@inproceedings{do-etal-2023-hyperrouter,
title = "{H}yper{R}outer: Towards Efficient Training and Inference of Sparse Mixture of Experts",
author = "Do, Truong Giang and
Khiem, Le and
Pham, Quang and
Nguyen, TrungTin and
Doan, Thanh-Nam and
Nguyen, Binh and
Liu, Chenghao and
Ramasamy, Savitha and
Li, Xiaoli and
Hoi, Steven",
editor = "Bouamor, Houda and
Pino, Juan and
Bali, Kalika",
booktitle = "Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing",
month = dec,
year = "2023",
address = "Singapore",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.emnlp-main.351/",
doi = "10.18653/v1/2023.emnlp-main.351",
pages = "5754--5765",
abstract = "By routing input tokens to only a few split experts, Sparse Mixture-of-Experts has enabled efficient training of large language models. Recent findings suggest that fixing the routers can achieve competitive performance by alleviating the collapsing problem, where all experts eventually learn similar representations. However, this strategy has two key limitations: (i) the policy derived from random routers might be sub-optimal, and (ii) it requires extensive resources during training and evaluation, leading to limited efficiency gains. This work introduces HyperRouter, which dynamically generates the router`s parameters through a fixed hypernetwork and trainable embeddings to achieve a balance between training the routers and freezing them to learn an improved routing policy. Extensive experiments across a wide range of tasks demonstrate the superior performance and efficiency gains of HyperRouter compared to existing routing methods. Our implementation is publicly available at \url{https://github.com/giangdip2410/HyperRouter}."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="do-etal-2023-hyperrouter">
<titleInfo>
<title>HyperRouter: Towards Efficient Training and Inference of Sparse Mixture of Experts</title>
</titleInfo>
<name type="personal">
<namePart type="given">Truong</namePart>
<namePart type="given">Giang</namePart>
<namePart type="family">Do</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Le</namePart>
<namePart type="family">Khiem</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Quang</namePart>
<namePart type="family">Pham</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">TrungTin</namePart>
<namePart type="family">Nguyen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Thanh-Nam</namePart>
<namePart type="family">Doan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Binh</namePart>
<namePart type="family">Nguyen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chenghao</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Savitha</namePart>
<namePart type="family">Ramasamy</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaoli</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Steven</namePart>
<namePart type="family">Hoi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2023-12</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Houda</namePart>
<namePart type="family">Bouamor</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Juan</namePart>
<namePart type="family">Pino</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kalika</namePart>
<namePart type="family">Bali</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Singapore</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>By routing input tokens to only a few split experts, Sparse Mixture-of-Experts has enabled efficient training of large language models. Recent findings suggest that fixing the routers can achieve competitive performance by alleviating the collapsing problem, where all experts eventually learn similar representations. However, this strategy has two key limitations: (i) the policy derived from random routers might be sub-optimal, and (ii) it requires extensive resources during training and evaluation, leading to limited efficiency gains. This work introduces HyperRouter, which dynamically generates the router‘s parameters through a fixed hypernetwork and trainable embeddings to achieve a balance between training the routers and freezing them to learn an improved routing policy. Extensive experiments across a wide range of tasks demonstrate the superior performance and efficiency gains of HyperRouter compared to existing routing methods. Our implementation is publicly available at https://github.com/giangdip2410/HyperRouter.</abstract>
<identifier type="citekey">do-etal-2023-hyperrouter</identifier>
<identifier type="doi">10.18653/v1/2023.emnlp-main.351</identifier>
<location>
<url>https://aclanthology.org/2023.emnlp-main.351/</url>
</location>
<part>
<date>2023-12</date>
<extent unit="page">
<start>5754</start>
<end>5765</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T HyperRouter: Towards Efficient Training and Inference of Sparse Mixture of Experts
%A Do, Truong Giang
%A Khiem, Le
%A Pham, Quang
%A Nguyen, TrungTin
%A Doan, Thanh-Nam
%A Nguyen, Binh
%A Liu, Chenghao
%A Ramasamy, Savitha
%A Li, Xiaoli
%A Hoi, Steven
%Y Bouamor, Houda
%Y Pino, Juan
%Y Bali, Kalika
%S Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing
%D 2023
%8 December
%I Association for Computational Linguistics
%C Singapore
%F do-etal-2023-hyperrouter
%X By routing input tokens to only a few split experts, Sparse Mixture-of-Experts has enabled efficient training of large language models. Recent findings suggest that fixing the routers can achieve competitive performance by alleviating the collapsing problem, where all experts eventually learn similar representations. However, this strategy has two key limitations: (i) the policy derived from random routers might be sub-optimal, and (ii) it requires extensive resources during training and evaluation, leading to limited efficiency gains. This work introduces HyperRouter, which dynamically generates the router‘s parameters through a fixed hypernetwork and trainable embeddings to achieve a balance between training the routers and freezing them to learn an improved routing policy. Extensive experiments across a wide range of tasks demonstrate the superior performance and efficiency gains of HyperRouter compared to existing routing methods. Our implementation is publicly available at https://github.com/giangdip2410/HyperRouter.
%R 10.18653/v1/2023.emnlp-main.351
%U https://aclanthology.org/2023.emnlp-main.351/
%U https://doi.org/10.18653/v1/2023.emnlp-main.351
%P 5754-5765
Markdown (Informal)
[HyperRouter: Towards Efficient Training and Inference of Sparse Mixture of Experts](https://aclanthology.org/2023.emnlp-main.351/) (Do et al., EMNLP 2023)
ACL
- Truong Giang Do, Le Khiem, Quang Pham, TrungTin Nguyen, Thanh-Nam Doan, Binh Nguyen, Chenghao Liu, Savitha Ramasamy, Xiaoli Li, and Steven Hoi. 2023. HyperRouter: Towards Efficient Training and Inference of Sparse Mixture of Experts. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 5754–5765, Singapore. Association for Computational Linguistics.