Accelerated Diffusion Models via Speculative Sampling

TL;DR

This paper extends speculative sampling to diffusion models, enabling faster sample generation by reducing function evaluations while maintaining exactness, with strategies that do not require additional training.

Contribution

It introduces a novel speculative sampling method for diffusion models, including a training-free drafting strategy that accelerates inference without sacrificing sample quality.

Findings

Halves the number of function evaluations needed for sampling

Achieves significant speedup in diffusion model inference

Generates exact samples from the target diffusion model

Abstract

Speculative sampling is a popular technique for accelerating inference in Large Language Models by generating candidate tokens using a fast draft model and accepting or rejecting them based on the target model's distribution. While speculative sampling was previously limited to discrete sequences, we extend it to diffusion models, which generate samples via continuous, vector-valued Markov chains. In this context, the target model is a high-quality but computationally expensive diffusion model. We propose various drafting strategies, including a simple and effective approach that does not require training a draft model and is applicable out of the box to any diffusion model. Our experiments demonstrate significant generation speedup on various diffusion models, halving the number of function evaluations, while generating exact samples from the target model.