HydraOpt: Navigating the Efficiency-Performance Trade-off of Adapter Merging

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ceritli, Taha, Bohdal, Ondrej, Ozay, Mete, Moon, Jijoong, Lee, Kyeng-Hun, Ko, Hyeonmok, Michieli, Umberto
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918102798696448
author Ceritli, Taha
Bohdal, Ondrej
Ozay, Mete
Moon, Jijoong
Lee, Kyeng-Hun
Ko, Hyeonmok
Michieli, Umberto
author_facet Ceritli, Taha
Bohdal, Ondrej
Ozay, Mete
Moon, Jijoong
Lee, Kyeng-Hun
Ko, Hyeonmok
Michieli, Umberto
contents Large language models (LLMs) often leverage adapters, such as low-rank-based adapters, to achieve strong performance on downstream tasks. However, storing a separate adapter for each task significantly increases memory requirements, posing a challenge for resource-constrained environments such as mobile devices. Although model merging techniques can reduce storage costs, they typically result in substantial performance degradation. In this work, we introduce HydraOpt, a new model merging technique that capitalizes on the inherent similarities between the matrices of low-rank adapters. Unlike existing methods that produce a fixed trade-off between storage size and performance, HydraOpt allows us to navigate this spectrum of efficiency and performance. Our experiments show that HydraOpt significantly reduces storage size (48% reduction) compared to storing all adapters, while achieving competitive performance (0.2-1.8% drop). Furthermore, it outperforms existing merging techniques in terms of performance at the same or slightly worse storage efficiency.
format Preprint
id arxiv_https___arxiv_org_abs_2507_17706
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle HydraOpt: Navigating the Efficiency-Performance Trade-off of Adapter Merging
Ceritli, Taha
Bohdal, Ondrej
Ozay, Mete
Moon, Jijoong
Lee, Kyeng-Hun
Ko, Hyeonmok
Michieli, Umberto
Machine Learning
Large language models (LLMs) often leverage adapters, such as low-rank-based adapters, to achieve strong performance on downstream tasks. However, storing a separate adapter for each task significantly increases memory requirements, posing a challenge for resource-constrained environments such as mobile devices. Although model merging techniques can reduce storage costs, they typically result in substantial performance degradation. In this work, we introduce HydraOpt, a new model merging technique that capitalizes on the inherent similarities between the matrices of low-rank adapters. Unlike existing methods that produce a fixed trade-off between storage size and performance, HydraOpt allows us to navigate this spectrum of efficiency and performance. Our experiments show that HydraOpt significantly reduces storage size (48% reduction) compared to storing all adapters, while achieving competitive performance (0.2-1.8% drop). Furthermore, it outperforms existing merging techniques in terms of performance at the same or slightly worse storage efficiency.
title HydraOpt: Navigating the Efficiency-Performance Trade-off of Adapter Merging
topic Machine Learning
url https://arxiv.org/abs/2507.17706