Saved in:
Bibliographic Details
Main Authors: Liu, Fuliang, Li, Xue, Zhao, Ketai, Gao, Yinxi, Zhou, Ziyan, Zhang, Zhonghui, Wang, Zhibin, Dou, Wanchun, Zhong, Sheng, Tian, Chen
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.19278
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917225093398528
author Liu, Fuliang
Li, Xue
Zhao, Ketai
Gao, Yinxi
Zhou, Ziyan
Zhang, Zhonghui
Wang, Zhibin
Dou, Wanchun
Zhong, Sheng
Tian, Chen
author_facet Liu, Fuliang
Li, Xue
Zhao, Ketai
Gao, Yinxi
Zhou, Ziyan
Zhang, Zhonghui
Wang, Zhibin
Dou, Wanchun
Zhong, Sheng
Tian, Chen
contents Speculative decoding is an effective and lossless approach for accelerating LLM inference. However, existing widely adopted model-based draft designs, such as EAGLE3, improve accuracy at the cost of multi-step autoregressive inference, resulting in high drafting latency and ultimately rendering the drafting stage itself a performance bottleneck. Inspired by diffusion-based large language models (dLLMs), we propose DART, which leverages parallel generation to reduce drafting latency. DART predicts logits for multiple future masked positions in parallel within a single forward pass based on hidden states of the target model, thereby eliminating autoregressive rollouts in the draft model while preserving a lightweight design. Based on these parallel logit predictions, we further introduce an efficient tree pruning algorithm that constructs high-quality draft token trees with N-gram-enforced semantic continuity. DART substantially reduces draft-stage overhead while preserving high draft accuracy, leading to significantly improved end-to-end decoding speed. Experimental results demonstrate that DART achieves a 2.03x--3.44x wall-clock time speedup across multiple datasets, surpassing EAGLE3 by 30% on average and offering a practical speculative decoding framework. Code is released at https://github.com/fvliang/DART.
format Preprint
id arxiv_https___arxiv_org_abs_2601_19278
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle DART: Diffusion-Inspired Speculative Decoding for Fast LLM Inference
Liu, Fuliang
Li, Xue
Zhao, Ketai
Gao, Yinxi
Zhou, Ziyan
Zhang, Zhonghui
Wang, Zhibin
Dou, Wanchun
Zhong, Sheng
Tian, Chen
Computation and Language
Speculative decoding is an effective and lossless approach for accelerating LLM inference. However, existing widely adopted model-based draft designs, such as EAGLE3, improve accuracy at the cost of multi-step autoregressive inference, resulting in high drafting latency and ultimately rendering the drafting stage itself a performance bottleneck. Inspired by diffusion-based large language models (dLLMs), we propose DART, which leverages parallel generation to reduce drafting latency. DART predicts logits for multiple future masked positions in parallel within a single forward pass based on hidden states of the target model, thereby eliminating autoregressive rollouts in the draft model while preserving a lightweight design. Based on these parallel logit predictions, we further introduce an efficient tree pruning algorithm that constructs high-quality draft token trees with N-gram-enforced semantic continuity. DART substantially reduces draft-stage overhead while preserving high draft accuracy, leading to significantly improved end-to-end decoding speed. Experimental results demonstrate that DART achieves a 2.03x--3.44x wall-clock time speedup across multiple datasets, surpassing EAGLE3 by 30% on average and offering a practical speculative decoding framework. Code is released at https://github.com/fvliang/DART.
title DART: Diffusion-Inspired Speculative Decoding for Fast LLM Inference
topic Computation and Language
url https://arxiv.org/abs/2601.19278