WeightLoRA: Keep Only Necessary Adapters

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Veprikov, Andrey, Solodkin, Vladimir, Zyl, Alexander, Savchenko, Andrey, Beznosikov, Aleksandr
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866909848961024000
author Veprikov, Andrey
Solodkin, Vladimir
Zyl, Alexander
Savchenko, Andrey
Beznosikov, Aleksandr
author_facet Veprikov, Andrey
Solodkin, Vladimir
Zyl, Alexander
Savchenko, Andrey
Beznosikov, Aleksandr
contents The widespread utilization of language models in modern applications is inconceivable without Parameter-Efficient Fine-Tuning techniques, such as low-rank adaptation ($\texttt{LoRA}$), which adds trainable adapters to selected layers. Although $\texttt{LoRA}$ may obtain accurate solutions, it requires significant memory to train large models and intuition on which layers to add adapters. In this paper, we propose a novel method, $\texttt{WeightLoRA}$, which overcomes this issue by adaptive selection of the most critical $\texttt{LoRA}$ heads throughout the optimization process. As a result, we can significantly reduce the number of trainable parameters while maintaining the capability to obtain consistent or even superior metric values. We conduct experiments for a series of competitive benchmarks and DeBERTa, BART, and Llama models, comparing our method with different adaptive approaches. The experimental results demonstrate the efficacy of $\texttt{WeightLoRA}$ and the superior performance of $\texttt{WeightLoRA+}$ in almost all cases.
format Preprint
id arxiv_https___arxiv_org_abs_2506_02724
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle WeightLoRA: Keep Only Necessary Adapters
Veprikov, Andrey
Solodkin, Vladimir
Zyl, Alexander
Savchenko, Andrey
Beznosikov, Aleksandr
Machine Learning
Optimization and Control
The widespread utilization of language models in modern applications is inconceivable without Parameter-Efficient Fine-Tuning techniques, such as low-rank adaptation ($\texttt{LoRA}$), which adds trainable adapters to selected layers. Although $\texttt{LoRA}$ may obtain accurate solutions, it requires significant memory to train large models and intuition on which layers to add adapters. In this paper, we propose a novel method, $\texttt{WeightLoRA}$, which overcomes this issue by adaptive selection of the most critical $\texttt{LoRA}$ heads throughout the optimization process. As a result, we can significantly reduce the number of trainable parameters while maintaining the capability to obtain consistent or even superior metric values. We conduct experiments for a series of competitive benchmarks and DeBERTa, BART, and Llama models, comparing our method with different adaptive approaches. The experimental results demonstrate the efficacy of $\texttt{WeightLoRA}$ and the superior performance of $\texttt{WeightLoRA+}$ in almost all cases.
title WeightLoRA: Keep Only Necessary Adapters
topic Machine Learning
Optimization and Control
url https://arxiv.org/abs/2506.02724