@inproceedings{wen-etal-2026-prism,
title = "{PRISM}: Post-Training {LLM} Pruning via Residual Importance Signal Merging and Intra-Block Sparsity Allocation",
author = "Wen, Keyu and
Zhu, Jianing and
HU, Kaiyang and
Yang, Lijun and
Lin, Rui",
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.606/",
pages = "12779--12795",
ISBN = "979-8-89176-462-0",
abstract = "Post-training pruning compresses Large Language Models (LLMs) without full-model retraining, but performance degrades sharply at high sparsity. We identify two limitations in existing methods: local importance estimates miss global spectral structure, and uniform within-block sparsity ignores heterogeneous sensitivity across transformer projections. We propose PRISM, a post-training pruning framework with training-free mask construction and optional fixed-mask reconstruction optimization. PRISM augments activation-based scoring with two adaptive-rank SVD-derived signals for principal-subspace and residual-aware importance, combines the factors through parameter-free CV-based fusion, and allocates sparsity non-uniformly across attention and MLP projections using a gradient-weighted matrix-level sensitivity proxy. Across six Qwen3 and LLaMA models, PRISM consistently outperforms existing pruning baselines, yielding up to 16.8{\%} relative improvement in WikiText-2 perplexity under the evaluated sparsity settings."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wen-etal-2026-prism">
<titleInfo>
<title>PRISM: Post-Training LLM Pruning via Residual Importance Signal Merging and Intra-Block Sparsity Allocation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Keyu</namePart>
<namePart type="family">Wen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jianing</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kaiyang</namePart>
<namePart type="family">HU</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lijun</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rui</namePart>
<namePart type="family">Lin</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>Post-training pruning compresses Large Language Models (LLMs) without full-model retraining, but performance degrades sharply at high sparsity. We identify two limitations in existing methods: local importance estimates miss global spectral structure, and uniform within-block sparsity ignores heterogeneous sensitivity across transformer projections. We propose PRISM, a post-training pruning framework with training-free mask construction and optional fixed-mask reconstruction optimization. PRISM augments activation-based scoring with two adaptive-rank SVD-derived signals for principal-subspace and residual-aware importance, combines the factors through parameter-free CV-based fusion, and allocates sparsity non-uniformly across attention and MLP projections using a gradient-weighted matrix-level sensitivity proxy. Across six Qwen3 and LLaMA models, PRISM consistently outperforms existing pruning baselines, yielding up to 16.8% relative improvement in WikiText-2 perplexity under the evaluated sparsity settings.</abstract>
<identifier type="citekey">wen-etal-2026-prism</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.606/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>12779</start>
<end>12795</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T PRISM: Post-Training LLM Pruning via Residual Importance Signal Merging and Intra-Block Sparsity Allocation
%A Wen, Keyu
%A Zhu, Jianing
%A HU, Kaiyang
%A Yang, Lijun
%A Lin, Rui
%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 wen-etal-2026-prism
%X Post-training pruning compresses Large Language Models (LLMs) without full-model retraining, but performance degrades sharply at high sparsity. We identify two limitations in existing methods: local importance estimates miss global spectral structure, and uniform within-block sparsity ignores heterogeneous sensitivity across transformer projections. We propose PRISM, a post-training pruning framework with training-free mask construction and optional fixed-mask reconstruction optimization. PRISM augments activation-based scoring with two adaptive-rank SVD-derived signals for principal-subspace and residual-aware importance, combines the factors through parameter-free CV-based fusion, and allocates sparsity non-uniformly across attention and MLP projections using a gradient-weighted matrix-level sensitivity proxy. Across six Qwen3 and LLaMA models, PRISM consistently outperforms existing pruning baselines, yielding up to 16.8% relative improvement in WikiText-2 perplexity under the evaluated sparsity settings.
%U https://aclanthology.org/2026.findings-emnlp.606/
%P 12779-12795
Markdown (Informal)
[PRISM: Post-Training LLM Pruning via Residual Importance Signal Merging and Intra-Block Sparsity Allocation](https://aclanthology.org/2026.findings-emnlp.606/) (Wen et al., Findings 2026)
ACL