@inproceedings{tropeano-etal-2026-improving,
title = "Improving Efficiency of Ternary Large Language Models through Pruning",
author = "Tropeano, Pietro and
Maistro, Maria and
Ruotsalo, Tuukka and
Lioma, Christina and
Aletras, Nikolaos",
editor = "Schimanski, Tobias and
Usmanova, Aida and
Morio, Gaku and
Ni, Jingwei and
Dutia, Kalyan and
Vaghefi, Seyed Saeid and
Stammbach, Dominik and
Leippold, Markus",
booktitle = "Proceedings of the 3rd Workshop on Natural Language Processing Meets Climate Change ({C}limate{NLP} 2026)",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.climatenlp-1.5/",
pages = "51--84",
ISBN = "979-8-89176-465-1",
abstract = "As the environmental impact of Large Language Models (LLMs) grows, improving model efficiency has become critical for sustainable NLP. Recent advances in model compression have shown that ternary quantisation (i.e., using only {\{}-1, 0, 1{\}} as weights) substantially reduces memory footprint, inference time, and energy usage of LLMs with minimal performance degradation. However, whether these already efficient models can be further pruned to reduce resource requirements without compromising effectiveness remains unexplored. We investigate this question across six models, four sparsity levels, and six structured and unstructured pruning techniques. We find that even minimal pruning yields substantial efficiency gains: pruning fewer than 5{\%} of parameters reduces inference time and energy consumption by up to 21{\%}, and memory usage by up to 13{\%}, with minimal accuracy degradation. We also find that, for pruned models with less than a 5{\%} accuracy drop, the most energy and inference-time efficient configurations are achieved by pruning entire attention layers, whereas the most memory efficient configurations are obtained by pruning entire MLP layers. These findings highlight pruning as a complementary technique to ternary quantisation for reducing environmental footprint of LLMs while maintaining model effectiveness."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="tropeano-etal-2026-improving">
<titleInfo>
<title>Improving Efficiency of Ternary Large Language Models through Pruning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Pietro</namePart>
<namePart type="family">Tropeano</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Maistro</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tuukka</namePart>
<namePart type="family">Ruotsalo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christina</namePart>
<namePart type="family">Lioma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nikolaos</namePart>
<namePart type="family">Aletras</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>Proceedings of the 3rd Workshop on Natural Language Processing Meets Climate Change (ClimateNLP 2026)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Tobias</namePart>
<namePart type="family">Schimanski</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aida</namePart>
<namePart type="family">Usmanova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gaku</namePart>
<namePart type="family">Morio</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jingwei</namePart>
<namePart type="family">Ni</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kalyan</namePart>
<namePart type="family">Dutia</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Seyed</namePart>
<namePart type="given">Saeid</namePart>
<namePart type="family">Vaghefi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dominik</namePart>
<namePart type="family">Stammbach</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Markus</namePart>
<namePart type="family">Leippold</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">979-8-89176-465-1</identifier>
</relatedItem>
<abstract>As the environmental impact of Large Language Models (LLMs) grows, improving model efficiency has become critical for sustainable NLP. Recent advances in model compression have shown that ternary quantisation (i.e., using only {-1, 0, 1} as weights) substantially reduces memory footprint, inference time, and energy usage of LLMs with minimal performance degradation. However, whether these already efficient models can be further pruned to reduce resource requirements without compromising effectiveness remains unexplored. We investigate this question across six models, four sparsity levels, and six structured and unstructured pruning techniques. We find that even minimal pruning yields substantial efficiency gains: pruning fewer than 5% of parameters reduces inference time and energy consumption by up to 21%, and memory usage by up to 13%, with minimal accuracy degradation. We also find that, for pruned models with less than a 5% accuracy drop, the most energy and inference-time efficient configurations are achieved by pruning entire attention layers, whereas the most memory efficient configurations are obtained by pruning entire MLP layers. These findings highlight pruning as a complementary technique to ternary quantisation for reducing environmental footprint of LLMs while maintaining model effectiveness.</abstract>
<identifier type="citekey">tropeano-etal-2026-improving</identifier>
<location>
<url>https://aclanthology.org/2026.climatenlp-1.5/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>51</start>
<end>84</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Improving Efficiency of Ternary Large Language Models through Pruning
%A Tropeano, Pietro
%A Maistro, Maria
%A Ruotsalo, Tuukka
%A Lioma, Christina
%A Aletras, Nikolaos
%Y Schimanski, Tobias
%Y Usmanova, Aida
%Y Morio, Gaku
%Y Ni, Jingwei
%Y Dutia, Kalyan
%Y Vaghefi, Seyed Saeid
%Y Stammbach, Dominik
%Y Leippold, Markus
%S Proceedings of the 3rd Workshop on Natural Language Processing Meets Climate Change (ClimateNLP 2026)
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-465-1
%F tropeano-etal-2026-improving
%X As the environmental impact of Large Language Models (LLMs) grows, improving model efficiency has become critical for sustainable NLP. Recent advances in model compression have shown that ternary quantisation (i.e., using only {-1, 0, 1} as weights) substantially reduces memory footprint, inference time, and energy usage of LLMs with minimal performance degradation. However, whether these already efficient models can be further pruned to reduce resource requirements without compromising effectiveness remains unexplored. We investigate this question across six models, four sparsity levels, and six structured and unstructured pruning techniques. We find that even minimal pruning yields substantial efficiency gains: pruning fewer than 5% of parameters reduces inference time and energy consumption by up to 21%, and memory usage by up to 13%, with minimal accuracy degradation. We also find that, for pruned models with less than a 5% accuracy drop, the most energy and inference-time efficient configurations are achieved by pruning entire attention layers, whereas the most memory efficient configurations are obtained by pruning entire MLP layers. These findings highlight pruning as a complementary technique to ternary quantisation for reducing environmental footprint of LLMs while maintaining model effectiveness.
%U https://aclanthology.org/2026.climatenlp-1.5/
%P 51-84Markdown (Informal)
[Improving Efficiency of Ternary Large Language Models through Pruning](https://aclanthology.org/2026.climatenlp-1.5/) (Tropeano et al., ClimateNLP 2026)
ACL
- Pietro Tropeano, Maria Maistro, Tuukka Ruotsalo, Christina Lioma, and Nikolaos Aletras. 2026. Improving Efficiency of Ternary Large Language Models through Pruning. In Proceedings of the 3rd Workshop on Natural Language Processing Meets Climate Change (ClimateNLP 2026), pages 51–84, Budapest, Hungary. Association for Computational Linguistics.