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

TL;DR

Cross-LoRA is a data-free, efficient transfer framework that enables LoRA modules to adapt across different large language models without additional training, broadening their applicability.

Contribution

It introduces a novel data-free transfer method for LoRA modules across heterogeneous LLMs using subspace alignment and projection techniques.

Findings

Achieves up to 5.26% performance gains on certain benchmarks.

Maintains comparable performance to directly trained LoRA adapters.

Enables lightweight adaptation in 20 minutes on a standard GPU.

Abstract

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.