<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/">
    <channel>
        <title>jinxiaokuang</title>
        <link>https://www.jinxiaokuang.top/</link>
        <description>Hello World</description>
        <lastBuildDate>Tue, 06 May 2025 03:23:39 GMT</lastBuildDate>
        <docs>https://validator.w3.org/feed/docs/rss2.html</docs>
        <generator>https://github.com/jpmonette/feed</generator>
        <language>en-US</language>
        <copyright>All rights reserved 2025, jinxiaokuang</copyright>
        <item>
            <title><![CDATA[后端开发学习笔记]]></title>
            <link>https://www.jinxiaokuang.top/notes/backend-development-1</link>
            <guid>https://www.jinxiaokuang.top/notes/backend-development-1</guid>
            <pubDate>Fri, 14 Feb 2025 00:00:00 GMT</pubDate>
            <description><![CDATA[记录Java后端开发学习经历]]></description>
            <content:encoded><![CDATA[<div id="notion-article" class="mx-auto overflow-hidden "><main class="notion light-mode notion-page notion-block-1be0bdda378280c08874d42e998f82eb"><div class="notion-viewport"></div><div class="notion-collection-page-properties"></div><div></div><div class="notion-callout notion-gray_background_co notion-block-1be0bdda378280deb918ebb5f7d01829"><div class="notion-page-icon-inline notion-page-icon-span"><span class="notion-page-icon" role="img" aria-label="😀">😀</span></div><div class="notion-callout-text">这里存放后端开发学习笔记</div></div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-1be0bdda3782807f8cd9e7452a172c35" data-id="1be0bdda3782807f8cd9e7452a172c35"><span><div id="1be0bdda3782807f8cd9e7452a172c35" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1be0bdda3782807f8cd9e7452a172c35" title="📘 笔记汇总"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">📘 笔记汇总</span></span></h2><a class="notion-page-link notion-block-1be0bdda378280be916fc1ec9b47a263" href="/1be0bdda378280be916fc1ec9b47a263"><span class="notion-page-title"><div class="notion-page-icon-inline notion-page-icon-span"><span class="notion-page-title-icon notion-page-icon" role="img" aria-label="📘">📘</span></div><span class="notion-page-title-text">阶段1 编程语言基础</span></span></a><a class="notion-page-link notion-block-1be0bdda378280bb8238d376e1609369" href="/1be0bdda378280bb8238d376e1609369"><span class="notion-page-title"><div class="notion-page-icon-inline notion-page-icon-span"><span class="notion-page-title-icon notion-page-icon" role="img" aria-label="📘">📘</span></div><span class="notion-page-title-text">阶段2 计算机体系基础知识</span></span></a><a class="notion-page-link notion-block-1be0bdda3782807c9834f8dad8d4a6ba" href="/1be0bdda3782807c9834f8dad8d4a6ba"><span class="notion-page-title"><div class="notion-page-icon-inline notion-page-icon-span"><span class="notion-page-title-icon notion-page-icon" role="img" aria-label="📘">📘</span></div><span class="notion-page-title-text">阶段3 企业开发基础</span></span></a><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-1be0bdda37828002870ef7aac15ed389" data-id="1be0bdda37828002870ef7aac15ed389"><span><div id="1be0bdda37828002870ef7aac15ed389" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1be0bdda37828002870ef7aac15ed389" title="📎 参考资源"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">📎 参考资源</span></span></h2><ul class="notion-list notion-list-disc notion-block-1be0bdda378280269900dd8120f21ef7"><li><a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://www.codefather.cn/course/1789189862986850306/section/1789190431398928386?type=#heading-74">Java 学习路线 | 25 年最新零基 - 📚2025 年最新编程学习路线(汇总) - 编程导航教程</a></li></ul><div class="notion-blank notion-block-1be0bdda378280899ef6f33c800bc36e"> </div><div class="notion-blank notion-block-1be0bdda37828090b9e2dd14db1591b3"> </div><div class="notion-blank notion-block-1be0bdda378280bcac6ad0a7e8833082"> </div><div class="notion-blank notion-block-1be0bdda378280348e27fb88c39fb0e6"> </div><div class="notion-blank notion-block-1be0bdda3782809dbac0c5af4d903ab5"> </div><div class="notion-blank notion-block-1be0bdda3782800ba353dc4df1ba19cc"> </div><div class="notion-blank notion-block-1be0bdda3782809c8537d85c9dfc752c"> </div><div class="notion-callout notion-gray_background_co notion-block-1be0bdda378280e1a7f6f101ad1fa6e6"><div class="notion-page-icon-inline notion-page-icon-span"><span class="notion-page-icon" role="img" aria-label="💡">💡</span></div><div class="notion-callout-text">欢迎您在底部评论区留言，一起交流~</div></div><div class="notion-blank notion-block-1be0bdda37828003b47bc79f3fa2d0ec"> </div></main></div>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[LeetCode算法笔记]]></title>
            <link>https://www.jinxiaokuang.top/notes/leetcode-notes-1</link>
            <guid>https://www.jinxiaokuang.top/notes/leetcode-notes-1</guid>
            <pubDate>Sat, 16 Nov 2024 00:00:00 GMT</pubDate>
