@inproceedings{dbouk-etal-2026-techniques,
title = "Techniques for Peak Memory Reduction for {L}o{RA} Fine-tuning of {LLM}s on Edge Devices",
author = "Dbouk, Hassan and
Reisser, Matthias and
Mandke, Prathamesh and
Navali, Likhita Arun and
Louizos, Christos",
editor = "Atanasova, Pepa and
Haffari, Gholamreza and
Wang, Minghan",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-industry.88/",
pages = "1466--1500",
ISBN = "979-8-89176-460-6",
abstract = "Fine-tuning of Large Language Models (LLMs) using Low-Rank Adaptation (LoRA) on an end-user{'}s data offers personalized experiences while keeping data private, but faces severe memory constraints on consumer hardware. Peak memory during fine-tuning often exceeds device limits, especially for models with billions of parameters and long-context training data. This paper introduces a suite of complementary techniques to reduce memory footprint: (1) base model quantization with on-the-fly dequantization, (2) memory-efficient checkpointing combining selective activation caching and disk offloading, (3) softmax approximation using semantically relevant token subsets, and (4) logits masking. Only softmax approximation alters the training objective and we discuss the resulting quality trade-off both analytically and empirically. Experiments on Llama-3.2 3B and Qwen-2.5 3B demonstrate up to $26\times$ and $28\times$ reduction in peak memory, enabling fine-tuning on resource-constrained devices."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="dbouk-etal-2026-techniques">
<titleInfo>
<title>Techniques for Peak Memory Reduction for LoRA Fine-tuning of LLMs on Edge Devices</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hassan</namePart>
<namePart type="family">Dbouk</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Matthias</namePart>
<namePart type="family">Reisser</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Prathamesh</namePart>
<namePart type="family">Mandke</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Likhita</namePart>
<namePart type="given">Arun</namePart>
<namePart type="family">Navali</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christos</namePart>
<namePart type="family">Louizos</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 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track</title>
</titleInfo>
<name type="personal">
<namePart type="given">Pepa</namePart>
<namePart type="family">Atanasova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gholamreza</namePart>
<namePart type="family">Haffari</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Minghan</namePart>
<namePart type="family">Wang</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">979-8-89176-460-6</identifier>
</relatedItem>
<abstract>Fine-tuning of Large Language Models (LLMs) using Low-Rank Adaptation (LoRA) on an end-user’s data offers personalized experiences while keeping data private, but faces severe memory constraints on consumer hardware. Peak memory during fine-tuning often exceeds device limits, especially for models with billions of parameters and long-context training data. This paper introduces a suite of complementary techniques to reduce memory footprint: (1) base model quantization with on-the-fly dequantization, (2) memory-efficient checkpointing combining selective activation caching and disk offloading, (3) softmax approximation using semantically relevant token subsets, and (4) logits masking. Only softmax approximation alters the training objective and we discuss the resulting quality trade-off both analytically and empirically. Experiments on Llama-3.2 3B and Qwen-2.5 3B demonstrate up to 26\times and 28\times reduction in peak memory, enabling fine-tuning on resource-constrained devices.</abstract>
<identifier type="citekey">dbouk-etal-2026-techniques</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-industry.88/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>1466</start>
<end>1500</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Techniques for Peak Memory Reduction for LoRA Fine-tuning of LLMs on Edge Devices
%A Dbouk, Hassan
%A Reisser, Matthias
%A Mandke, Prathamesh
%A Navali, Likhita Arun
%A Louizos, Christos
%Y Atanasova, Pepa
%Y Haffari, Gholamreza
%Y Wang, Minghan
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-460-6
%F dbouk-etal-2026-techniques
%X Fine-tuning of Large Language Models (LLMs) using Low-Rank Adaptation (LoRA) on an end-user’s data offers personalized experiences while keeping data private, but faces severe memory constraints on consumer hardware. Peak memory during fine-tuning often exceeds device limits, especially for models with billions of parameters and long-context training data. This paper introduces a suite of complementary techniques to reduce memory footprint: (1) base model quantization with on-the-fly dequantization, (2) memory-efficient checkpointing combining selective activation caching and disk offloading, (3) softmax approximation using semantically relevant token subsets, and (4) logits masking. Only softmax approximation alters the training objective and we discuss the resulting quality trade-off both analytically and empirically. Experiments on Llama-3.2 3B and Qwen-2.5 3B demonstrate up to 26\times and 28\times reduction in peak memory, enabling fine-tuning on resource-constrained devices.
%U https://aclanthology.org/2026.emnlp-industry.88/
%P 1466-1500
Markdown (Informal)
[Techniques for Peak Memory Reduction for LoRA Fine-tuning of LLMs on Edge Devices](https://aclanthology.org/2026.emnlp-industry.88/) (Dbouk et al., EMNLP 2026)
ACL