Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Camacho, Albert Manuel Orozco, Horoi, Stefan, Wolf, Guy, Belilovsky, Eugene
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2412.15467
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866913620711964672
author Camacho, Albert Manuel Orozco
Horoi, Stefan
Wolf, Guy
Belilovsky, Eugene
author_facet Camacho, Albert Manuel Orozco
Horoi, Stefan
Wolf, Guy
Belilovsky, Eugene
contents Combining multiple machine learning models has long been a technique for enhancing performance, particularly in distributed settings. Traditional approaches, such as model ensembles, work well, but are expensive in terms of memory and compute. Recently, methods based on averaging model parameters have achieved good results in some settings and have gained popularity. However, merging models initialized differently that do not share a part of their training trajectories can yield worse results than simply using the base models, even after aligning their neurons. In this paper, we introduce a novel approach, Non-uniform Parameter-wise Model Merging, or NP Merge, which merges models by learning the contribution of each parameter to the final model using gradient-based optimization. We empirically demonstrate the effectiveness of our method for merging models of various architectures in multiple settings, outperforming past methods. We also extend NP Merge to handle the merging of multiple models, showcasing its scalability and robustness.
format Preprint
id arxiv_https___arxiv_org_abs_2412_15467
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Non-Uniform Parameter-Wise Model Merging
Camacho, Albert Manuel Orozco
Horoi, Stefan
Wolf, Guy
Belilovsky, Eugene
Machine Learning
Artificial Intelligence
Combining multiple machine learning models has long been a technique for enhancing performance, particularly in distributed settings. Traditional approaches, such as model ensembles, work well, but are expensive in terms of memory and compute. Recently, methods based on averaging model parameters have achieved good results in some settings and have gained popularity. However, merging models initialized differently that do not share a part of their training trajectories can yield worse results than simply using the base models, even after aligning their neurons. In this paper, we introduce a novel approach, Non-uniform Parameter-wise Model Merging, or NP Merge, which merges models by learning the contribution of each parameter to the final model using gradient-based optimization. We empirically demonstrate the effectiveness of our method for merging models of various architectures in multiple settings, outperforming past methods. We also extend NP Merge to handle the merging of multiple models, showcasing its scalability and robustness.
title Non-Uniform Parameter-Wise Model Merging
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2412.15467