            <description><![CDATA[记录Leetcode算法刷题经历]]></description>
            <content:encoded><![CDATA[<div id="notion-article" class="mx-auto overflow-hidden "><main class="notion light-mode notion-page notion-block-1be0bdda3782808bbcd9cd02beffd4ef"><div class="notion-viewport"></div><div class="notion-collection-page-properties"></div><div></div><div class="notion-callout notion-gray_background_co notion-block-1be0bdda3782807e8ca2e893617b837d"><div class="notion-page-icon-inline notion-page-icon-span"><span class="notion-page-icon" role="img" aria-label="😀">😀</span></div><div class="notion-callout-text">这里存放Leetcode刷题笔记</div></div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-1be0bdda3782807cac48c478934bb3e3" data-id="1be0bdda3782807cac48c478934bb3e3"><span><div id="1be0bdda3782807cac48c478934bb3e3" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1be0bdda3782807cac48c478934bb3e3" title="📘 学习笔记"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">📘 学习笔记</span></span></h2><h4 class="notion-h notion-h3 notion-h-indent-1 notion-block-1be0bdda378280bebd6ecf1cbd0f06f4" data-id="1be0bdda378280bebd6ecf1cbd0f06f4"><span><div id="1be0bdda378280bebd6ecf1cbd0f06f4" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1be0bdda378280bebd6ecf1cbd0f06f4" title="代码随想录"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">代码随想录</span></span></h4><a class="notion-page-link notion-block-1be0bdda37828069b806c72bd29dc696" href="/1be0bdda37828069b806c72bd29dc696"><span class="notion-page-title"><div class="notion-page-icon-inline notion-page-icon-span"><span class="notion-page-title-icon notion-page-icon" role="img" aria-label="📘">📘</span></div><span class="notion-page-title-text">数组</span></span></a><a class="notion-page-link notion-block-1be0bdda378280bfba50c97020876764" href="/1be0bdda378280bfba50c97020876764"><span class="notion-page-title"><div class="notion-page-icon-inline notion-page-icon-span"><span class="notion-page-title-icon notion-page-icon" role="img" aria-label="📘">📘</span></div><span class="notion-page-title-text">链表</span></span></a><a class="notion-page-link notion-block-1be0bdda3782804396d5f5c52b183d7c" href="/1be0bdda3782804396d5f5c52b183d7c"><span class="notion-page-title"><div class="notion-page-icon-inline notion-page-icon-span"><span class="notion-page-title-icon notion-page-icon" role="img" aria-label="📘">📘</span></div><span class="notion-page-title-text">字符串</span></span></a><div class="notion-blank notion-block-1be0bdda378280c4860dc55077bc2d45"> </div><h4 class="notion-h notion-h3 notion-h-indent-1 notion-block-1be0bdda37828048847ae3da0763cb04" data-id="1be0bdda37828048847ae3da0763cb04"><span><div id="1be0bdda37828048847ae3da0763cb04" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1be0bdda37828048847ae3da0763cb04" title="Hot100"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">Hot100</span></span></h4><div class="notion-to-do notion-block-1be0bdda37828076b4e0d1d6dd700f7b"><div class="notion-to-do-item"><span class="notion-property notion-property-checkbox"><div class="notion-property-checkbox-unchecked"></div></span><div class="notion-to-do-body">TODO</div></div><div class="notion-to-do-children"></div></div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-1be0bdda37828088af5ccf217a8cc336" data-id="1be0bdda37828088af5ccf217a8cc336"><span><div id="1be0bdda37828088af5ccf217a8cc336" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1be0bdda37828088af5ccf217a8cc336" title="📎 参考资源"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">📎 参考资源</span></span></h2><ul class="notion-list notion-list-disc notion-block-1be0bdda37828089b166dc2bd3c477eb"><li>LeetCode Hot100：<a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://leetcode.cn/studyplan/top-100-liked/">https://leetcode.cn/studyplan/top-100-liked/</a></li></ul><ul class="notion-list notion-list-disc notion-block-1be0bdda37828047a0e3da826c3f0646"><li>‣<!-- -->：<a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://www.programmercarl.com/">https://www.programmercarl.com/</a></li></ul><ul class="notion-list notion-list-disc notion-block-1be0bdda378280dcb66dfc6b31cb9561"><li>算法可视化：<a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://labuladong.online/algo/">labuladong 的算法笔记 | labuladong 的算法笔记</a></li></ul><ul class="notion-list notion-list-disc notion-block-1be0bdda378280bc92acfab57d6140ff"><li>LeetCode题解：<a target="_blank" rel="noopener noreferrer" class="notion-link" href="http://www.cyc2018.xyz/%E7%AE%97%E6%B3%95/Leetcode%20%E9%A2%98%E8%A7%A3/Leetcode%20%E9%A2%98%E8%A7%A3%20-%20%E7%9B%AE%E5%BD%95.html">Leetcode 题解 | CS-Notes 面试笔记</a></li></ul><hr class="notion-hr notion-block-1be0bdda378280488fa2cfb4b9e9cef7"/><div class="notion-blank notion-block-1be0bdda378280579bc3de3321dcb100"> </div><div class="notion-callout notion-gray_background_co notion-block-1be0bdda37828054b5adde46f0560f35"><div class="notion-page-icon-inline notion-page-icon-span"><span class="notion-page-icon" role="img" aria-label="💡">💡</span></div><div class="notion-callout-text">欢迎您在底部评论区留言，一起交流~</div></div><div class="notion-blank notion-block-1be0bdda378280f3a390f85837054d4c"> </div></main></div>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[示例文章]]></title>
