LoRASuite: Efficient LoRA Adaptation Across Large Language Model Upgrades

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Li, Yanan, Meng, Fanxu, Zhang, Muhan, Zhu, Shiai, Wang, Shangguang, Xu, Mengwei
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908371071795200
author Li, Yanan
Meng, Fanxu
Zhang, Muhan
Zhu, Shiai
Wang, Shangguang
Xu, Mengwei
author_facet Li, Yanan
Meng, Fanxu
Zhang, Muhan
Zhu, Shiai
Wang, Shangguang
Xu, Mengwei
contents As Large Language Models (LLMs) are frequently updated, LoRA weights trained on earlier versions quickly become obsolete. The conventional practice of retraining LoRA weights from scratch on the latest model is costly, time-consuming, and environmentally detrimental, particularly as the diversity of LLMs and downstream tasks expands. This motivates a critical question: "How can we efficiently leverage existing LoRA weights to adapt to newer model versions?" To address this, we propose LoRASuite, a modular approach tailored specifically to various types of LLM updates. First, we compute a transfer matrix utilizing known parameters from both old and new LLMs. Next, we allocate corresponding layers and attention heads based on centered kernel alignment and cosine similarity metrics, respectively. A subsequent small-scale, skillful fine-tuning step ensures numerical stability. Experimental evaluations demonstrate that LoRASuite consistently surpasses small-scale vanilla LoRA methods. Notably, on backbone LLMs such as MiniCPM and Qwen, LoRASuite even exceeds the performance of full-scale LoRA retraining, with average improvements of +1.4 and +6.6 points on math tasks, respectively. Additionally, LoRASuite significantly reduces memory consumption by 5.5 GB and computational time by 78.23%.
format Preprint
id arxiv_https___arxiv_org_abs_2505_13515
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LoRASuite: Efficient LoRA Adaptation Across Large Language Model Upgrades
Li, Yanan
Meng, Fanxu
Zhang, Muhan
Zhu, Shiai
Wang, Shangguang
Xu, Mengwei
Machine Learning
Artificial Intelligence
Computation and Language
As Large Language Models (LLMs) are frequently updated, LoRA weights trained on earlier versions quickly become obsolete. The conventional practice of retraining LoRA weights from scratch on the latest model is costly, time-consuming, and environmentally detrimental, particularly as the diversity of LLMs and downstream tasks expands. This motivates a critical question: "How can we efficiently leverage existing LoRA weights to adapt to newer model versions?" To address this, we propose LoRASuite, a modular approach tailored specifically to various types of LLM updates. First, we compute a transfer matrix utilizing known parameters from both old and new LLMs. Next, we allocate corresponding layers and attention heads based on centered kernel alignment and cosine similarity metrics, respectively. A subsequent small-scale, skillful fine-tuning step ensures numerical stability. Experimental evaluations demonstrate that LoRASuite consistently surpasses small-scale vanilla LoRA methods. Notably, on backbone LLMs such as MiniCPM and Qwen, LoRASuite even exceeds the performance of full-scale LoRA retraining, with average improvements of +1.4 and +6.6 points on math tasks, respectively. Additionally, LoRASuite significantly reduces memory consumption by 5.5 GB and computational time by 78.23%.
title LoRASuite: Efficient LoRA Adaptation Across Large Language Model Upgrades
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2505.13515