Saved in:
Bibliographic Details
Main Authors: Zhong, Yibo, Jiang, Haoxiang, Li, Lincan, Nakada, Ryumei, Liu, Tianci, Zhang, Linjun, Yao, Huaxiu, Wang, Haoyu
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2410.01870
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909921001340928
author Zhong, Yibo
Jiang, Haoxiang
Li, Lincan
Nakada, Ryumei
Liu, Tianci
Zhang, Linjun
Yao, Huaxiu
Wang, Haoyu
author_facet Zhong, Yibo
Jiang, Haoxiang
Li, Lincan
Nakada, Ryumei
Liu, Tianci
Zhang, Linjun
Yao, Huaxiu
Wang, Haoyu
contents Fine-tuning large pre-trained foundation models often yields excellent downstream performance but is prohibitively expensive when updating all parameters. Parameter-efficient fine-tuning (PEFT) methods such as LoRA alleviate this by introducing lightweight update modules, yet they commonly rely on weight-agnostic linear approximations, limiting their expressiveness. In this work, we propose PEANuT, a novel PEFT framework that introduces weight-aware neural tweakers, compact neural modules that generate task-adaptive updates conditioned on frozen pre-trained weights. PEANuT provides a flexible yet efficient way to capture complex update patterns without full model tuning. We theoretically show that PEANuT achieves equivalent or greater expressivity than existing linear PEFT methods with comparable or fewer parameters. Extensive experiments across four benchmarks with over twenty datasets demonstrate that PEANuT consistently outperforms strong baselines in both NLP and vision tasks, while maintaining low computational overhead.
format Preprint
id arxiv_https___arxiv_org_abs_2410_01870
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle PEANuT: Parameter-Efficient Adaptation with Weight-aware Neural Tweakers
Zhong, Yibo
Jiang, Haoxiang
Li, Lincan
Nakada, Ryumei
Liu, Tianci
Zhang, Linjun
Yao, Huaxiu
Wang, Haoyu
Machine Learning
Computation and Language
Fine-tuning large pre-trained foundation models often yields excellent downstream performance but is prohibitively expensive when updating all parameters. Parameter-efficient fine-tuning (PEFT) methods such as LoRA alleviate this by introducing lightweight update modules, yet they commonly rely on weight-agnostic linear approximations, limiting their expressiveness. In this work, we propose PEANuT, a novel PEFT framework that introduces weight-aware neural tweakers, compact neural modules that generate task-adaptive updates conditioned on frozen pre-trained weights. PEANuT provides a flexible yet efficient way to capture complex update patterns without full model tuning. We theoretically show that PEANuT achieves equivalent or greater expressivity than existing linear PEFT methods with comparable or fewer parameters. Extensive experiments across four benchmarks with over twenty datasets demonstrate that PEANuT consistently outperforms strong baselines in both NLP and vision tasks, while maintaining low computational overhead.
title PEANuT: Parameter-Efficient Adaptation with Weight-aware Neural Tweakers
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2410.01870