SARA: Singular-Value Based Adaptive Low-Rank Adaption

TL;DR

SARA introduces an adaptive low-rank fine-tuning method using SVD to automatically determine layer ranks, improving parameter efficiency and performance across diverse tasks.

Contribution

The paper proposes SARA, a novel method that adaptively determines layer ranks during fine-tuning, addressing manual rank selection and layer importance issues in LoRA.

Findings

Effectively finds optimal layer ranks adaptively

Reduces parameters with Mixture-of-SARA

Demonstrates strong performance on complex tasks

Abstract

With the increasing number of parameters in large pre-trained models, LoRA as a parameter-efficient fine-tuning(PEFT) method is widely used for not adding inference overhead. The LoRA method assumes that weight changes during fine-tuning can be approximated by low-rank matrices. However, the rank values need to be manually verified to match different downstream tasks, and they cannot accommodate the varying importance of different layers in the model. In this work, we first analyze the relationship between the performance of different layers and their ranks using SVD. Based on this, we design the Singular-Value Based Adaptive Low-Rank Adaption(SARA), which adaptively finds the rank during initialization by performing SVD on the pre-trained weights. Additionally, we explore the Mixture-of-SARA(Mo-SARA), which significantly reduces the number of parameters by fine-tuning only multiple parallel sets of singular values controlled by a router. Extensive experiments on various complex tasks demonstrate the simplicity and parameter efficiency of our methods. They can effectively and adaptively find the most suitable rank for each layer of each model.