From 2:4 to 8:16 sparsity patterns in LLMs for Outliers and Weights with Variance Correction

TL;DR

This paper investigates 8:16 semi-structured sparsity in large language models, demonstrating its superior flexibility and performance in compression, especially for outliers, with techniques like variance correction enhancing results.

Contribution

It introduces 8:16 sparsity patterns that outperform 2:4 sparsity, and applies variance correction and weight equalization to improve sparse model performance.

Findings

8:16 sparsity surpasses performance thresholds at equivalent memory.

Structured sparsity for outliers is competitive with unstructured methods.

Variance correction and weight equalization improve sparse model accuracy.

Abstract

As large language models (LLMs) grow in size, efficient compression techniques like quantization and sparsification are critical. While quantization maintains performance with reduced precision, structured sparsity methods, such as N:M sparsification, often fall short due to limited flexibility, and sensitivity to outlier weights. We explore 8:16 semi-structured sparsity, demonstrating its ability to surpass the Performance Threshold-where a compressed model matches the accuracy of its uncompressed or smaller counterpart under equivalent memory constraints. Compared to 2:4 sparsity, 8:16 offers greater flexibility with minimal storage overhead (0.875 vs. 0.75 bits/element). We also apply sparse structured patterns for salient weights, showing that structured sparsity for outliers is competitive with unstructured approaches leading to equivalent or better results. Finally, we demonstrate that simple techniques such as variance correction and SmoothQuant like weight equalization improve sparse models performance.