            <link>https://www.jinxiaokuang.top/essay/example-1</link>
            <guid>https://www.jinxiaokuang.top/essay/example-1</guid>
            <pubDate>Mon, 20 Jan 2025 00:00:00 GMT</pubDate>
            <description><![CDATA[NotionNext主题的文章模板]]></description>
            <content:encoded><![CDATA[NotionNext主题的文章模板]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[多模态学习链接]]></title>
            <link>https://www.jinxiaokuang.top/technology/multimodal-learning-11</link>
            <guid>https://www.jinxiaokuang.top/technology/multimodal-learning-11</guid>
            <pubDate>Tue, 25 Feb 2025 00:00:00 GMT</pubDate>
            <description><![CDATA[介绍一些多模态的学习资源链接]]></description>
            <content:encoded><![CDATA[<div id="notion-article" class="mx-auto overflow-hidden "><main class="notion light-mode notion-page notion-block-1a70bdda378281a2bfc4fe4ec3fc59aa"><div class="notion-viewport"></div><div class="notion-collection-page-properties"></div><div></div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-1a70bdda37828182b4ced8f680120b2f" data-id="1a70bdda37828182b4ced8f680120b2f"><span><div id="1a70bdda37828182b4ced8f680120b2f" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda37828182b4ced8f680120b2f" title="Awesome Multimodal😯"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title"><b>Awesome Multimodal😯</b></span></span></h2><h3 class="notion-h notion-h2 notion-h-indent-1 notion-block-1a70bdda37828188b271f615ab6df1f9" data-id="1a70bdda37828188b271f615ab6df1f9"><span><div id="1a70bdda37828188b271f615ab6df1f9" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda37828188b271f615ab6df1f9" title="Paper"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">Paper</span></span></h3><div class="notion-row"><a target="_blank" rel="noopener noreferrer" class="notion-bookmark notion-block-1a70bdda378280279aeefb5025a7b183" href="https://github.com/friedrichor/Awesome-Multimodal-Papers/tree/main"><div><div class="notion-bookmark-title">GitHub - friedrichor/Awesome-Multimodal-Papers: A curated list of awesome Multimodal studies.</div><div class="notion-bookmark-description">A curated list of awesome Multimodal studies. Contribute to friedrichor/Awesome-Multimodal-Papers development by creating an account on GitHub.</div><div class="notion-bookmark-link"><div class="notion-bookmark-link-icon"><img src="https://www.notion.so/image/https%3A%2F%2Fgithub.com%2Ffluidicon.png?table=block&amp;id=1a70bdda-3782-8027-9aee-fb5025a7b183&amp;t=1a70bdda-3782-8027-9aee-fb5025a7b183" alt="GitHub - friedrichor/Awesome-Multimodal-Papers: A curated list of awesome Multimodal studies." loading="lazy" decoding="async"/></div><div class="notion-bookmark-link-text">https://github.com/friedrichor/Awesome-Multimodal-Papers/tree/main</div></div></div><div class="notion-bookmark-image"><img style="object-fit:cover" src="https://www.notion.so/image/https%3A%2F%2Fopengraph.githubassets.com%2F796e076c988690654069f90b9d5e095cb351ea22130138da45238ed05cc80694%2Ffriedrichor%2FAwesome-Multimodal-Papers?table=block&amp;id=1a70bdda-3782-8027-9aee-fb5025a7b183&amp;t=1a70bdda-3782-8027-9aee-fb5025a7b183" alt="GitHub - friedrichor/Awesome-Multimodal-Papers: A curated list of awesome Multimodal studies." loading="lazy" decoding="async"/></div></a></div><h3 class="notion-h notion-h2 notion-h-indent-1 notion-block-1a70bdda378281a189e6ef9b0f2b415e" data-id="1a70bdda378281a189e6ef9b0f2b415e"><span><div id="1a70bdda378281a189e6ef9b0f2b415e" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda378281a189e6ef9b0f2b415e" title="Model &amp; Code"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">Model &amp; Code</span></span></h3><div class="notion-row"><a target="_blank" rel="noopener noreferrer" class="notion-bookmark notion-block-1a70bdda378280dc97fbc7cd1367fa85" href="https://github.com/BradyFU/Awesome-Multimodal-Large-Language-Models"><div><div class="notion-bookmark-title">GitHub - BradyFU/Awesome-Multimodal-Large-Language-Models: :sparkles::sparkles:Latest Advances on Multimodal Large Language Models</div><div class="notion-bookmark-description">:sparkles::sparkles:Latest Advances on Multimodal Large Language Models - BradyFU/Awesome-Multimodal-Large-Language-Models</div><div class="notion-bookmark-link"><div