---
title: "DARTree：利用自回归草稿树进行扩散投机解码"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.13524v1"
permalink: "https://feed.astralor.com/items/itm_3928289022811be714dc61a4"
published: "2026-08-13T17:43:44.000Z"
score: 77
channel: "ai"
tags: ["DARTree","扩散模型","投机解码","推理加速"]
---

# DARTree：利用自回归草稿树进行扩散投机解码

## AI 点评

> 投机解码是大模型推理加速的关键路径，DARTree巧妙地把扩散草稿模型的并行性和自回归校正头的序列修正结合成树结构，且无需重新训练。对做LLM推理优化和部署的工程师来说，这是值得关注的一篇论文；对研究者来说，它展示了树形剪枝与批量评分的工程技巧。

## 速览

投机解码通过并行验证多个草稿token来无损加速自回归语言模型。扩散式草稿模型可并行预测整个token块，但其逐位置分布是边际分布，未沿草稿路径条件化。DARTree是一种训练免费方法，将预训练的自回归校正头从链扩展到树，批量构建固定宽度候选树，并通过最佳优先剪枝选择验证树。在7个数学、代码和聊天基准上，DARTree在四种模型-温度配置中均取得最高平均接受长度和加速，单轮最多接受12.97个token，较DFlash提升98.6%，较Domino提升27.9%，最高无损加速达9.73倍。

## 关键点

- DARTree是首个将预训练自回归校正头从链扩展到树的训练免费投机解码方法。
- 通过批量扩展和最佳优先剪枝，解耦AR头推理与顺序堆操作，提升效率。
- 在7个基准、4种模型-温度配置中均实现最高平均接受长度和加速。
- 单轮最多接受12.97个token，比DFlash高98.6%，比Domino高27.9%，最高无损加速9.73倍。

## 全文

Speculative decoding losslessly accelerates autoregressive language models by verifying multiple draft tokens in parallel. Diffusion-based drafters further reduce proposal latency by predicting an entire token block in parallel, but their position-wise distributions are marginal rather than conditioned on tokens selected along each draft path. Existing recurrent correction incorporates causal information along a single draft chain, whereas diffusion-based tree construction broadens candidate coverage without carrying this correction along individual branches. We introduce DARTree, a training-free speculative decoding method that extends a pretrained AR correction head from chains to trees. DARTree first constructs a fixed-width candidate tree by expanding and scoring all nodes at each depth in a single batch, and then only applies best-first pruning to select the verification tree, decoupling AR-head inference from sequential heap operations. Across seven math, code, and chat benchmarks, DARTree achieves the highest average acceptance length and speedup in all four model--temperature configurations, accepting up to 12.97 tokens per verification round, 98.6\% more than DFlash and 27.9\% more than Domino in the same setting, and reaching up to 9.73$\times$ lossless speedup over locally measured autoregressive decoding.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_3928289022811be714dc61a4