Skip to content

Commit

Permalink
Site updated: 2024-09-23 20:28:25
Browse files Browse the repository at this point in the history
  • Loading branch information
YeQiuO committed Sep 23, 2024
1 parent 2b1ca9a commit f853ff8
Show file tree
Hide file tree
Showing 137 changed files with 5,109 additions and 5,026 deletions.
4 changes: 2 additions & 2 deletions -2024/Docker/index.html
Original file line number Diff line number Diff line change
Expand Up @@ -320,12 +320,12 @@ <h2 id="如何容器化-java-应用">如何容器化 Java 应用</h2>

<div class="post-nav">
<div class="post-nav-item">
<a href="/Design-Pattern/" rel="prev" title="设计模式">
<a href="/-NLP/Design-Pattern/" rel="prev" title="设计模式">
<i class="fa fa-angle-left"></i> 设计模式
</a>
</div>
<div class="post-nav-item">
<a href="/DL-basic/" rel="next" title="深度学习基础概念">
<a href="/-NLP/DL-basic/" rel="next" title="深度学习基础概念">
深度学习基础概念 <i class="fa fa-angle-right"></i>
</a>
</div>
Expand Down
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
50 changes: 25 additions & 25 deletions DL-basic/index.html → -NLP/DL-basic/index.html
Original file line number Diff line number Diff line change
Expand Up @@ -22,31 +22,31 @@
<meta name="description" content="机器学习是采用真实世界中采样的一小部分数据,并且希望模型能够对于未知数据有不错的泛化性能。">
<meta property="og:type" content="article">
<meta property="og:title" content="深度学习基础概念">
<meta property="og:url" content="https://yeqiuo.github.io/DL-basic/index.html">
<meta property="og:url" content="https://yeqiuo.github.io/-NLP/DL-basic/index.html">
<meta property="og:site_name" content="">
<meta property="og:description" content="机器学习是采用真实世界中采样的一小部分数据,并且希望模型能够对于未知数据有不错的泛化性能。">
<meta property="og:locale" content="zh_CN">
<meta property="og:image" content="https://yeqiuo.github.io/DL-basic/606386-20180722194316424-288674381.png">
<meta property="og:image" content="https://yeqiuo.github.io/DL-basic/o_L1.png">
<meta property="og:image" content="https://yeqiuo.github.io/DL-basic/o_L2.png">
<meta property="og:image" content="https://yeqiuo.github.io/DL-basic/overview_chatgpt.png">
<meta property="og:image" content="https://yeqiuo.github.io/DL-basic/image-20240716101748533.png">
<meta property="og:image" content="https://yeqiuo.github.io/DL-basic/image-20240715191912939.png">
<meta property="og:image" content="https://yeqiuo.github.io/DL-basic/image-20240715191936226.png">
<meta property="og:image" content="https://yeqiuo.github.io/DL-basic/image-20240715192056550.png">
<meta property="og:image" content="https://yeqiuo.github.io/-NLP/DL-basic/606386-20180722194316424-288674381.png">
<meta property="og:image" content="https://yeqiuo.github.io/-NLP/DL-basic/o_L1.png">
<meta property="og:image" content="https://yeqiuo.github.io/-NLP/DL-basic/o_L2.png">
<meta property="og:image" content="https://yeqiuo.github.io/-NLP/DL-basic/overview_chatgpt.png">
<meta property="og:image" content="https://yeqiuo.github.io/-NLP/DL-basic/image-20240716101748533.png">
<meta property="og:image" content="https://yeqiuo.github.io/-NLP/DL-basic/image-20240715191912939.png">
<meta property="og:image" content="https://yeqiuo.github.io/-NLP/DL-basic/image-20240715191936226.png">
<meta property="og:image" content="https://yeqiuo.github.io/-NLP/DL-basic/image-20240715192056550.png">
<meta property="article:published_time" content="2024-06-24T01:28:08.000Z">
<meta property="article:modified_time" content="2024-09-17T04:22:34.273Z">
<meta property="article:author" content="ZheWen Wang">
<meta property="article:tag" content="深度学习基础">
<meta name="twitter:card" content="summary">
<meta name="twitter:image" content="https://yeqiuo.github.io/DL-basic/606386-20180722194316424-288674381.png">
<meta name="twitter:image" content="https://yeqiuo.github.io/-NLP/DL-basic/606386-20180722194316424-288674381.png">


