---
id: 20261029-T0-15
title: "分布式LLM推理新方案：用通信高效的推测解码降低词表传输开销"
title_en: "Communication-Efficient Speculative Decoding Cuts Vocabulary-Scale Overhead in Distributed LLM Inference"
url: https://ai.daily.yangsir.net/daily/20261029-T0-15
issue_date: 2026-10-29
publish_date: 2026-10-28T04:00:00.000Z
category: research
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2610.35957
---

# 分布式LLM推理新方案：用通信高效的推测解码降低词表传输开销

arXiv 新论文研究设备与服务器协同的分布式 LLM 推理。这类方案的核心问题是词表规模的通信开销限制了推测解码的加速效果。论文提出通信高效的推测解码方法，目标是降低这一开销。对端侧与云端协同推理的场景来说，如果通信瓶颈被缓解，设备端推理的实用性和响应速度都有提升空间。论文为预印本，尚未经过同行评审。

## English Version

**Communication-Efficient Speculative Decoding Cuts Vocabulary-Scale Overhead in Distributed LLM Inference**

A new arXiv paper addresses distributed LLM inference where devices and servers collaborate. The core problem: vocabulary-scale communication overhead limits the speedup from speculative decoding. The authors propose a communication-efficient speculative decoding method aimed at reducing that overhead. For device-cloud collaborative inference, easing the communication bottleneck could improve both practicality and response speed on-device. The paper is a preprint and has not been peer-reviewed.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2610.35957)

**详情页**：https://ai.daily.yangsir.net/daily/20261029-T0-15

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*