ROSAQ: Rotation-based Saliency-Aware Weight Quantization for Efficiently Compressing Large Language Models

TL;DR

ROSAQ introduces a rotation-based, saliency-aware weight quantization method for large language models, leveraging PCA and mixed-precision to improve compression efficiency and inference speed.

Contribution

It proposes a novel PCA-based rotation and saliency detection approach for weight quantization, enhancing model compression and inference speed.

Findings

Outperforms baseline saliency-aware quantization methods.

Achieves 2.3x speed-up in token generation.

Improves model compression with minimal accuracy loss.

Abstract

Quantization has been widely studied as an effective technique for reducing the memory requirement of large language models (LLMs), potentially improving the latency time as well. Utilizing the characteristic of rotational invariance of transformer, we propose the rotation-based saliency-aware weight quantization (ROSAQ), which identifies salient channels in the projection feature space, not in the original feature space, where the projected "principal" dimensions are naturally considered as "salient" features. The proposed ROSAQ consists of 1) PCA-based projection, which first performs principal component analysis (PCA) on a calibration set and transforms via the PCA projection, 2) Salient channel dentification, which selects dimensions corresponding to the K-largest eigenvalues as salient channels, and 3) Saliency-aware quantization with mixed-precision, which uses FP16 for salient dimensions and INT3/4 for other dimensions. Experiment results show that ROSAQ shows improvements over the baseline saliency-aware quantization on the original feature space and other existing quantization methods. With kernel fusion, ROSAQ presents about 2.3x speed up over FP16 implementation in generating 256 tokens with a batch size of 64.