DoRA: Weight-Decomposed Low-Rank Adaptation

TL;DR

This paper introduces DoRA, a weight-decomposed low-rank adaptation method that improves fine-tuning accuracy and stability of pre-trained models without increasing inference costs.

Contribution

The paper proposes DoRA, a novel weight decomposition approach that enhances LoRA's learning capacity and stability, narrowing the accuracy gap with full fine-tuning.

Findings

DoRA outperforms LoRA on multiple benchmarks

It improves fine-tuning stability and capacity

Effective across various models and tasks

Abstract

Among the widely used parameter-efficient fine-tuning (PEFT) methods, LoRA and its variants have gained considerable popularity because of avoiding additional inference costs. However, there still often exists an accuracy gap between these methods and full fine-tuning (FT). In this work, we first introduce a novel weight decomposition analysis to investigate the inherent differences between FT and LoRA. Aiming to resemble the learning capacity of FT from the findings, we propose Weight-Decomposed Low-Rank Adaptation (DoRA). DoRA decomposes the pre-trained weight into two components, magnitude and direction, for fine-tuning, specifically employing LoRA for directional updates to efficiently minimize the number of trainable parameters. By employing \ours, we enhance both the learning capacity and training stability of LoRA while avoiding any additional inference overhead. \ours~consistently outperforms LoRA on fine-tuning LLaMA, LLaVA, and VL-BART on various downstream tasks, such as commonsense reasoning, visual instruction tuning, and image/video-text understanding. Code is available at https://github.com/NVlabs/DoRA.