<link rel="canonical" href="https://yeqiuo.github.io/DL-basic/">
<link rel="canonical" href="https://yeqiuo.github.io/-NLP/DL-basic/">



<script class="next-config" data-name="page" type="application/json">{"sidebar":"","isHome":false,"isPost":true,"lang":"zh-CN","comments":true,"permalink":"https://yeqiuo.github.io/DL-basic/","path":"DL-basic/","title":"深度学习基础概念"}</script>
<script class="next-config" data-name="page" type="application/json">{"sidebar":"","isHome":false,"isPost":true,"lang":"zh-CN","comments":true,"permalink":"https://yeqiuo.github.io/-NLP/DL-basic/","path":"-NLP/DL-basic/","title":"深度学习基础概念"}</script>

<script class="next-config" data-name="calendar" type="application/json">""</script>
<title>深度学习基础概念 | 闻</title>
Expand Down Expand Up @@ -199,7 +199,7 @@


<article itemscope itemtype="http://schema.org/Article" class="post-content" lang="zh-CN">
<link itemprop="mainEntityOfPage" href="https://yeqiuo.github.io/DL-basic/">
<link itemprop="mainEntityOfPage" href="https://yeqiuo.github.io/-NLP/DL-basic/">

<span hidden itemprop="author" itemscope itemtype="http://schema.org/Person">
<meta itemprop="image" content="/images/avatar.gif">
Expand Down Expand Up @@ -289,7 +289,7 @@ <h2 id="噪声">噪声</h2>
<p><span
class="math display">\[\varepsilon^2=E_D\left[(y_D-y)^2\right]\]</span></p>
<h2 id="偏差-方差窘境">偏差-方差窘境</h2>
<img data-src="/DL-basic/606386-20180722194316424-288674381.png" class="" title="bias-variance-tradeoff">
<img data-src="/-NLP/DL-basic/606386-20180722194316424-288674381.png" class="" title="bias-variance-tradeoff">
<ul>
<li>偏向左侧时,训练数据的扰动不足以使学习器产生显著变化,此时偏差主导泛化误差,称为<strong>欠拟合现象</strong>
<ul>
Expand All @@ -306,8 +306,8 @@ <h2 id="偏差-方差窘境">偏差-方差窘境</h2>
<li>正则化在损失函数中加入一个与模型复杂度相关的惩罚项,以限制模型的复杂度并防止过拟合
<a target="_blank" rel="noopener" href="https://www.cnblogs.com/zingp/p/10375691.html">参考文章</a>
<ul>
<li>L1正则化是指权值向量w中各个元素的绝对值之和,可以使得参数稀疏化,即得到的参数是一个稀疏矩阵,进而可以用于特征选择。二维平面图像:<img data-src="/DL-basic/o_L1.png" class="" title="image"></li>
<li>L2正则化是指权值向量w中各个元素的平方和然后再求平方根,让权值尽可能小,如果参数足够小,数据偏移得多一点也不会对结果造成什么影响,专业一点的说法是<strong>抗扰动能力强</strong>。二维平面图像:<img data-src="/DL-basic/o_L2.png" class="" title="image"></li>
<li>L1正则化是指权值向量w中各个元素的绝对值之和,可以使得参数稀疏化,即得到的参数是一个稀疏矩阵,进而可以用于特征选择。二维平面图像:<img data-src="/-NLP/DL-basic/o_L1.png" class="" title="image"></li>
<li>L2正则化是指权值向量w中各个元素的平方和然后再求平方根,让权值尽可能小,如果参数足够小,数据偏移得多一点也不会对结果造成什么影响,专业一点的说法是<strong>抗扰动能力强</strong>。二维平面图像:<img data-src="/-NLP/DL-basic/o_L2.png" class="" title="image"></li>
</ul></li>
</ul>
<h1 id="模型训练">模型训练</h1>
Expand All @@ -327,26 +327,26 @@ <h2 id="强化学习">强化学习</h2>
<ul>
<li>归一化奖励分数、将奖励分数做白化处理</li>
</ul>
<img data-src="/DL-basic/overview_chatgpt.png" class="" title="overview_chatgpt">
<img data-src="/-NLP/DL-basic/overview_chatgpt.png" class="" title="overview_chatgpt">
<h3 id="rlhf">RLHF</h3>
<blockquote>
<p>Reinforcement Learning from Human
Feedback,以<strong>强化学习方式依据人类反馈优化语言模型</strong>,是
OpenAI 提出的生成领域(Decoder_only)的新训练范式</p>
</blockquote>
<p>Policy Gradient</p>
<style>.timgyuyjbgqa{zoom:50%;}</style>
<img data-src="/DL-basic/image-20240716101748533.png" class="timgyuyjbgqa" alt="image-20240716101748533">
<style>.mhhpjyorqakg{zoom:50%;}</style>
<img data-src="/-NLP/DL-basic/image-20240716101748533.png" class="mhhpjyorqakg" alt="image-20240716101748533">
<p>RM+PPO</p>
<p>步骤一、预训练语言模型</p>
<style>.csueuqxkylgs{zoom: 50%;}</style>
<img data-src="/DL-basic/image-20240715191912939.png" class="csueuqxkylgs" alt="image-20240715191912939">
<style>.hpjbtvqnhnei{zoom: 50%;}</style>
<img data-src="/-NLP/DL-basic/image-20240715191912939.png" class="hpjbtvqnhnei" alt="image-20240715191912939">
<p>步骤二、训练奖励模型 RM</p>
<style>.nwphvczwezmi{zoom: 50%;}</style>
<img data-src="/DL-basic/image-20240715191936226.png" class="nwphvczwezmi" alt="image-20240715191936226">
<style>.csiczvauxwow{zoom: 50%;}</style>
<img data-src="/-NLP/DL-basic/image-20240715191936226.png" class="csiczvauxwow" alt="image-20240715191936226">
<p>步骤三、用强化学习 PPO 微调</p>
<style>.ixsesjsgnxkj{zoom:50%;}</style>
<img data-src="/DL-basic/image-20240715192056550.png" class="ixsesjsgnxkj" alt="image-20240715192056550">
<style>.ulflbfxbhcpv{zoom:50%;}</style>
<img data-src="/-NLP/DL-basic/image-20240715192056550.png" class="ulflbfxbhcpv" alt="image-20240715192056550">
<p>DPO</p>
<p>KTO</p>
<p>ORPO</p>
Expand Down
File renamed without changes
File renamed without changes
File renamed without changes
12 changes: 6 additions & 6 deletions Design-Pattern/index.html → -NLP/Design-Pattern/index.html
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,7 @@
<meta name="description" content="项目架构的设计模式,和Java开发中的具体实现">
<meta property="og:type" content="article">
<meta property="og:title" content="设计模式">
<meta property="og:url" content="https://yeqiuo.github.io/Design-Pattern/index.html">
<meta property="og:url" content="https://yeqiuo.github.io/-NLP/Design-Pattern/index.html">
<meta property="og:site_name" content="">
<meta property="og:description" content="项目架构的设计模式,和Java开发中的具体实现">
<meta property="og:locale" content="zh_CN">
Expand All @@ -33,11 +33,11 @@
<meta name="twitter:card" content="summary">