class="notion-bookmark-link-icon"><img src="https://www.notion.so/image/https%3A%2F%2Fgithub.com%2Ffluidicon.png?table=block&amp;id=1a70bdda-3782-80dc-97fb-c7cd1367fa85&amp;t=1a70bdda-3782-80dc-97fb-c7cd1367fa85" alt="GitHub - BradyFU/Awesome-Multimodal-Large-Language-Models: :sparkles::sparkles:Latest Advances on Multimodal Large Language Models" loading="lazy" decoding="async"/></div><div class="notion-bookmark-link-text">https://github.com/BradyFU/Awesome-Multimodal-Large-Language-Models</div></div></div><div class="notion-bookmark-image"><img style="object-fit:cover" src="https://www.notion.so/image/https%3A%2F%2Fopengraph.githubassets.com%2Fcbe792bf7b65205c240c840e8278b0be4899523843f279e4021defbe2461a16c%2FBradyFU%2FAwesome-Multimodal-Large-Language-Models?table=block&amp;id=1a70bdda-3782-80dc-97fb-c7cd1367fa85&amp;t=1a70bdda-3782-80dc-97fb-c7cd1367fa85" alt="GitHub - BradyFU/Awesome-Multimodal-Large-Language-Models: :sparkles::sparkles:Latest Advances on Multimodal Large Language Models" loading="lazy" decoding="async"/></div></a></div><h3 class="notion-h notion-h2 notion-h-indent-1 notion-block-1a70bdda378281eb8b1ec327dedd2b16" data-id="1a70bdda378281eb8b1ec327dedd2b16"><span><div id="1a70bdda378281eb8b1ec327dedd2b16" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda378281eb8b1ec327dedd2b16" title="Reference"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">Reference</span></span></h3><div class="notion-row"><a target="_blank" rel="noopener noreferrer" class="notion-bookmark notion-block-1a70bdda378280e8be61f884214adff7" href="https://l12k1v7ghxm.feishu.cn/wiki/MEZHwaLlmirEI8k2Em4cg2cZn8e"><div><div class="notion-bookmark-title">l12k1v7ghxm.feishu.cn</div><div class="notion-bookmark-link"><div class="notion-bookmark-link-text">https://l12k1v7ghxm.feishu.cn/wiki/MEZHwaLlmirEI8k2Em4cg2cZn8e</div></div></div></a></div><div class="notion-row"><a target="_blank" rel="noopener noreferrer" class="notion-bookmark notion-block-1a70bdda378280db9b0cc1354137d404" href="https://cmu-mmml.github.io/spring2024/"><div><div class="notion-bookmark-title">11-777 MMML</div><div class="notion-bookmark-description">Multimodal Machine Learning 11-777- at CMU</div><div class="notion-bookmark-link"><div class="notion-bookmark-link-icon"><img src="https://www.notion.so/image/https%3A%2F%2Fcmu-mmml.github.io%2Fspring2024%2Fimages%2Fcmu-icon.png?table=block&amp;id=1a70bdda-3782-80db-9b0c-c1354137d404&amp;t=1a70bdda-3782-80db-9b0c-c1354137d404" alt="11-777 MMML" loading="lazy" decoding="async"/></div><div class="notion-bookmark-link-text">https://cmu-mmml.github.io/spring2024/</div></div></div></a></div><h3 class="notion-h notion-h2 notion-h-indent-1 notion-block-1a70bdda3782811e8c63fcc96dec8818" data-id="1a70bdda3782811e8c63fcc96dec8818"><span><div id="1a70bdda3782811e8c63fcc96dec8818" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda3782811e8c63fcc96dec8818" title="Theory"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">Theory</span></span></h3><div class="notion-row"><a target="_blank" rel="noopener noreferrer" class="notion-bookmark notion-block-1a70bdda378280c39a3cf708a30a276b" href="https://github.com/pliang279/awesome-multimodal-ml"><div><div class="notion-bookmark-title">GitHub - pliang279/awesome-multimodal-ml: Reading list for research topics in multimodal machine learning</div><div class="notion-bookmark-description">Reading list for research topics in multimodal machine learning - pliang279/awesome-multimodal-ml</div><div class="notion-bookmark-link"><div class="notion-bookmark-link-icon"><img src="https://www.notion.so/image/https%3A%2F%2Fgithub.com%2Ffluidicon.png?table=block&amp;id=1a70bdda-3782-80c3-9a3c-f708a30a276b&amp;t=1a70bdda-3782-80c3-9a3c-f708a30a276b" alt="GitHub - pliang279/awesome-multimodal-ml: Reading list for research topics in multimodal machine learning" loading="lazy" decoding="async"/></div><div class="notion-bookmark-link-text">https://github.com/pliang279/awesome-multimodal-ml</div></div></div><div class="notion-bookmark-image"><img style="object-fit:cover" src="https://www.notion.so/image/https%3A%2F%2Fopengraph.githubassets.com%2F91392e0970caf8f22512c821b4187811f637a4d7f61dd82533a09e76564be774%2Fpliang279%2Fawesome-multimodal-ml?table=block&amp;id=1a70bdda-3782-80c3-9a3c-f708a30a276b&amp;t=1a70bdda-3782-80c3-9a3c-f708a30a276b" alt="GitHub - pliang279/awesome-multimodal-ml: Reading list for research topics in multimodal machine learning" loading="lazy" decoding="async"/></div></a></div><div class="notion-row"><a target="_blank" rel="noopener noreferrer" class="notion-bookmark notion-block-1a70bdda378280e18d24df30b9d235b5" href="http://github.com"><div><div class="notion-bookmark-title">GitHub · Build and ship software on a single, collaborative platform</div><div class="notion-bookmark-description">Join the world&#x27;s most widely adopted, AI-powered developer platform where millions of developers, businesses, and the largest open source community build software that advances humanity.