SDQ: Sparse Decomposed Quantization for LLM Inference

TL;DR

SDQ introduces a novel compression technique combining structured sparsity and quantization, significantly improving LLM inference efficiency while maintaining high accuracy, thus enabling broader deployment of large models.

Contribution

The paper presents SDQ, a new method that leverages structured sparsity and quantization for efficient large language model inference, achieving high throughput with minimal quality loss.

Findings

4x effective compute throughput achieved

Less than 1% quality drop

Efficient deployment of large models

Abstract

Recently, large language models (LLMs) have shown surprising performance in task-specific workloads as well as general tasks with the given prompts. However, to achieve unprecedented performance, recent LLMs use billions to trillions of parameters, which hinder the wide adaptation of those models due to their extremely large compute and memory requirements. To resolve the issue, various model compression methods are being actively investigated. In this work, we propose SDQ (Sparse Decomposed Quantization) to exploit both structured sparsity and quantization to achieve both high compute and memory efficiency. From our evaluations, we observe that SDQ can achieve 4x effective compute throughput with <1% quality drop.