Accelerating Diffusion Large Language Models with SlowFast Sampling: The Three Golden Principles

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wei, Qingyan, Zhang, Yaojie, Liu, Zhiyuan, Zeng, Puyu, Wang, Yuxuan, Qi, Biqing, Liu, Dongrui, Zhang, Linfeng
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917370521452544
author Wei, Qingyan
Zhang, Yaojie
Liu, Zhiyuan
Zeng, Puyu
Wang, Yuxuan
Qi, Biqing
Liu, Dongrui
Zhang, Linfeng
author_facet Wei, Qingyan
Zhang, Yaojie
Liu, Zhiyuan
Zeng, Puyu
Wang, Yuxuan
Qi, Biqing
Liu, Dongrui
Zhang, Linfeng
contents Diffusion-based language models (dLLMs) have emerged as a promising alternative to traditional autoregressive LLMs by enabling parallel token generation and significantly reducing inference latency. However, existing sampling strategies for dLLMs, such as confidence-based or semi-autoregressive decoding, often suffer from static behavior, leading to suboptimal efficiency and limited flexibility. In this paper, we propose SlowFast Sampling, a novel dynamic sampling strategy that adaptively alternates between exploratory and accelerated decoding stages. Our method is guided by three golden principles: certainty principle, convergence principle, and positional principle, which govern when and where tokens can be confidently and efficiently decoded. We further integrate our strategy with dLLM-Cache to reduce redundant computation. Extensive experiments across benchmarks and models show that SlowFast Sampling achieves up to 15.63$\times$ speedup on LLaDA with minimal accuracy drop, and up to 34.22$\times$ when combined with caching. Notably, our approach outperforms strong autoregressive baselines like LLaMA3 8B in throughput, demonstrating that well-designed sampling can unlock the full potential of dLLMs for fast and high-quality generation.
format Preprint
id arxiv_https___arxiv_org_abs_2506_10848
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Accelerating Diffusion Large Language Models with SlowFast Sampling: The Three Golden Principles
Wei, Qingyan
Zhang, Yaojie
Liu, Zhiyuan
Zeng, Puyu
Wang, Yuxuan
Qi, Biqing
Liu, Dongrui
Zhang, Linfeng
Computation and Language
Artificial Intelligence
Machine Learning
Diffusion-based language models (dLLMs) have emerged as a promising alternative to traditional autoregressive LLMs by enabling parallel token generation and significantly reducing inference latency. However, existing sampling strategies for dLLMs, such as confidence-based or semi-autoregressive decoding, often suffer from static behavior, leading to suboptimal efficiency and limited flexibility. In this paper, we propose SlowFast Sampling, a novel dynamic sampling strategy that adaptively alternates between exploratory and accelerated decoding stages. Our method is guided by three golden principles: certainty principle, convergence principle, and positional principle, which govern when and where tokens can be confidently and efficiently decoded. We further integrate our strategy with dLLM-Cache to reduce redundant computation. Extensive experiments across benchmarks and models show that SlowFast Sampling achieves up to 15.63$\times$ speedup on LLaDA with minimal accuracy drop, and up to 34.22$\times$ when combined with caching. Notably, our approach outperforms strong autoregressive baselines like LLaMA3 8B in throughput, demonstrating that well-designed sampling can unlock the full potential of dLLMs for fast and high-quality generation.
title Accelerating Diffusion Large Language Models with SlowFast Sampling: The Three Golden Principles
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2506.10848