@inproceedings{li-etal-2025-mobilora,
title = "{M}obi{L}o{RA}: Accelerating {L}o{RA}-based {LLM} Inference on Mobile Devices via Context-aware {KV} Cache Optimization",
author = "Li, Borui and
Wang, Yitao and
Ma, Haoran and
Chen, Ligeng and
Xiao, Jun and
Wang, Shuai",
editor = "Che, Wanxiang and
Nabende, Joyce and
Shutova, Ekaterina and
Pilehvar, Mohammad Taher",
booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.acl-long.1140/",
doi = "10.18653/v1/2025.acl-long.1140",
pages = "23400--23410",
ISBN = "979-8-89176-251-0",
abstract = "Deploying large language models (LLMs) with low-rank adaptation (LoRA) on mobile devices is promising due to their capability to complete diverse domain-specific tasks while ensuring privacy and accessibility. In this paper, we introduce MobiLoRA to accelerate LoRA-based LLM inference on mobile devices. MobiLoRA focuses on optimizing the key-value (KV) caches due to the limited computing and memory resources of mobile devices. The key insight of MobiLoRA lies in the utilization of two contexts for on-device LoRA serving: semantic-level contexts, such as prompts with shared prefixes, and system-level contexts, such as the application status (e.g., foreground or killed) of LLM requests. Specifically, for semantic-level contexts, MobiLoRA proposes similarity-aware delta encoding, which leverages token-wise similarity in KV caches across LoRA adapters for efficient storage and reuse. Furthermore, MobiLoRA advocates context-aware KV cache management to optimize cache retention and eviction considering the system-level contexts. We fully implement MobiLoRA and compare it with state-of-the-art LLM serving frameworks using real-world mobile device traces. Results show that MobiLoRA accelerates LoRA-based LLM inference by 57.6{\%} on mobile devices."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="li-etal-2025-mobilora">
<titleInfo>
<title>MobiLoRA: Accelerating LoRA-based LLM Inference on Mobile Devices via Context-aware KV Cache Optimization</title>
</titleInfo>
<name type="personal">
<namePart type="given">Borui</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yitao</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haoran</namePart>
<namePart type="family">Ma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ligeng</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Xiao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shuai</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wanxiang</namePart>
<namePart type="family">Che</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joyce</namePart>
<namePart type="family">Nabende</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Shutova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohammad</namePart>
<namePart type="given">Taher</namePart>
<namePart type="family">Pilehvar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Vienna, Austria</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-251-0</identifier>
</relatedItem>
<abstract>Deploying large language models (LLMs) with low-rank adaptation (LoRA) on mobile devices is promising due to their capability to complete diverse domain-specific tasks while ensuring privacy and accessibility. In this paper, we introduce MobiLoRA to accelerate LoRA-based LLM inference on mobile devices. MobiLoRA focuses on optimizing the key-value (KV) caches due to the limited computing and memory resources of mobile devices. The key insight of MobiLoRA lies in the utilization of two contexts for on-device LoRA serving: semantic-level contexts, such as prompts with shared prefixes, and system-level contexts, such as the application status (e.g., foreground or killed) of LLM requests. Specifically, for semantic-level contexts, MobiLoRA proposes similarity-aware delta encoding, which leverages token-wise similarity in KV caches across LoRA adapters for efficient storage and reuse. Furthermore, MobiLoRA advocates context-aware KV cache management to optimize cache retention and eviction considering the system-level contexts. We fully implement MobiLoRA and compare it with state-of-the-art LLM serving frameworks using real-world mobile device traces. Results show that MobiLoRA accelerates LoRA-based LLM inference by 57.6% on mobile devices.</abstract>
<identifier type="citekey">li-etal-2025-mobilora</identifier>
<identifier type="doi">10.18653/v1/2025.acl-long.1140</identifier>
<location>
<url>https://aclanthology.org/2025.acl-long.1140/</url>
</location>
<part>
<date>2025-07</date>
<extent unit="page">
<start>23400</start>
<end>23410</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T MobiLoRA: Accelerating LoRA-based LLM Inference on Mobile Devices via Context-aware KV Cache Optimization
%A Li, Borui
%A Wang, Yitao
%A Ma, Haoran
%A Chen, Ligeng
%A Xiao, Jun
%A Wang, Shuai
%Y Che, Wanxiang
%Y Nabende, Joyce
%Y Shutova, Ekaterina
%Y Pilehvar, Mohammad Taher
%S Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2025
%8 July
%I Association for Computational Linguistics
%C Vienna, Austria
%@ 979-8-89176-251-0
%F li-etal-2025-mobilora
%X Deploying large language models (LLMs) with low-rank adaptation (LoRA) on mobile devices is promising due to their capability to complete diverse domain-specific tasks while ensuring privacy and accessibility. In this paper, we introduce MobiLoRA to accelerate LoRA-based LLM inference on mobile devices. MobiLoRA focuses on optimizing the key-value (KV) caches due to the limited computing and memory resources of mobile devices. The key insight of MobiLoRA lies in the utilization of two contexts for on-device LoRA serving: semantic-level contexts, such as prompts with shared prefixes, and system-level contexts, such as the application status (e.g., foreground or killed) of LLM requests. Specifically, for semantic-level contexts, MobiLoRA proposes similarity-aware delta encoding, which leverages token-wise similarity in KV caches across LoRA adapters for efficient storage and reuse. Furthermore, MobiLoRA advocates context-aware KV cache management to optimize cache retention and eviction considering the system-level contexts. We fully implement MobiLoRA and compare it with state-of-the-art LLM serving frameworks using real-world mobile device traces. Results show that MobiLoRA accelerates LoRA-based LLM inference by 57.6% on mobile devices.
%R 10.18653/v1/2025.acl-long.1140
%U https://aclanthology.org/2025.acl-long.1140/
%U https://doi.org/10.18653/v1/2025.acl-long.1140
%P 23400-23410
Markdown (Informal)
[MobiLoRA: Accelerating LoRA-based LLM Inference on Mobile Devices via Context-aware KV Cache Optimization](https://aclanthology.org/2025.acl-long.1140/) (Li et al., ACL 2025)
ACL