Guardado en:
Detalles Bibliográficos
Autores principales: Schafhalter, Peter, Liao, Shun, Zhou, Yanqi, Yeh, Chih-Kuan, Kandoor, Arun, Laudon, James
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:https://arxiv.org/abs/2410.10181
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866910663835648000
author Schafhalter, Peter
Liao, Shun
Zhou, Yanqi
Yeh, Chih-Kuan
Kandoor, Arun
Laudon, James
author_facet Schafhalter, Peter
Liao, Shun
Zhou, Yanqi
Yeh, Chih-Kuan
Kandoor, Arun
Laudon, James
contents Domain-specific adaptation is critical to maximizing the performance of pre-trained language models (PLMs) on one or multiple targeted tasks, especially under resource-constrained use cases, such as edge devices. However, existing methods often struggle to balance domain-specific performance, retention of general knowledge, and efficiency for training and inference. To address these challenges, we propose Modular Domain Experts (MoDE). MoDE is a mixture-of-experts architecture that augments a general PLMs with modular, domain-specialized experts. These experts are trained independently and composed together via a lightweight training process. In contrast to standard low-rank adaptation methods, each MoDE expert consists of several transformer layers which scale better with more training examples and larger parameter counts. Our evaluation demonstrates that MoDE achieves comparable target performances to full parameter fine-tuning while achieving 1.65% better retention performance. Moreover, MoDE's architecture enables flexible sharding configurations and improves training speeds by up to 38% over state-of-the-art distributed training configurations.
format Preprint
id arxiv_https___arxiv_org_abs_2410_10181
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Scalable Multi-Domain Adaptation of Language Models using Modular Experts
Schafhalter, Peter
Liao, Shun
Zhou, Yanqi
Yeh, Chih-Kuan
Kandoor, Arun
Laudon, James
Computation and Language
Artificial Intelligence
Domain-specific adaptation is critical to maximizing the performance of pre-trained language models (PLMs) on one or multiple targeted tasks, especially under resource-constrained use cases, such as edge devices. However, existing methods often struggle to balance domain-specific performance, retention of general knowledge, and efficiency for training and inference. To address these challenges, we propose Modular Domain Experts (MoDE). MoDE is a mixture-of-experts architecture that augments a general PLMs with modular, domain-specialized experts. These experts are trained independently and composed together via a lightweight training process. In contrast to standard low-rank adaptation methods, each MoDE expert consists of several transformer layers which scale better with more training examples and larger parameter counts. Our evaluation demonstrates that MoDE achieves comparable target performances to full parameter fine-tuning while achieving 1.65% better retention performance. Moreover, MoDE's architecture enables flexible sharding configurations and improves training speeds by up to 38% over state-of-the-art distributed training configurations.
title Scalable Multi-Domain Adaptation of Language Models using Modular Experts
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2410.10181