DuoGPT: Training-free Dual Sparsity through Activation-aware Pruning in LLMs

TL;DR

DuoGPT introduces a training-free dual sparsity framework for large language models that combines weight pruning with activation sparsity, improving efficiency while maintaining accuracy.

Contribution

It extends the OBC framework with activation-aware calibration and GPU-optimized solutions for scalable, dual-sparse LLM workloads.

Findings

Outperforms state-of-the-art pruning methods by up to 9.17% accuracy

Achieves 1.39× speedup over dense models

Scalable to billion-parameter LLMs

Abstract

Large language models (LLMs) deliver strong performance but are difficult to deploy due to high memory and compute costs. While pruning reduces these demands, most methods ignore activation sparsity observed at runtime. We reinterpret activation sparsity as dynamic structured weight sparsity and propose DuoGPT, a unified framework that constructs dual-sparse (spMspV) workloads by combining unstructured weight pruning with activation sparsity. To preserve accuracy, we extend the Optimal Brain Compression (OBC) framework with activation-aware calibration and introduce output residuals from the dense model as correction terms. We further optimize the solution for efficient GPU execution, enabling scalability to billion-parameter LLMs. Evaluations on LLaMA-2 and LLaMA-3 show that DuoGPT outperforms state-of-the-art structured pruning methods by up to 9.17% accuracy at an iso-speedup of 1.39$\times$ compared to the baseline dense model. Code is available at Github.