</div><div class="notion-bookmark-link"><div class="notion-bookmark-link-icon"><img src="https://www.notion.so/image/https%3A%2F%2Fgithub.com%2Ffluidicon.png?table=block&amp;id=1a70bdda-3782-80e1-8d24-df30b9d235b5&amp;t=1a70bdda-3782-80e1-8d24-df30b9d235b5" alt="GitHub · Build and ship software on a single, collaborative platform" loading="lazy" decoding="async"/></div><div class="notion-bookmark-link-text">http://github.com</div></div></div><div class="notion-bookmark-image"><img style="object-fit:cover" src="https://www.notion.so/image/https%3A%2F%2Fgithub.githubassets.com%2Fassets%2Fhome24-5939032587c9.jpg?table=block&amp;id=1a70bdda-3782-80e1-8d24-df30b9d235b5&amp;t=1a70bdda-3782-80e1-8d24-df30b9d235b5" alt="GitHub · Build and ship software on a single, collaborative platform" loading="lazy" decoding="async"/></div></a></div></main></div>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Transformer]]></title>
            <link>https://www.jinxiaokuang.top/technology/transformer-1</link>
            <guid>https://www.jinxiaokuang.top/technology/transformer-1</guid>
            <pubDate>Thu, 27 Feb 2025 00:00:00 GMT</pubDate>
            <description><![CDATA[介绍Transformer的架构，了解其内部原理]]></description>
            <content:encoded><![CDATA[<div id="notion-article" class="mx-auto overflow-hidden "><main class="notion light-mode notion-page notion-block-1a70bdda378280a4935bfced9ee81095"><div class="notion-viewport"></div><div class="notion-collection-page-properties"></div><div></div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-1a70bdda378281639e32fb78df00b32a" data-id="1a70bdda378281639e32fb78df00b32a"><span><div id="1a70bdda378281639e32fb78df00b32a" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda378281639e32fb78df00b32a" title="Overview"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">Overview</span></span></h2><div class="notion-text notion-block-1a70bdda3782816c8ca0d1091c543455"><b>Transformer</b>是一种用于自然语言处理（NLP）和其他序列到序列（sequence-to-sequence）任务的深度学习模型架构，它在2017年由Vaswani等人首次提出。Transformer架构引入了<b>自注意力机制</b>（self-attention mechanism），这是一个关键的创新，使其在处理序列数据时表现出色。</div><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-1a70bdda378281949876f76af5ab89c1" data-id="1a70bdda378281949876f76af5ab89c1"><span><div id="1a70bdda378281949876f76af5ab89c1" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda378281949876f76af5ab89c1" title="Method"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">Method</span></span></h2><h3 class="notion-h notion-h2 notion-h-indent-1 notion-block-1a70bdda3782814d854bd9f60c003656" data-id="1a70bdda3782814d854bd9f60c003656"><span><div id="1a70bdda3782814d854bd9f60c003656" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda3782814d854bd9f60c003656" title="整体架构"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">整体架构</span></span></h3><div class="notion-text notion-block-1a70bdda3782816fbb4fe2ae498a6309">主要由输入部分（输入输出嵌入与位置编码）、多层编码器、多层解码器以及输出部分（输出线性层与Softmax）四大部分组成。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-1a70bdda378281ceab02f75bac06911e"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:480px;max-width:100%;flex-direction:column"><img style="object-fit:cover" src="https://cdn.nlark.com/yuque/0/2024/png/33947789/1729050593342-ea89d8c1-fa15-4540-b952-712a37d96891.png?t=1a70bdda-3782-81ce-ab02-f75bac06911e" alt="Transformer整体架构" loading="lazy" decoding="async"/><figcaption class="notion-asset-caption">Transformer整体架构</figcaption></div></figure><h4 class="notion-h notion-h3 notion-h-indent-2 notion-block-1a70bdda378281bcbdb6ddcd208f72b3" data-id="1a70bdda378281bcbdb6ddcd208f72b3"><span><div id="1a70bdda378281bcbdb6ddcd208f72b3" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda378281bcbdb6ddcd208f72b3" title="输入部分"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">输入部分</span></span></h4><ul class="notion-list notion-list-disc notion-block-1a70bdda37828102acd2cdb70b6ae9ad"><li><b>源文本嵌入层</b>：将源文本中的词汇数字表示转换为向量表示，捕捉词汇间的关系。</li></ul><ul class="notion-list notion-list-disc notion-block-1a70bdda3782815192c1c5a610b5621e"><li><b>位置编码器</b>：为输入序列的每个位置生成位置向量，以便模型能够理解序列中的位置信息。</li></ul><ul class="notion-list notion-list-disc notion-block-1a70bdda37828124ba56c316673209b2"><li><b>目标文本嵌入层（在解码器中使用）</b>：将目标文本中的词汇数字表示转换为向量表示。</li></ul><h4 class="notion-h notion-h3 notion-h-indent-2 notion-block-1a70bdda3782816db517dede8d9fa9a1" data-id="1a70bdda3782816db517dede8d9fa9a1"><span><div id="1a70bdda3782816db517dede8d9fa9a1" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda3782816db517dede8d9fa9a1" title="编码器部分"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">编码器部分</span></span></h4><ul class="notion-list notion-list-disc notion-block-1a70bdda37828118920dddc1b57c7ce1"><li>由N个编码器层堆叠而成。