<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Speculative Decoding on 硅基 AGI · 智能体学习与测评</title><link>https://guijiagi.com/tags/speculative-decoding/</link><description>Recent content in Speculative Decoding on 硅基 AGI · 智能体学习与测评</description><generator>Hugo</generator><language>zh-cn</language><copyright>本站内容采用 CC BY-NC-SA 4.0 国际许可协议授权</copyright><lastBuildDate>Mon, 13 Jul 2026 04:40:00 +0800</lastBuildDate><atom:link href="https://guijiagi.com/tags/speculative-decoding/index.xml" rel="self" type="application/rss+xml"/><item><title>大模型推理的Speculative Decoding原理详解</title><link>https://guijiagi.com/posts/article-71/</link><pubDate>Mon, 13 Jul 2026 04:40:00 +0800</pubDate><guid>https://guijiagi.com/posts/article-71/</guid><description>深入解析Speculative Decoding的数学原理、工程实现和最新变体，包括投机采样、批量投机等前沿技术</description></item><item><title>Speculative Decoding实战：推理速度提升3倍</title><link>https://guijiagi.com/posts/speculative-decoding-practice/</link><pubDate>Tue, 30 Jun 2026 11:45:00 +0800</pubDate><guid>https://guijiagi.com/posts/speculative-decoding-practice/</guid><description>深入实践Speculative Decoding技术，实现大模型推理速度3倍提升</description></item><item><title>多Token预测：大模型推理加速的新范式</title><link>https://guijiagi.com/posts/multi-token-prediction-mtp-inference-acceleration/</link><pubDate>Tue, 30 Jun 2026 09:20:00 +0800</pubDate><guid>https://guijiagi.com/posts/multi-token-prediction-mtp-inference-acceleration/</guid><description>深入解析多Token预测(MTP)技术原理、训练方法和推理加速效果，对比分析主流实现方案</description></item><item><title>多Token预测：大模型推理加速的新范式</title><link>https://guijiagi.com/posts/multi-token-prediction/</link><pubDate>Tue, 30 Jun 2026 09:20:00 +0800</pubDate><guid>https://guijiagi.com/posts/multi-token-prediction/</guid><description>深入介绍多Token预测（MTP/Multi-Token Prediction）技术，如何通过将每次生成多个token来加速大模型推理</description></item><item><title>投机解码原理：Draft Model 如何加速推理</title><link>https://guijiagi.com/posts/speculative-decoding-principles/</link><pubDate>Sun, 28 Jun 2026 11:07:00 +0800</pubDate><guid>https://guijiagi.com/posts/speculative-decoding-principles/</guid><description>深入解析投机解码的算法原理、Draft Model 设计、接受率优化与2026年最新工程实践</description></item><item><title>Speculative Decoding 实战：推理速度提升 3 倍的配置指南</title><link>https://guijiagi.com/posts/speculative-decoding-practical-3x-speedup-guide/</link><pubDate>Sun, 28 Jun 2026 10:00:00 +0800</pubDate><guid>https://guijiagi.com/posts/speculative-decoding-practical-3x-speedup-guide/</guid><description>Speculative Decoding投机解码实战配置指南：原理详解、Draft模型选择与3倍加速实测</description></item></channel></rss>