Mix Data or Merge Models? Optimizing for Diverse Multi-Task Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Aakanksha, Ahmadian, Arash, Goldfarb-Tarrant, Seraphina, Ermis, Beyza, Fadaee, Marzieh, Hooker, Sara
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917802215997440
author Aakanksha
Ahmadian, Arash
Goldfarb-Tarrant, Seraphina
Ermis, Beyza
Fadaee, Marzieh
Hooker, Sara
author_facet Aakanksha
Ahmadian, Arash
Goldfarb-Tarrant, Seraphina
Ermis, Beyza
Fadaee, Marzieh
Hooker, Sara
contents Large Language Models (LLMs) have been adopted and deployed worldwide for a broad variety of applications. However, ensuring their safe use remains a significant challenge. Preference training and safety measures often overfit to harms prevalent in Western-centric datasets, and safety protocols frequently fail to extend to multilingual settings. In this work, we explore model merging in a diverse multi-task setting, combining safety and general-purpose tasks within a multilingual context. Each language introduces unique and varied learning challenges across tasks. We find that objective-based merging is more effective than mixing data, with improvements of up to 8% and 10% in general performance and safety respectively. We also find that language-based merging is highly effective -- by merging monolingually fine-tuned models, we achieve a 4% increase in general performance and 7% reduction in harm across all languages on top of the data mixtures method using the same available data. Overall, our comprehensive study of merging approaches provides a useful framework for building strong and safe multilingual models.
format Preprint
id arxiv_https___arxiv_org_abs_2410_10801
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Mix Data or Merge Models? Optimizing for Diverse Multi-Task Learning
Aakanksha
Ahmadian, Arash
Goldfarb-Tarrant, Seraphina
Ermis, Beyza
Fadaee, Marzieh
Hooker, Sara
Computation and Language
Machine Learning
Large Language Models (LLMs) have been adopted and deployed worldwide for a broad variety of applications. However, ensuring their safe use remains a significant challenge. Preference training and safety measures often overfit to harms prevalent in Western-centric datasets, and safety protocols frequently fail to extend to multilingual settings. In this work, we explore model merging in a diverse multi-task setting, combining safety and general-purpose tasks within a multilingual context. Each language introduces unique and varied learning challenges across tasks. We find that objective-based merging is more effective than mixing data, with improvements of up to 8% and 10% in general performance and safety respectively. We also find that language-based merging is highly effective -- by merging monolingually fine-tuned models, we achieve a 4% increase in general performance and 7% reduction in harm across all languages on top of the data mixtures method using the same available data. Overall, our comprehensive study of merging approaches provides a useful framework for building strong and safe multilingual models.
title Mix Data or Merge Models? Optimizing for Diverse Multi-Task Learning
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2410.10801