<link rel="canonical" href="https://yeqiuo.github.io/Design-Pattern/">
<link rel="canonical" href="https://yeqiuo.github.io/-NLP/Design-Pattern/">



<script class="next-config" data-name="page" type="application/json">{"sidebar":"","isHome":false,"isPost":true,"lang":"zh-CN","comments":true,"permalink":"https://yeqiuo.github.io/Design-Pattern/","path":"Design-Pattern/","title":"设计模式"}</script>
<script class="next-config" data-name="page" type="application/json">{"sidebar":"","isHome":false,"isPost":true,"lang":"zh-CN","comments":true,"permalink":"https://yeqiuo.github.io/-NLP/Design-Pattern/","path":"-NLP/Design-Pattern/","title":"设计模式"}</script>

<script class="next-config" data-name="calendar" type="application/json">""</script>
<title>设计模式 | 闻</title>
Expand Down Expand Up @@ -190,7 +190,7 @@


<article itemscope itemtype="http://schema.org/Article" class="post-content" lang="zh-CN">
<link itemprop="mainEntityOfPage" href="https://yeqiuo.github.io/Design-Pattern/">
<link itemprop="mainEntityOfPage" href="https://yeqiuo.github.io/-NLP/Design-Pattern/">

<span hidden itemprop="author" itemscope itemtype="http://schema.org/Person">
<meta itemprop="image" content="/images/avatar.gif">
Expand Down Expand Up @@ -283,8 +283,8 @@ <h1 id="适配器模式">适配器模式</h1>

