ReSpinQuant: Efficient Layer-Wise LLM Quantization via Subspace Residual Rotation Approximation

TL;DR

ReSpinQuant introduces an efficient layer-wise quantization method for LLMs that combines high accuracy with minimal inference overhead by using offline activation rotation fusion and residual subspace rotation.

Contribution

It proposes a novel quantization framework that achieves layer-wise adaptation with negligible overhead, outperforming global rotation methods and matching expensive layer-wise approaches.

Findings

ReSpinQuant outperforms global rotation methods in accuracy.

ReSpinQuant matches the accuracy of layer-wise methods.

ReSpinQuant incurs only negligible inference overhead.

Abstract

Rotation-based Post-Training Quantization (PTQ) has emerged as a promising solution for mitigating activation outliers in the quantization of Large Language Models (LLMs). Global rotation methods achieve inference efficiency by fusing activation rotations into attention and FFN blocks, but suffer from limited expressivity as they are constrained to use a single learnable rotation matrix across all layers. To tackle this, layer-wise transformation methods emerged, achieving superior accuracy through localized adaptation. However, layer-wise methods cannot fuse activation rotation matrices into weights, requiring online computations and causing significant overhead. In this paper, we propose ReSpinQuant, a quantization framework that resolves such overhead by leveraging offline activation rotation fusion and matching basis using efficient residual subspace rotation. This design reconciles the high expressivity of layer-wise adaptation with only negligible inference overhead. Extensive experiments on W4A4 and W3A3 quantization demonstrate that ReSpinQuant achieves state-of-the-art performance, outperforming global rotation methods and matching the accuracy of computationally expensive layer-wise methods with minimal overhead.