AWP: Activation-Aware Weight Pruning and Quantization with Projected Gradient Descent

TL;DR

This paper introduces AWP, a unified activation-aware weight pruning and quantization method for large language models, leveraging projected gradient descent to improve compression efficiency and performance.

Contribution

It proposes a novel unified approach combining pruning and quantization with theoretical convergence guarantees, outperforming existing methods.

Findings

AWP outperforms state-of-the-art pruning and quantization techniques.

The method provides theoretical convergence guarantees.

Experimental results validate the effectiveness of AWP on LLMs.

Abstract

To address the enormous size of Large Language Models (LLMs), model compression methods, such as quantization and pruning, are often deployed, especially on edge devices. In this work, we focus on layer-wise post-training quantization and pruning. Drawing connections between activation-aware weight pruning and sparse approximation problems, and motivated by the success of Iterative Hard Thresholding (IHT), we propose a unified method for Activation-aware Weight pruning and quantization via Projected gradient descent (AWP). Our experiments demonstrate that AWP outperforms state-of-the-art LLM pruning and quantization methods. Theoretical convergence guarantees of the proposed method for pruning are also provided.