</li></ul><ul class="notion-list notion-list-disc notion-block-1a70bdda378281249732f7daad377222"><li><b>每个编码器层由两个子层连接结构组成</b>：第一个子层是一个多头自注意力子层，第二个子层是一个前馈全连接子层。每个子层后都接有一个规范化层和一个残差连接。</li></ul><h4 class="notion-h notion-h3 notion-h-indent-2 notion-block-1a70bdda3782810b9241e119cba3a7e7" data-id="1a70bdda3782810b9241e119cba3a7e7"><span><div id="1a70bdda3782810b9241e119cba3a7e7" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda3782810b9241e119cba3a7e7" title="解码器部分"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">解码器部分</span></span></h4><ul class="notion-list notion-list-disc notion-block-1a70bdda37828170b984d1b5103e34e3"><li>由N个解码器层堆叠而成。</li></ul><ul class="notion-list notion-list-disc notion-block-1a70bdda378281cbab04e7d29a0f563b"><li><b>每个解码器层由三个子层连接结构组成</b>：第一个子层是一个带掩码的多头自注意力子层，第二个子层是一个多头注意力子层（编码器到解码器），第三个子层是一个前馈全连接子层。每个子层后都接有一个规范化层和一个残差连接。</li></ul><h4 class="notion-h notion-h3 notion-h-indent-2 notion-block-1a70bdda3782819894addf525f54762d" data-id="1a70bdda3782819894addf525f54762d"><span><div id="1a70bdda3782819894addf525f54762d" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda3782819894addf525f54762d" title="输出部分"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">输出部分</span></span></h4><ul class="notion-list notion-list-disc notion-block-1a70bdda37828117a368fcd04c661e2c"><li><b>线性层</b>：将解码器输出的向量转换为最s终的输出维度。</li></ul><ul class="notion-list notion-list-disc notion-block-1a70bdda378281cebc7ec625aa20dde0"><li><b>Softmax层</b>：将线性层的输出转换为概率分布，以便进行最终的预测。</li></ul><div class="notion-text notion-block-1a70bdda378281e18e17db560e7c5b34"><b>Encoder-Decoder（编码器-解码器）</b>左边是N个编码器，右边是N个解码器，Transformer中的N为6。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-1a70bdda3782816cbf66f646c0c16c89"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:336px;max-width:100%;flex-direction:column"><img style="object-fit:cover" src="https://cdn.nlark.com/yuque/0/2024/webp/33947789/1729051000260-5ba7fbec-b5e3-4921-9470-7bdacec112ed.webp?t=1a70bdda-3782-816c-bf66-f646c0c16c89" alt="Encoder-Decoder（编码器-解码器）" loading="lazy" decoding="async"/><figcaption class="notion-asset-caption">Encoder-Decoder（编码器-解码器）</figcaption></div></figure><h3 class="notion-h notion-h2 notion-h-indent-1 notion-block-1a70bdda378281c5af05ebf13a0ef84c" data-id="1a70bdda378281c5af05ebf13a0ef84c"><span><div id="1a70bdda378281c5af05ebf13a0ef84c" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda378281c5af05ebf13a0ef84c" title="Self-Attention"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">Self-Attention</span></span></h3><div class="notion-text notion-block-1a70bdda378281648a2cf23505417a8e">随着模型处理输入序列的每个单词，自注意力<b>会关注整个输入序列的所有单词</b>，帮助模型对本单词更好地进行编码。在处理过程中，自注意力机制会将对所有相关单词的理解融入到我们正在处理的单词中。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-1a70bdda378281a8a793ef4d084efc4e"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:384px;max-width:100%;flex-direction:column"><img style="object-fit:cover" src="https://cdn.nlark.com/yuque/0/2024/webp/33947789/1729051428051-bce33d78-3339-47e9-b35c-723fb6a67dfa.webp?t=1a70bdda-3782-81a8-a793-ef4d084efc4e" alt="Scaled Dot-Product Attention（缩放点积注意力）" loading="lazy" decoding="async"/><figcaption class="notion-asset-caption">Scaled Dot-Product Attention（缩放点积注意力）</figcaption></div></figure><div class="notion-text notion-block-1a70bdda37828170b307ee2cdf0862ac">上图是 Self-Attention 的结构，在计算的时候需要用到矩阵<b>Q(查询),K(键值),V(值)</b>。在实际中，Self-Attention 接收的是输入(单词的表示向量x组成的矩阵X) 或者上一个 Encoder block 的输出。而<b>Q,K,V</b>正是通过 Self-Attention 的输入进行线性变换得到的。</div><h4 class="notion-h notion-h3 notion-h-indent-2 notion-block-1a70bdda378281e99e91c347fcd31b73" data-id="1a70bdda378281e99e91c347fcd31b73"><span><div id="1a70bdda378281e99e91c347fcd31b73" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda378281e99e91c347fcd31b73" title="自注意力的计算"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">自注意力的计算</span></span></h4><ol start="1" class="notion-list notion-list-numbered notion-block-1a70bdda3782815786f6d4036dd0e297"><li><b>Q、K和V矩阵</b></li><ol class="notion-list notion-list-numbered notion-block-1a70bdda3782815786f6d4036dd0e297"><ul class="notion-list notion-list-disc notion-block-1a70bdda3782813f9139f454f1e65786"><li><b>Query矩阵（Q）</b>：表示当前的关注点或信息需求，用于与Key矩阵进行匹配。</li></ul><ul class="notion-list notion-list-disc notion-block-1a70bdda37828157b39af0b83dd33926"><li><b>Key矩阵（K）</b>：包含输入序列中各个位置的标识信息，用于被Query矩阵查询匹配。</li></ul><ul class="notion-list notion-list-disc notion-block-1a70bdda3782819689b5f77a4281367c"><li><b>Value矩阵（V）</b>：存储了与Key矩阵相对应的实际值或信息内容，当Query与某个Key匹配时，相应的Value将被用来计算输出。