<div class="post-nav">
<div class="post-nav-item">
<a href="/-developer/Java-Interview/" rel="prev" title="Java 面试常见问题和知识点">
<i class="fa fa-angle-left"></i> Java 面试常见问题和知识点
<a href="/OS/" rel="prev" title="操作系统知识点">
<i class="fa fa-angle-left"></i> 操作系统知识点
</a>
</div>
<div class="post-nav-item">
Expand Down
File renamed without changes
16 changes: 8 additions & 8 deletions Faiss/index.html → -NLP/Faiss/index.html
Original file line number Diff line number Diff line change
Expand Up @@ -22,24 +22,24 @@
<meta name="description" content="faiss 索引类型、代码实践">
<meta property="og:type" content="article">
<meta property="og:title" content="Faiss 向量数据库">
<meta property="og:url" content="https://yeqiuo.github.io/Faiss/index.html">
<meta property="og:url" content="https://yeqiuo.github.io/-NLP/Faiss/index.html">
<meta property="og:site_name" content="">
<meta property="og:description" content="faiss 索引类型、代码实践">
<meta property="og:locale" content="zh_CN">
<meta property="og:image" content="https://yeqiuo.github.io/Faiss/image-20240911200426522.png">
<meta property="og:image" content="https://yeqiuo.github.io/-NLP/Faiss/image-20240911200426522.png">
<meta property="article:published_time" content="2024-09-11T07:30:17.000Z">
<meta property="article:modified_time" content="2024-09-17T06:11:24.793Z">
<meta property="article:author" content="ZheWen Wang">
<meta property="article:tag" content="faiss">
<meta name="twitter:card" content="summary">
<meta name="twitter:image" content="https://yeqiuo.github.io/Faiss/image-20240911200426522.png">
<meta name="twitter:image" content="https://yeqiuo.github.io/-NLP/Faiss/image-20240911200426522.png">


<link rel="canonical" href="https://yeqiuo.github.io/Faiss/">
<link rel="canonical" href="https://yeqiuo.github.io/-NLP/Faiss/">



<script class="next-config" data-name="page" type="application/json">{"sidebar":"","isHome":false,"isPost":true,"lang":"zh-CN","comments":true,"permalink":"https://yeqiuo.github.io/Faiss/","path":"Faiss/","title":"Faiss 向量数据库"}</script>
<script class="next-config" data-name="page" type="application/json">{"sidebar":"","isHome":false,"isPost":true,"lang":"zh-CN","comments":true,"permalink":"https://yeqiuo.github.io/-NLP/Faiss/","path":"-NLP/Faiss/","title":"Faiss 向量数据库"}</script>

<script class="next-config" data-name="calendar" type="application/json">""</script>
<title>Faiss 向量数据库 | 闻</title>
Expand Down Expand Up @@ -192,7 +192,7 @@


<article itemscope itemtype="http://schema.org/Article" class="post-content" lang="zh-CN">
<link itemprop="mainEntityOfPage" href="https://yeqiuo.github.io/Faiss/">
<link itemprop="mainEntityOfPage" href="https://yeqiuo.github.io/-NLP/Faiss/">

