DGQ: Distribution-Aware Group Quantization for Text-to-Image Diffusion Models

TL;DR

This paper introduces DGQ, a novel distribution-aware group quantization method that effectively reduces the computational and memory demands of text-to-image diffusion models while maintaining high image quality and text-image alignment at low bit-widths.

Contribution

DGQ adaptively handles outliers and applies prompt-specific logarithmic scales, enabling low-bit quantization of diffusion models without additional fine-tuning.

Findings

Achieves effective low-bit quantization on MS-COCO and PartiPrompts datasets.

Preserves image quality and text-image alignment at sub-8-bit levels.

First to quantize diffusion models without extra fine-tuning.

Abstract

Despite the widespread use of text-to-image diffusion models across various tasks, their computational and memory demands limit practical applications. To mitigate this issue, quantization of diffusion models has been explored. It reduces memory usage and computational costs by compressing weights and activations into lower-bit formats. However, existing methods often struggle to preserve both image quality and text-image alignment, particularly in lower-bit($<$ 8bits) quantization. In this paper, we analyze the challenges associated with quantizing text-to-image diffusion models from a distributional perspective. Our analysis reveals that activation outliers play a crucial role in determining image quality. Additionally, we identify distinctive patterns in cross-attention scores, which significantly affects text-image alignment. To address these challenges, we propose Distribution-aware Group Quantization (DGQ), a method that identifies and adaptively handles pixel-wise and channel-wise outliers to preserve image quality. Furthermore, DGQ applies prompt-specific logarithmic quantization scales to maintain text-image alignment. Our method demonstrates remarkable performance on datasets such as MS-COCO and PartiPrompts. We are the first to successfully achieve low-bit quantization of text-to-image diffusion models without requiring additional fine-tuning of weight quantization parameters. Code is available at https://github.com/ugonfor/DGQ.