</li></ul></ol></ol><ol start="2" class="notion-list notion-list-numbered notion-block-1a70bdda3782811f9a19c752f29c1b17"><li><b>点积计算：</b></li><ol class="notion-list notion-list-numbered notion-block-1a70bdda3782811f9a19c752f29c1b17"><ul class="notion-list notion-list-disc notion-block-1a70bdda3782814a96e4d210d1aa5a8f"><li>通过计算Query矩阵和Key矩阵之间的点积（即对应元素相乘后求和），来衡量Query与每个Key之间的相似度或匹配程度。</li></ul></ol></ol><ol start="3" class="notion-list notion-list-numbered notion-block-1a70bdda3782810f9addede1de0b33bc"><li><b>缩放因子</b></li><ol class="notion-list notion-list-numbered notion-block-1a70bdda3782810f9addede1de0b33bc"><ul class="notion-list notion-list-disc notion-block-1a70bdda37828123b831dfa6e00281ce"><li>由于点积操作的结果可能非常大，尤其是在输入维度较高的情况下，这可能导致softmax函数在计算注意力权重时进入饱和区。为了避免这个问题，缩放点积注意力引入了一个缩放因子，通常是输入维度的平方根。点积结果除以这个缩放因子，可以使得softmax函数的输入保持在一个合理的范围内。</li></ul></ol></ol><ol start="4" class="notion-list notion-list-numbered notion-block-1a70bdda3782810fa8f1eb729c1b7cfc"><li><b>Softmax函数</b></li><ol class="notion-list notion-list-numbered notion-block-1a70bdda3782810fa8f1eb729c1b7cfc"><ul class="notion-list notion-list-disc notion-block-1a70bdda378281f996d1cafb8136b0a0"><li>将缩放后的点积结果输入到softmax函数中，计算每个Key相对于Query的注意力权重。Softmax函数将原始得分转换为概率分布，使得所有Key的注意力权重之和为1。</li></ul></ol></ol><ol start="5" class="notion-list notion-list-numbered notion-block-1a70bdda378281339624c04728a53ec7"><li><b>加权求和</b></li><ol class="notion-list notion-list-numbered notion-block-1a70bdda378281339624c04728a53ec7"><ul class="notion-list notion-list-disc notion-block-1a70bdda378281c9b4fcd129f752a03f"><li>使用计算出的注意力权重对Value矩阵进行加权求和，得到最终的输出。这个过程根据注意力权重的大小，将更多的关注放在与Query更匹配的Value上。</li></ul></ol></ol><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-1a70bdda378281a2b29cef8d666aaa1a"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:576px;max-width:100%;flex-direction:column"><img style="object-fit:cover" src="https://cdn.nlark.com/yuque/0/2024/png/33947789/1729051753334-237219a4-fc8a-4d13-bcb8-0042c7e4c372.png?t=1a70bdda-3782-81a2-b29c-ef8d666aaa1a" alt="自注意力计算" loading="lazy" decoding="async"/><figcaption class="notion-asset-caption">自注意力计算</figcaption></div></figure><h4 class="notion-h notion-h3 notion-h-indent-2 notion-block-1a70bdda378281ea8c7def54f4b6dd87" data-id="1a70bdda378281ea8c7def54f4b6dd87"><span><div id="1a70bdda378281ea8c7def54f4b6dd87" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda378281ea8c7def54f4b6dd87" title="Multi-Head Attention（多头注意力）"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">Multi-Head Attention（多头注意力）</span></span></h4><div class="notion-text notion-block-1a70bdda378281889b3cea1e452303e3">它允许模型同时关注来自不同位置的信息。通过分割原始的输入向量到多个头（head），每个头都能<b>独立地</b>学习<b>不同的注意力权重</b>，从而增强模型对输入序列中不同部分的关注能力。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-1a70bdda37828111a0e1d4f338b76f3a"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:192px;max-width:100%;flex-direction:column"><img style="object-fit:cover" src="https://cdn.nlark.com/yuque/0/2024/webp/33947789/1729051873238-a684227b-74e6-431e-8eb8-a439bee230a2.webp?t=1a70bdda-3782-8111-a0e1-d4f338b76f3a" alt="Multi-Head Attention（多头注意力）" loading="lazy" decoding="async"/><figcaption class="notion-asset-caption">Multi-Head Attention（多头注意力）</figcaption></div></figure><h3 class="notion-h notion-h2 notion-h-indent-1 notion-block-1a70bdda37828105a1c0fa592e4f4533" data-id="1a70bdda37828105a1c0fa592e4f4533"><span><div id="1a70bdda37828105a1c0fa592e4f4533" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda37828105a1c0fa592e4f4533" title="Encoder编码器"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">Encoder编码器</span></span></h3><div class="notion-text notion-block-1a70bdda378281bf8cecdda33909fb70">Transformer中的编码器部分一共6个相同的编码器层组成。每个编码器层都有两个子层，即多头自注意力层(Multi-Head Attention)和<b>逐位置的前馈神经网络</b>(Position-wise Feed-Forward Network)。在每个子层后面都有<b>残差连接Add</b>（图中的虚线）和<b>层归一化LN</b>操作，二者合起来称为<b>Add&amp;Norm</b>操作。Add &amp; Norm 层计算公式如下：</div><div class="notion-text notion-block-1a70bdda378281bdbae3e478bd354ad3">$ (X+(X))\(X+(X)) $</div><div class="notion-text notion-block-1a70bdda37828154ab9dd52b9c9fb715">其中X表示 Multi-Head Attention 或者 Feed Forward 的输入，MultiHeadAttention(X) 和 FeedForward(X) 表示输出(输出与输入 X 维度是一样的，所以可以相加)。</div><div class="notion-text notion-block-1a70bdda378281188e48ea0efab306c5"><b>Add的作用</b>：加入残差块的目的是为了防止在网络的训练过程中发生退化的问题，退化的意思就是深度神经网络通过增加网络的层数，Loss逐渐减小，然后趋于稳定达到饱和，然后再继续增加网络层数，Loss反而增大。</div><div class="notion-text notion-block-1a70bdda378281ca8628f535d30695de"><b>Norm的作用</b>：Layer Normalization(LN)是在同一个样本中不同神经元之间进行归一化，而BN是在同一个batch中不同样本之间的同一位置的神经元之间进行归一化。BN是对于相同的维度进行归一化，但是NLP中输入的都是词向量，一个300维的词向量，单独去分析它的每一维是没有意义地，在每一维上进行归一化也是适合地，因此这里选用的是LN。</div><div class="notion-text notion-block-1a70bdda3782813c9565f9ec24d4921c"><b>全连接层作用</b>：全连接层是一个两层的神经网络，先线性变换，然后ReLU非线性，再线性变换。