Dual LoRA: Enhancing LoRA with Magnitude and Direction Updates

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Xu, Yixing, Li, Chao, Yin, Xuanwu, Tiwari, Spandan, Li, Dong, Sirasao, Ashish, Barsoum, Emad
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915702928048128
author Xu, Yixing
Li, Chao
Yin, Xuanwu
Tiwari, Spandan
Li, Dong
Sirasao, Ashish
Barsoum, Emad
author_facet Xu, Yixing
Li, Chao
Yin, Xuanwu
Tiwari, Spandan
Li, Dong
Sirasao, Ashish
Barsoum, Emad
contents Low-rank adaptation (LoRA) is one of the most popular methods among parameter-efficient fine-tuning (PEFT) methods to adapt pre-trained large language models (LLMs) to specific downstream tasks. However, the model trained based on LoRA often has an unsatisfactory performance due to its low-rank assumption. In this paper, we propose a novel method called Dual LoRA to improve the performance by incorporating an inductive bias into the original LoRA. Specifically, we separate low-rank matrices into two groups: the magnitude group to control whether or not and how far we should update a parameter and the direction group to decide whether this parameter should move forward or backward, to better simulate the parameter updating process of the full fine-tuning based on gradient-based optimization algorithms. We show that this can be simply achieved by adding a ReLU function to the magnitude group and a sign function to the direction group. We conduct several experiments over a wide range of NLP tasks, including natural language understanding (NLU) and commonsense reasoning datasets on RoBERTa, DeBERTa, and LLaMA-1/2/3 as baseline models. The results show that we consistently outperform LoRA and its state-of-the-art variants with the same number of trainable parameters.
format Preprint
id arxiv_https___arxiv_org_abs_2512_03402
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Dual LoRA: Enhancing LoRA with Magnitude and Direction Updates
Xu, Yixing
Li, Chao
Yin, Xuanwu
Tiwari, Spandan
Li, Dong
Sirasao, Ashish
Barsoum, Emad
Computation and Language
Low-rank adaptation (LoRA) is one of the most popular methods among parameter-efficient fine-tuning (PEFT) methods to adapt pre-trained large language models (LLMs) to specific downstream tasks. However, the model trained based on LoRA often has an unsatisfactory performance due to its low-rank assumption. In this paper, we propose a novel method called Dual LoRA to improve the performance by incorporating an inductive bias into the original LoRA. Specifically, we separate low-rank matrices into two groups: the magnitude group to control whether or not and how far we should update a parameter and the direction group to decide whether this parameter should move forward or backward, to better simulate the parameter updating process of the full fine-tuning based on gradient-based optimization algorithms. We show that this can be simply achieved by adding a ReLU function to the magnitude group and a sign function to the direction group. We conduct several experiments over a wide range of NLP tasks, including natural language understanding (NLU) and commonsense reasoning datasets on RoBERTa, DeBERTa, and LLaMA-1/2/3 as baseline models. The results show that we consistently outperform LoRA and its state-of-the-art variants with the same number of trainable parameters.
title Dual LoRA: Enhancing LoRA with Magnitude and Direction Updates
topic Computation and Language
url https://arxiv.org/abs/2512.03402