Q-Diffusion: Quantizing Diffusion Models

TL;DR

Q-Diffusion introduces a novel post-training quantization method tailored for diffusion models, enabling 4-bit compression with minimal performance loss and faster inference, thus enhancing efficiency for image synthesis tasks.

Contribution

The paper presents a new PTQ technique specifically designed for diffusion models, addressing their unique multi-timestep architecture and activation distributions.

Findings

Quantizes diffusion models to 4-bit with minimal FID increase

Maintains high generation quality in text-guided image synthesis

Achieves training-free quantization with significant efficiency gains

Abstract

Diffusion models have achieved great success in image synthesis through iterative noise estimation using deep neural networks. However, the slow inference, high memory consumption, and computation intensity of the noise estimation model hinder the efficient adoption of diffusion models. Although post-training quantization (PTQ) is considered a go-to compression method for other tasks, it does not work out-of-the-box on diffusion models. We propose a novel PTQ method specifically tailored towards the unique multi-timestep pipeline and model architecture of the diffusion models, which compresses the noise estimation network to accelerate the generation process. We identify the key difficulty of diffusion model quantization as the changing output distributions of noise estimation networks over multiple time steps and the bimodal activation distribution of the shortcut layers within the noise estimation network. We tackle these challenges with timestep-aware calibration and split shortcut quantization in this work. Experimental results show that our proposed method is able to quantize full-precision unconditional diffusion models into 4-bit while maintaining comparable performance (small FID change of at most 2.34 compared to >100 for traditional PTQ) in a training-free manner. Our approach can also be applied to text-guided image generation, where we can run stable diffusion in 4-bit weights with high generation quality for the first time.