DLP: Dynamic Layerwise Pruning in Large Language Models

2025-05-27

AI Summary

To address the sharp performance degradation of large language models (LLMs) under high sparsity caused by fixed-layer pruning ratios, this paper proposes a dynamic inter-layer pruning method. Our approach introduces a novel weight-activation co-aware importance scoring mechanism that eliminates predefined sparsity constraints; instead, it adaptively allocates layer-wise pruning ratios through joint modeling of per-layer weight distributions and input activation statistics. The framework is lightweight, fully compatible with parameter-efficient fine-tuning (PEFT), and seamlessly integrates with mainstream compression techniques. Under 70% overall sparsity, our method reduces perplexity by 7.79 and improves average accuracy by 2.7% on LLaMA2-7B—outperforming existing state-of-the-art pruning approaches significantly.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Computer Vision: Large Vision Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Pruning has recently been widely adopted to reduce the parameter scale and improve the inference efficiency of Large Language Models (LLMs). Mainstream pruning techniques often rely on uniform layerwise pruning strategies, which can lead to severe performance degradation at high sparsity levels. Recognizing the varying contributions of different layers in LLMs, recent studies have shifted their focus toward non-uniform layerwise pruning. However, these approaches often rely on pre-defined values, which can result in suboptimal performance. To overcome these limitations, we propose a novel method called Dynamic Layerwise Pruning (DLP). This approach adaptively determines the relative importance of each layer by integrating model weights with input activation information, assigning pruning rates accordingly. Experimental results show that DLP effectively preserves model performance at high sparsity levels across multiple LLMs. Specifically, at 70% sparsity, DLP reduces the perplexity of LLaMA2-7B by 7.79 and improves the average accuracy by 2.7% compared to state-of-the-art methods. Moreover, DLP is compatible with various existing LLM compression techniques and can be seamlessly integrated into Parameter-Efficient Fine-Tuning (PEFT). We release the code at https://github.com/ironartisan/DLP to facilitate future research.

Problem

• Dynamic pruning for efficient Large Language Models
• Adaptive layer importance for optimal sparsity performance
• Compatibility with existing compression and fine-tuning techniques

Innovation

• Dynamic layerwise pruning adapts to layer importance
• Integrates model weights with input activation data
• Compatible with existing LLM compression techniques