MAR: Efficient Large Language Models via Module-aware Architecture Refinement

TL;DR

This paper introduces MAR, a framework that combines SSMs and activation sparsification to create energy-efficient large language models with maintained performance.

Contribution

MAR is a novel two-stage approach integrating SSMs and SNNs with new techniques to reduce energy costs while preserving LLM performance.

Findings

MAR restores dense model performance under resource constraints.

It significantly reduces inference energy consumption.

Outperforms comparable or larger efficient models.

Abstract

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.