<?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>RoPE on 硅基 AGI · 智能体学习与测评</title><link>https://guijiagi.com/tags/rope/</link><description>Recent content in RoPE 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>Thu, 16 Jul 2026 10:36:00 +0800</lastBuildDate><atom:link href="https://guijiagi.com/tags/rope/index.xml" rel="self" type="application/rss+xml"/><item><title>长上下文模型的技术挑战：从注意力衰减到有效利用</title><link>https://guijiagi.com/posts/b2-d60bc32e/</link><pubDate>Thu, 16 Jul 2026 10:36:00 +0800</pubDate><guid>https://guijiagi.com/posts/b2-d60bc32e/</guid><description>分析大模型处理超长上下文面临的技术挑战，涵盖注意力衰减、Needle-in-Haystack测试与长上下文优化策略</description></item><item><title>深度解析RoPE位置编码及其变体：从原理到演进</title><link>https://guijiagi.com/posts/article-80/</link><pubDate>Mon, 13 Jul 2026 06:10:00 +0800</pubDate><guid>https://guijiagi.com/posts/article-80/</guid><description>系统解析Rotary Position Embedding的数学原理、关键变体与工程实现，理解为什么它成为主流位置编码方案</description></item><item><title>大模型上下文窗口的极限与突破</title><link>https://guijiagi.com/posts/article-27/</link><pubDate>Sun, 12 Jul 2026 21:20:00 +0800</pubDate><guid>https://guijiagi.com/posts/article-27/</guid><description>从32K到10M+，上下文窗口扩展的技术路径、瓶颈分析与未来展望</description></item><item><title>大模型上下文窗口的极限与突破</title><link>https://guijiagi.com/posts/b2-188174a6/</link><pubDate>Sun, 12 Jul 2026 21:20:00 +0800</pubDate><guid>https://guijiagi.com/posts/b2-188174a6/</guid><description>从32K到10M+，上下文窗口扩展的技术路径、瓶颈分析与未来展望</description></item><item><title>RoPE旋转位置编码原理与改进</title><link>https://guijiagi.com/posts/rope-rotation-position-encoding/</link><pubDate>Thu, 02 Jul 2026 10:51:00 +0800</pubDate><guid>https://guijiagi.com/posts/rope-rotation-position-encoding/</guid><description>深入解析RoPE旋转位置编码的数学原理、长度外推方法及2026年最新改进</description></item><item><title>长上下文技术演进：从位置编码到环形注意力</title><link>https://guijiagi.com/posts/long-context-evolution-positional-encoding-to-ring-attention/</link><pubDate>Tue, 30 Jun 2026 09:20:00 +0800</pubDate><guid>https://guijiagi.com/posts/long-context-evolution-positional-encoding-to-ring-attention/</guid><description>全面梳理大模型长上下文技术的演进路线，涵盖位置编码外推、分块注意力、环形注意力等前沿方案</description></item><item><title>长上下文技术演进：从位置编码到环形注意力</title><link>https://guijiagi.com/posts/long-context-techniques-2026/</link><pubDate>Tue, 30 Jun 2026 09:20:00 +0800</pubDate><guid>https://guijiagi.com/posts/long-context-techniques-2026/</guid><description>系统梳理大模型长上下文处理技术的演进历程，从位置编码方案到环形注意力机制</description></item><item><title>位置编码深度对比：RoPE vs ALiBi vs NoPE 实测分析</title><link>https://guijiagi.com/posts/positional-encoding-comparison/</link><pubDate>Sun, 28 Jun 2026 11:02:00 +0800</pubDate><guid>https://guijiagi.com/posts/positional-encoding-comparison/</guid><description>深入对比RoPE旋转位置编码、ALiBi线性偏置、NoPE无位置编码三种方案的数学原理、长文本外推能力与实测性能</description></item><item><title>位置编码深度解析：从绝对位置到 RoPE 与 ALiBi</title><link>https://guijiagi.com/posts/positional-encoding/</link><pubDate>Thu, 25 Jun 2026 12:00:00 +0800</pubDate><guid>https://guijiagi.com/posts/positional-encoding/</guid><description>系统梳理 Transformer 位置编码的演进历程，深入分析 Sinusoidal、Learned、Relative、RoPE、ALiBi 等方案的数学原理与工程实现。</description></item><item><title>位置编码全解：绝对/相对/RoPE/ALiBi</title><link>https://guijiagi.com/posts/positional-encoding-guide/</link><pubDate>Thu, 25 Jun 2026 10:00:00 +0800</pubDate><guid>https://guijiagi.com/posts/positional-encoding-guide/</guid><description>系统解析 Transformer 位置编码的演进脉络与长度外推技术</description></item><item><title>旋转位置编码 RoPE：从原理到长度外推</title><link>https://guijiagi.com/posts/rope-rotation-embedding/</link><pubDate>Thu, 25 Jun 2026 10:00:00 +0800</pubDate><guid>https://guijiagi.com/posts/rope-rotation-embedding/</guid><description>深入解析 RoPE 数学原理及主流长度外推方法 NTK/YaRN/Longrope</description></item></channel></rss>