Cross-LoRA: A Data-Free LoRA Transfer Framework across Heterogeneous LLMs

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Xia, Feifan, Liao, Mingyang, Fang, Yuyang, Li, Defang, Xie, Yantong, Li, Weikang, Li, Yang, Xia, Deguo, Huang, Jizhou
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908481007648768
author Xia, Feifan
Liao, Mingyang
Fang, Yuyang
Li, Defang
Xie, Yantong
Li, Weikang
Li, Yang
Xia, Deguo
Huang, Jizhou
author_facet Xia, Feifan
Liao, Mingyang
Fang, Yuyang
Li, Defang
Xie, Yantong
Li, Weikang
Li, Yang
Xia, Deguo
Huang, Jizhou
contents Traditional parameter-efficient fine-tuning (PEFT) methods such as LoRA are tightly coupled with the base model architecture, which constrains their applicability across heterogeneous pretrained large language models (LLMs). To address this limitation, we introduce Cross-LoRA, a data-free framework for transferring LoRA modules between diverse base models without requiring additional training data. Cross-LoRA consists of two key components: (a) LoRA-Align, which performs subspace alignment between source and target base models through rank-truncated singular value decomposition (SVD) and Frobenius-optimal linear transformation, ensuring compatibility under dimension mismatch; and (b) LoRA-Shift, which applies the aligned subspaces to project source LoRA weight updates into the target model parameter space. Both components are data-free, training-free, and enable lightweight adaptation on a commodity GPU in 20 minutes. Experiments on ARCs, OBOA and HellaSwag show that Cross-LoRA achieves relative gains of up to 5.26% over base models. Across other commonsense reasoning benchmarks, Cross-LoRA maintains performance comparable to that of directly trained LoRA adapters.
format Preprint
id arxiv_https___arxiv_org_abs_2508_05232
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Cross-LoRA: A Data-Free LoRA Transfer Framework across Heterogeneous LLMs
Xia, Feifan
Liao, Mingyang
Fang, Yuyang
Li, Defang
Xie, Yantong
Li, Weikang
Li, Yang
Xia, Deguo
Huang, Jizhou
Machine Learning
Traditional parameter-efficient fine-tuning (PEFT) methods such as LoRA are tightly coupled with the base model architecture, which constrains their applicability across heterogeneous pretrained large language models (LLMs). To address this limitation, we introduce Cross-LoRA, a data-free framework for transferring LoRA modules between diverse base models without requiring additional training data. Cross-LoRA consists of two key components: (a) LoRA-Align, which performs subspace alignment between source and target base models through rank-truncated singular value decomposition (SVD) and Frobenius-optimal linear transformation, ensuring compatibility under dimension mismatch; and (b) LoRA-Shift, which applies the aligned subspaces to project source LoRA weight updates into the target model parameter space. Both components are data-free, training-free, and enable lightweight adaptation on a commodity GPU in 20 minutes. Experiments on ARCs, OBOA and HellaSwag show that Cross-LoRA achieves relative gains of up to 5.26% over base models. Across other commonsense reasoning benchmarks, Cross-LoRA maintains performance comparable to that of directly trained LoRA adapters.
title Cross-LoRA: A Data-Free LoRA Transfer Framework across Heterogeneous LLMs
topic Machine Learning
url https://arxiv.org/abs/2508.05232