<span hidden itemprop="author" itemscope itemtype="http://schema.org/Person">
<meta itemprop="image" content="/images/avatar.gif">
Expand Down Expand Up @@ -326,7 +326,7 @@ <h1 id="bm25-实现">BM25 实现</h1>
<figure class="highlight python"><table><tr><td class="code"><pre><span class="line"><span class="keyword">import</span> jieba</span><br><span class="line"><span class="keyword">from</span> langchain.schema <span class="keyword">import</span> Document</span><br><span class="line"><span class="keyword">from</span> langchain_community.retrievers <span class="keyword">import</span> BM25Retriever</span><br><span class="line"></span><br><span class="line"><span class="keyword">def</span> <span class="title function_">create_bm25_retriever</span>(<span class="params">docs, top_k</span>):</span><br><span class="line"> bm25_retriever = BM25Retriever.from_documents(</span><br><span class="line"> docs,</span><br><span class="line"> preprocess_func=jieba.lcut_for_search,</span><br><span class="line"> )</span><br><span class="line"> bm25_retriever.k = top_k</span><br><span class="line"> <span class="keyword">return</span> bm25_retriever</span><br><span class="line"></span><br><span class="line"><span class="keyword">if</span> __name__ == <span class="string">&quot;__main__&quot;</span>:</span><br><span class="line"> docs = [</span><br><span class="line"> Document(page_content=<span class="string">&quot;这是一篇关于机器学习的文档。&quot;</span>),</span><br><span class="line"> Document(page_content=<span class="string">&quot;自然语言处理是人工智能的一个重要领域。&quot;</span>),</span><br><span class="line"> Document(page_content=<span class="string">&quot;深度学习是机器学习的一个子集。&quot;</span>),</span><br><span class="line"> ]</span><br><span class="line"> top_k = <span class="number">2</span></span><br><span class="line"> bm25_retriever = create_bm25_retriever(docs, top_k)</span><br><span class="line"> query = <span class="string">&quot;机器学习&quot;</span></span><br><span class="line"> results = bm25_retriever.get_relevant_documents(query)</span><br><span class="line"> <span class="keyword">for</span> rank, doc <span class="keyword">in</span> <span class="built_in">enumerate</span>(results, start=<span class="number">1</span>):</span><br><span class="line"> <span class="built_in">print</span>(<span class="string">f&quot;Rank <span class="subst">&#123;rank&#125;</span>: <span class="subst">&#123;doc&#125;</span>&quot;</span>)</span><br></pre></td></tr></table></figure>
<p>Debug 查看分词结果
<code>bm25_retriever.vectorizer.doc_freqs</code></p>
<img data-src="/Faiss/image-20240911200426522.png" class="" title="image-20240911200426522">
<img data-src="/-NLP/Faiss/image-20240911200426522.png" class="" title="image-20240911200426522">

</div>

Expand All @@ -343,7 +343,7 @@ <h1 id="bm25-实现">BM25 实现</h1>

<div class="post-nav">
<div class="post-nav-item">
<a href="/LLM-Model/" rel="prev" title="LLM 结构和创新点">
<a href="/-NLP/LLM-Model/" rel="prev" title="LLM 结构和创新点">
<i class="fa fa-angle-left"></i> LLM 结构和创新点
</a>
</div>
Expand Down
8 changes: 4 additions & 4 deletions Java-Algorithm/index.html → -NLP/Java-Algorithm/index.html
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,7 @@
<meta name="description" content="1.汇总用 Java 写算法的一些小技巧;2.总结语雀上的所有算法笔记。方便每次笔试之前复习总结">
<meta property="og:type" content="article">
<meta property="og:title" content="数据结构和算法">
<meta property="og:url" content="https://yeqiuo.github.io/Java-Algorithm/index.html">
<meta property="og:url" content="https://yeqiuo.github.io/-NLP/Java-Algorithm/index.html">
<meta property="og:site_name" content="">
<meta property="og:description" content="1.汇总用 Java 写算法的一些小技巧;2.总结语雀上的所有算法笔记。方便每次笔试之前复习总结">
<meta property="og:locale" content="zh_CN">
Expand All @@ -33,11 +33,11 @@
<meta name="twitter:card" content="summary">


<link rel="canonical" href="https://yeqiuo.github.io/Java-Algorithm/">
<link rel="canonical" href="https://yeqiuo.github.io/-NLP/Java-Algorithm/">



<script class="next-config" data-name="page" type="application/json">{"sidebar":"","isHome":false,"isPost":true,"lang":"zh-CN","comments":true,"permalink":"https://yeqiuo.github.io/Java-Algorithm/","path":"Java-Algorithm/","title":"数据结构和算法"}</script>
<script class="next-config" data-name="page" type="application/json">{"sidebar":"","isHome":false,"isPost":true,"lang":"zh-CN","comments":true,"permalink":"https://yeqiuo.github.io/-NLP/Java-Algorithm/","path":"-NLP/Java-Algorithm/","title":"数据结构和算法"}</script>

<script class="next-config" data-name="calendar" type="application/json">""</script>
<title>数据结构和算法 | 闻</title>
Expand Down Expand Up @@ -191,7 +191,7 @@


<article itemscope itemtype="http://schema.org/Article" class="post-content" lang="zh-CN">
<link itemprop="mainEntityOfPage" href="https://yeqiuo.github.io/Java-Algorithm/">
<link itemprop="mainEntityOfPage" href="https://yeqiuo.github.io/-NLP/Java-Algorithm/">

<span hidden itemprop="author" itemscope itemtype="http://schema.org/Person">
<meta itemprop="image" content="/images/avatar.gif">
Expand Down
Loading

0 comments on commit f853ff8

Please sign in to comment.