# DFlash: Block Diffusion for Flash Speculative Decoding

**arXiv**: 2602.06036  
**Published**: 2026-02-05  
**Authors**: Jian Chen, Yesheng Liang, Zhijian Liu  
**GitHub**: https://github.com/z-lab/dflash  
**Stars**: 1953

## Abstract

DFlash: Block Diffusion for Flash Speculative Decoding

DFlash: Block Diffusion for Flash Speculative Decoding

Jian Chen

Yesheng Liang

Zhijian Liu

Abstract

Autoregressive large language models (LLMs) deliver strong performance but require inherently sequential decoding, leading to high inference latency and poor GPU utilization. Speculative decoding mitigates this bottleneck by using a fast draft model whose outputs are verified in parallel by the target LLM. However, existing methods still rely on

autoregressive drafting

, which remains sequential and constrains practical speedups.
Diffusion LLMs offer a promising alternative by enabling parallel generation, but current diffusion models typically underperform compared with autoregressive models.
In this paper, we introduce

DFlash

, a speculative decoding framework that employs a lightweight block diffusion model for parallel drafting. We show that speculative decoding provides a natural and effective setting for diffusion models. By generating draft tokens in a single forward pass, DFlash enables efficient drafting, and by conditioning the draft model on context features extracted from the target model, it achieves high-quality drafts with higher acceptance rates.
Experiments show that DFlash achieves over 6

×

\times

lossless acceleration across a range of models and tasks, delivering up to 2.5

×

\times

higher speedup than the state-of-the-art speculative decoding method EAGLE-3.

Links:

Code

(GitHub)

|

|

Models

(Hugging Face)

https://z-lab.ai/projects/dflash

1

Introduction

Large language models (LLMs) have enabled a wide range of powerful applications, including conversational agents

(Yang

et al.

,

2025

; Guo

et al.

,

2025

)

and automated programming tools. Despite their success, LLM inference remains dominated by a sequential, token-by-token generation process, where each output depends on the full preceding context. This inherent seriality creates a major performance bottleneck: inference is slow, memory-bound, and fails to fully utilize modern GPUs. With the recent emergence of long Chain-of-Thought (CoT) reasoning models

(OpenAI

et al.

,

2024

; Guo

et al.

,

2025

)

, this bottleneck has become increasingly critical, as prolonged inference times now dominate the generation process.

Speculative decoding

(Leviathan

et al.

,

2023

; Li

et al.

,

2025c

,

2024

,

b

; Cai

et al.

,

2024

)

has emerged as a primary solution to this bottleneck. This paradigm employs a lightweight

draft model

to speculate a sequence of future tokens, which are then verified in parallel by the large

target model

. While this approach achieves lossless acceleration and has been widely integrated into production frameworks, state-of-the-art methods like EAGLE-3

(Li

et al.

,

2025b

)

still rely on autoregressive drafting. This serial drafting process is not only inherently inefficient but also susceptible to error accumulation, which effectively caps achi

---

Full HTML content extracted from https://arxiv.org/html/2602.06036