기본 정보
연구 분야
프로젝트
발행물
구성원
article|
인용수 0
·2025
ReAx: Resource-efficient Asynchronous Execution for Accelerating LLM Fine-tuning at the Edge
Heung Sik Na, Daeseon Choi, Young‐Ho Gong, Young Geun Kim
IF 2IEEE Embedded Systems Letters
초록

With the widespread use of Large Language Models (LLMs), there is an increasing demand for personalized models that meet diverse needs of users. To enable private, network-independent personalization of LLMs, on-device fine-tuning is receiving much attention. However, on-device fine-tuning faces efficiency and scalability challenges as sequential execution of compute- and memory-intensive operations often under-utilizes resources. In this letter, we propose ReAx, a framework that accelerates on-device fine-tuning through resource efficient asynchronous parallel execution of memory- and compute-intensive operations. Without increasing memory usage, ReAx improves the average fine-tuning performance and energy consumption by 10.42% and 5.55%, respectively, compared to the baseline. As a positive side effect, asynchronous parameter updates induce gradient noise due to slight delays between streams, which act as a regularizer against adverse updates across mini-batches, without sacrificing accuracy.

키워드
Asynchronous communicationScalabilityPersonalizationEnergy consumptionEnhanced Data Rates for GSM EvolutionResource (disambiguation)Noise (video)Efficient energy useResource consumption
타입
article
IF / 인용수
2 / 0
게재 연도
2025