Saved in:
Bibliographic Details
Main Authors: Gao, Shangqian, Hua, Ting, Shirkavand, Reza, Lin, Chi-Heng, Tang, Zheng, Li, Zhengao, Yuan, Longge, Li, Fangyi, Zhang, Zeyu, Ganjdanesh, Alireza, Qian, Lou, Jie, Xu, Hsu, Yen-Chang
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2501.15316
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918269657546752
author Gao, Shangqian
Hua, Ting
Shirkavand, Reza
Lin, Chi-Heng
Tang, Zheng
Li, Zhengao
Yuan, Longge
Li, Fangyi
Zhang, Zeyu
Ganjdanesh, Alireza
Qian, Lou
Jie, Xu
Hsu, Yen-Chang
author_facet Gao, Shangqian
Hua, Ting
Shirkavand, Reza
Lin, Chi-Heng
Tang, Zheng
Li, Zhengao
Yuan, Longge
Li, Fangyi
Zhang, Zeyu
Ganjdanesh, Alireza
Qian, Lou
Jie, Xu
Hsu, Yen-Chang
contents Large Language Models (LLMs) have demonstrated remarkable abilities in tackling a wide range of complex tasks. However, their huge computational and memory costs raise significant challenges in deploying these models on resource-constrained devices or efficiently serving them. Prior approaches have attempted to alleviate these problems by permanently removing less important model structures, yet these methods often result in substantial performance degradation due to the permanent deletion of model parameters. In this work, we tried to mitigate this issue by reducing the number of active parameters without permanently removing them. Specifically, we introduce a differentiable dynamic pruning method that pushes dense models to maintain a fixed number of active parameters by converting their MLP layers into a Mixture of Experts (MoE) architecture. Our method, even without fine-tuning, consistently outperforms previous structural pruning techniques across diverse model families, including Phi-2, LLaMA-2, LLaMA-3, and Qwen-2.5.
format Preprint
id arxiv_https___arxiv_org_abs_2501_15316
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ToMoE: Converting Dense Large Language Models to Mixture-of-Experts through Dynamic Structural Pruning
Gao, Shangqian
Hua, Ting
Shirkavand, Reza
Lin, Chi-Heng
Tang, Zheng
Li, Zhengao
Yuan, Longge
Li, Fangyi
Zhang, Zeyu
Ganjdanesh, Alireza
Qian, Lou
Jie, Xu
Hsu, Yen-Chang
Machine Learning
Computation and Language
Large Language Models (LLMs) have demonstrated remarkable abilities in tackling a wide range of complex tasks. However, their huge computational and memory costs raise significant challenges in deploying these models on resource-constrained devices or efficiently serving them. Prior approaches have attempted to alleviate these problems by permanently removing less important model structures, yet these methods often result in substantial performance degradation due to the permanent deletion of model parameters. In this work, we tried to mitigate this issue by reducing the number of active parameters without permanently removing them. Specifically, we introduce a differentiable dynamic pruning method that pushes dense models to maintain a fixed number of active parameters by converting their MLP layers into a Mixture of Experts (MoE) architecture. Our method, even without fine-tuning, consistently outperforms previous structural pruning techniques across diverse model families, including Phi-2, LLaMA-2, LLaMA-3, and Qwen-2.5.
title ToMoE: Converting Dense Large Language Models to Mixture-of-Experts through Dynamic Structural Pruning
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2501.15316