Saved in:
Bibliographic Details
Main Authors: Cai, Junhong, Wang, Guiqin, Zhao, Kejie, Tang, Jianxiong, Wang, Xiang, Leng, Luziwei, Cheng, Ran, Ma, Yuxin, Guo, Qinghai
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.21503
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913052933226496
author Cai, Junhong
Wang, Guiqin
Zhao, Kejie
Tang, Jianxiong
Wang, Xiang
Leng, Luziwei
Cheng, Ran
Ma, Yuxin
Guo, Qinghai
author_facet Cai, Junhong
Wang, Guiqin
Zhao, Kejie
Tang, Jianxiong
Wang, Xiang
Leng, Luziwei
Cheng, Ran
Ma, Yuxin
Guo, Qinghai
contents Large Language Models (LLMs) excel across diverse domains but suffer from high energy costs due to quadratic attention and dense Feed-Forward Network (FFN) operations. To address these issues, we propose Module-aware Architecture Refinement (MAR), a two-stage framework that integrates State Space Models (SSMs) for linear-time sequence modeling and applies activation sparsification to reduce FFN costs. In addition, to mitigate low information density and temporal mismatch in integrating Spiking Neural Networks (SNNs) with SSMs, we design the Adaptive Ternary Multi-step Neuron (ATMN) and the Spike-aware Bidirectional Distillation Strategy (SBDS). Extensive experiments demonstrate that MAR effectively restores the performance of its dense counterpart under constrained resources while substantially reducing inference energy consumption. Furthermore, it outperforms efficient models of comparable or even larger scale, underscoring its potential for building efficient and practical LLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2601_21503
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle MAR: Efficient Large Language Models via Module-aware Architecture Refinement
Cai, Junhong
Wang, Guiqin
Zhao, Kejie
Tang, Jianxiong
Wang, Xiang
Leng, Luziwei
Cheng, Ran
Ma, Yuxin
Guo, Qinghai
Artificial Intelligence
Computation and Language
Machine Learning
Neural and Evolutionary Computing
Large Language Models (LLMs) excel across diverse domains but suffer from high energy costs due to quadratic attention and dense Feed-Forward Network (FFN) operations. To address these issues, we propose Module-aware Architecture Refinement (MAR), a two-stage framework that integrates State Space Models (SSMs) for linear-time sequence modeling and applies activation sparsification to reduce FFN costs. In addition, to mitigate low information density and temporal mismatch in integrating Spiking Neural Networks (SNNs) with SSMs, we design the Adaptive Ternary Multi-step Neuron (ATMN) and the Spike-aware Bidirectional Distillation Strategy (SBDS). Extensive experiments demonstrate that MAR effectively restores the performance of its dense counterpart under constrained resources while substantially reducing inference energy consumption. Furthermore, it outperforms efficient models of comparable or even larger scale, underscoring its potential for building efficient and practical LLMs.
title MAR: Efficient Large Language Models via Module-aware Architecture Refinement
topic Artificial Intelligence
Computation and Language
Machine Learning
Neural and Evolutionary Computing
url https://arxiv.org/abs/2601.21503