这两层网络就是为了将输入的Z映射到更加高维的空间中然后通过非线性函数ReLU进行筛选，筛选完后再变回原来的维度。经过6个encoder后输入到decoder中。其计算公式如下：</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-1a70bdda3782817dbbf8df845d05c247"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:288px;max-width:100%;flex-direction:column"><img style="object-fit:cover" src="https://cdn.nlark.com/yuque/0/2024/webp/33947789/1729052025395-8fb9b76a-f6c8-4d36-8b77-33bb889ff392.webp?t=1a70bdda-3782-817d-bbf8-df845d05c247" alt="Encoder（编码器）架构" loading="lazy" decoding="async"/><figcaption class="notion-asset-caption">Encoder（编码器）架构</figcaption></div></figure><h3 class="notion-h notion-h2 notion-h-indent-1 notion-block-1a70bdda378281ae8f02e29b2b1753c5" data-id="1a70bdda378281ae8f02e29b2b1753c5"><span><div id="1a70bdda378281ae8f02e29b2b1753c5" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda378281ae8f02e29b2b1753c5" title="Decoder解码器"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">Decoder解码器</span></span></h3><div class="notion-text notion-block-1a70bdda37828150a375e89b0269ab03">Transformer中的解码器部分同样一共6个相同的解码器层组成。每个解码器层都有三个子层，掩蔽自注意力层(Masked Self-Attention)、<b>Encoder-Decoder注意力层、逐位置的前馈神经网络</b>。同样，在每个子层后面都有残差连接（图中的虚线）和层归一化（LayerNorm）操作，二者合起来称为Add&amp;Norm操作。</div><figure class="notion-asset-wrapper notion-asset-wrapper-image notion-block-1a70bdda37828148babad00231987f44"><div style="position:relative;display:flex;justify-content:center;align-self:center;width:288px;max-width:100%;flex-direction:column"><img style="object-fit:cover" src="https://cdn.nlark.com/yuque/0/2024/webp/33947789/1729052080099-a67fb330-d0a2-4939-a581-c201c0f16632.webp?t=1a70bdda-3782-8148-baba-d00231987f44" alt="Decoder（解码器）架构" loading="lazy" decoding="async"/><figcaption class="notion-asset-caption">Decoder（解码器）架构</figcaption></div></figure><h2 class="notion-h notion-h1 notion-h-indent-0 notion-block-1a70bdda37828121a39bc2807c8ae14e" data-id="1a70bdda37828121a39bc2807c8ae14e"><span><div id="1a70bdda37828121a39bc2807c8ae14e" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda37828121a39bc2807c8ae14e" title="Reference"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">Reference</span></span></h2><h3 class="notion-h notion-h2 notion-h-indent-1 notion-block-1a70bdda378281118061f6ad9780b0f5" data-id="1a70bdda378281118061f6ad9780b0f5"><span><div id="1a70bdda378281118061f6ad9780b0f5" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda378281118061f6ad9780b0f5" title="Code"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">Code</span></span></h3><div class="notion-text notion-block-1a70bdda3782818f8178fe92e81d98b8"><b>非官方</b></div><div class="notion-text notion-block-1a70bdda378281f2a5a7e098d3a2e56c"><a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://github.com/graykode/nlp-tutorial/tree/master/5-1.Transformer">nlp-tutorial/5-1.Transformer at master · graykode/nlp-tutorial (github.com)</a></div><div class="notion-text notion-block-1a70bdda37828140a0a4ebca4a22d645"><b>官方</b></div><div class="notion-text notion-block-1a70bdda378281a9a638d530d70a8879"><a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://github.com/huggingface/transformers">huggingface/transformers: 🤗 Transformers: State-of-the-art Machine Learning for Pytorch, TensorFlow, and JAX. (github.com)</a></div><h3 class="notion-h notion-h2 notion-h-indent-1 notion-block-1a70bdda3782817f932bf6e73fd06b19" data-id="1a70bdda3782817f932bf6e73fd06b19"><span><div id="1a70bdda3782817f932bf6e73fd06b19" class="notion-header-anchor"></div><a class="notion-hash-link" href="#1a70bdda3782817f932bf6e73fd06b19" title="Explanation"><svg viewBox="0 0 16 16" width="16" height="16"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a><span class="notion-h-title">Explanation</span></span></h3><div class="notion-text notion-block-1a70bdda378281379452e79897638f20"><a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://www.youtube.com/watch?v=ugWDIIOHtPA&amp;list=PLJV_el3uVTsOK_ZK5L0Iv_EQoL1JefRL4&amp;index=60">李宏毅-史上最强transformer讲解</a></div><div class="notion-text notion-block-1a70bdda378281f1a5d5d7312996203c"><a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://zhuanlan.zhihu.com/p/338817680">Transformer模型详解（图解最完整版） - 知乎 (zhihu.com)</a></div><div class="notion-text notion-block-1a70bdda3782819fbeb3e3cbd60cdd5b"><a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://blog.csdn.net/2301_76161259/article/details/141172163">一文看懂 Transformer！超级详解，小白入门必看！-CSDN博客</a></div><div class="notion-text notion-block-1a70bdda37828112afa8ec18d648e10a"><a target="_blank" rel="noopener noreferrer" class="notion-link" href="https://blog.csdn.net/weixin_57345774/article/details/134919890">Transformer源码（带注释）-CSDN博客</a></div></main></div>]]></content:encoded>
        </item>
    </channel>
</rss>