Morse: Dual-Sampling for Lossless Acceleration of Diffusion Models

TL;DR

Morse introduces a dual-sampling framework that accelerates diffusion models losslessly by interleaving fast jump sampling with residual feedback, achieving up to 3.31X speedup without loss of quality.

Contribution

The paper proposes Morse, a novel dual-model approach with weight sharing that significantly speeds up diffusion model sampling losslessly, applicable to various image generation tasks.

Findings

Achieves 1.78X to 3.31X speedup over baseline models.

Maintains image quality while reducing sampling steps.

Generalizes to improve existing accelerated models like LCM-SDXL.

Abstract

In this paper, we present Morse, a simple dual-sampling framework for accelerating diffusion models losslessly. The key insight of Morse is to reformulate the iterative generation (from noise to data) process via taking advantage of fast jump sampling and adaptive residual feedback strategies. Specifically, Morse involves two models called Dash and Dot that interact with each other. The Dash model is just the pre-trained diffusion model of any type, but operates in a jump sampling regime, creating sufficient space for sampling efficiency improvement. The Dot model is significantly faster than the Dash model, which is learnt to generate residual feedback conditioned on the observations at the current jump sampling point on the trajectory of the Dash model, lifting the noise estimate to easily match the next-step estimate of the Dash model without jump sampling. By chaining the outputs of the Dash and Dot models run in a time-interleaved fashion, Morse exhibits the merit of flexibly attaining desired image generation performance while improving overall runtime efficiency. With our proposed weight sharing strategy between the Dash and Dot models, Morse is efficient for training and inference. Our method shows a lossless speedup of 1.78X to 3.31X on average over a wide range of sampling step budgets relative to 9 baseline diffusion models on 6 image generation tasks. Furthermore, we show that our method can be also generalized to improve the Latent Consistency Model (LCM-SDXL, which is already accelerated with consistency distillation technique) tailored for few-step text-to-image synthesis. The code and models are available at https://github.com/deep-optimization/Morse.