental)
TeX Source
view license
Current browse context:
cs.CV
new
|
recent
| 2026-09
Change to browse by:
cs
References & Citations
NASA ADS
Google Scholar
Semantic Scholar
export BibTeX citation
Loading...
BibTeX formatted citation
×
loading...
Data provided by:
Bookmark
Bibliographic Tools
Bibliographic and Citation Tools
Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media
Code, Data and Media Associated with this Article
alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
Demos
Demos
Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related Papers
Recommenders and Search Tools
Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
Author
Venue
Institution
Topic
About arXivLabs
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Which authors of this paper are endorsers? |
Disable MathJax (What is MathJax?)