Saved in:
Bibliographic Details
Main Authors: Do, Giang, Pham, Kha, Le, Hung, Tran, Truyen
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2411.19402
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916865906835456
author Do, Giang
Pham, Kha
Le, Hung
Tran, Truyen
author_facet Do, Giang
Pham, Kha
Le, Hung
Tran, Truyen
contents Sparse mixture of experts (SMoE) is an effective solution for scaling up model capacity without increasing the computational costs. A crucial component of SMoE is the router, responsible for directing the input to relevant experts; however, it also presents a major weakness, leading to routing inconsistencies and representation collapse issues. Instead of fixing the router like previous works, we propose an alternative that assigns experts to input via indirection, which employs the discrete representation of input that points to the expert. The discrete representations are learnt via vector quantization, resulting in a new architecture dubbed Vector-Quantized Mixture of Experts (VQMoE). We provide theoretical support and empirical evidence demonstrating the VQMoE's ability to overcome the challenges present in traditional routers. Through extensive evaluations on both large language models and vision tasks for pre-training and fine-tuning, we show that VQMoE achieves a 28% improvement in robustness compared to other SMoE routing methods, while maintaining strong performance in fine-tuning tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2411_19402
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle On the Role of Discrete Representation in Sparse Mixture of Experts
Do, Giang
Pham, Kha
Le, Hung
Tran, Truyen
Machine Learning
Sparse mixture of experts (SMoE) is an effective solution for scaling up model capacity without increasing the computational costs. A crucial component of SMoE is the router, responsible for directing the input to relevant experts; however, it also presents a major weakness, leading to routing inconsistencies and representation collapse issues. Instead of fixing the router like previous works, we propose an alternative that assigns experts to input via indirection, which employs the discrete representation of input that points to the expert. The discrete representations are learnt via vector quantization, resulting in a new architecture dubbed Vector-Quantized Mixture of Experts (VQMoE). We provide theoretical support and empirical evidence demonstrating the VQMoE's ability to overcome the challenges present in traditional routers. Through extensive evaluations on both large language models and vision tasks for pre-training and fine-tuning, we show that VQMoE achieves a 28% improvement in robustness compared to other SMoE routing methods, while maintaining strong performance in fine-tuning tasks.
title On the Role of Discrete Representation in Sparse Mixture of Experts
topic Machine Learning
url https://arxiv.org/abs/2411.19402