Enregistré dans:
Détails bibliographiques
Auteurs principaux: Do, Giang, Le, Hung, Tran, Truyen
Format: Preprint
Publié: 2024
Sujets:
Accès en ligne:https://arxiv.org/abs/2406.15883
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866916297748512768
author Do, Giang
Le, Hung
Tran, Truyen
author_facet Do, Giang
Le, Hung
Tran, Truyen
contents Sparse mixture of experts (SMoE) have emerged as an effective approach for scaling large language models while keeping a constant computational cost. Regardless of several notable successes of SMoE, effective training such architecture remains elusive due to the representation collapse problem, which in turn harms model performance and causes parameter redundancy. In this work, we present Similarity-based Sparse Mixture of Experts (SimSMoE), a novel similarity of neural network algorithm, that guarantees a solution to address the representation collapse issue between experts given a fixed FLOPs budget. We conduct extensive empirical evaluations on three large language models for both Pre-training and Fine-tuning tasks to illustrate the efficacy, robustness, and scalability of our method. The results demonstrate that SimSMoE significantly enhances existing routing policy and outperforms other SMoE training methods in performance for the tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2406_15883
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle SimSMoE: Solving Representational Collapse via Similarity Measure
Do, Giang
Le, Hung
Tran, Truyen
Computation and Language
Artificial Intelligence
Sparse mixture of experts (SMoE) have emerged as an effective approach for scaling large language models while keeping a constant computational cost. Regardless of several notable successes of SMoE, effective training such architecture remains elusive due to the representation collapse problem, which in turn harms model performance and causes parameter redundancy. In this work, we present Similarity-based Sparse Mixture of Experts (SimSMoE), a novel similarity of neural network algorithm, that guarantees a solution to address the representation collapse issue between experts given a fixed FLOPs budget. We conduct extensive empirical evaluations on three large language models for both Pre-training and Fine-tuning tasks to illustrate the efficacy, robustness, and scalability of our method. The results demonstrate that SimSMoE significantly enhances existing routing policy and outperforms other SMoE training methods in performance for the tasks.
title SimSMoE: Solving Representational Collapse via Similarity Measure
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2406.15883