<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[m0_62965652的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/m0_62965652</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; m0_62965652]]></copyright><item><title><![CDATA[RAG中文本和图片的对齐]]></title><link>https://blog.youkuaiyun.com/m0_62965652/article/details/143030486</link><guid>https://blog.youkuaiyun.com/m0_62965652/article/details/143030486</guid><author>m0_62965652</author><pubDate>Tue, 22 Oct 2024 16:13:29 +0800</pubDate><description><![CDATA[首先，分别从文本和图片中提取特征。文本可以通过预训练的语言模型（如 Transformer、BERT、GPT）进行编码。图片则可以通过卷积神经网络（如 ResNet、Vision Transformer）进行编码。]]></description><category></category></item><item><title><![CDATA[langchain中RecursiveCharacterTextSplitter文档切割以及与其他分割方法的区别]]></title><link>https://blog.youkuaiyun.com/m0_62965652/article/details/142999598</link><guid>https://blog.youkuaiyun.com/m0_62965652/article/details/142999598</guid><author>m0_62965652</author><pubDate>Tue, 22 Oct 2024 15:59:53 +0800</pubDate><description><![CDATA[文本分割]]></description><category></category></item><item><title><![CDATA[Albert理解]]></title><link>https://blog.youkuaiyun.com/m0_62965652/article/details/140165215</link><guid>https://blog.youkuaiyun.com/m0_62965652/article/details/140165215</guid><author>m0_62965652</author><pubDate>Thu, 04 Jul 2024 00:20:58 +0800</pubDate><description><![CDATA[在网上看到的非常好的文章，怕找不到转载一下：Albert理解 - 光彩照人 - 博客园 (cnblogs.com)参考文献：1909.11942 (arxiv.org)下载相关模型：GitHub - brightmart/albert_zh: A LITE BERT FOR SELF-SUPERVISED LEARNING OF LANGUAGE REPRESENTATIONS, 海量中文预训练ALBERT模型]]></description><category></category></item><item><title><![CDATA[XLNet中对AE和AR方法的改进-排列语言模型（Permutation Language Modeling）]]></title><link>https://blog.youkuaiyun.com/m0_62965652/article/details/140145193</link><guid>https://blog.youkuaiyun.com/m0_62965652/article/details/140145193</guid><author>m0_62965652</author><pubDate>Thu, 04 Jul 2024 00:20:51 +0800</pubDate><description><![CDATA[排列语言模型（Permutation Language Modeling）]]></description><category></category></item><item><title><![CDATA[numpy中where()]]></title><link>https://blog.youkuaiyun.com/m0_62965652/article/details/139866770</link><guid>https://blog.youkuaiyun.com/m0_62965652/article/details/139866770</guid><author>m0_62965652</author><pubDate>Sun, 23 Jun 2024 21:42:40 +0800</pubDate><description><![CDATA[中的每个元素进行判断,如果元素值大于5,则该位置的结果为True,否则为False。注意这里返回的是坐标.也就是说，当where()只传入一个参数时，返回的是符合条件的坐标。的这种根据条件从两个数组/值中进行选择赋值的操作,在数据处理中非常常见和有用。是NumPy库中一个非常有用的函数,它根据给定条件从一个数组中筛选元素。中元素的值,用10替换了大于5的元素,其余元素保持原值。是当条件为False时,输出数组中该位置应该赋予的值。是当条件为True时,输出数组中该位置应该赋予的值。大小相同的布尔数组。]]></description><category></category></item><item><title><![CDATA[numpy中的布尔索引机制]]></title><link>https://blog.youkuaiyun.com/m0_62965652/article/details/139866102</link><guid>https://blog.youkuaiyun.com/m0_62965652/article/details/139866102</guid><author>m0_62965652</author><pubDate>Sun, 23 Jun 2024 21:36:48 +0800</pubDate><description><![CDATA[具体来说,如果布尔数组的shape比原数组的shape小,NumPy会自动对布尔数组的shape进行扩展,使其与原数组的shape相匹配。扩展规则是:在布尔数组的shape前面补1,直到其shape与原数组的shape一致为止。也就是说，布尔索引就是当数组接收到一个布尔数组时，会将内部的布尔数组中的元素和自己的内部元素一一对应。,并从原数组中选取这些位置对应的元素。但在当数组大小不匹配时，NumPy有。布尔索引的工作原理是,的机制来处理这种情况。]]></description><category></category></item><item><title><![CDATA[全连接层中先升维再降维其中的含义：以FeedForward为例]]></title><link>https://blog.youkuaiyun.com/m0_62965652/article/details/139402960</link><guid>https://blog.youkuaiyun.com/m0_62965652/article/details/139402960</guid><author>m0_62965652</author><pubDate>Mon, 03 Jun 2024 21:42:39 +0800</pubDate><description><![CDATA[全连接层中先升维再降维]]></description><category></category></item><item><title><![CDATA[关于seq2seq模型loss使用交叉熵具体是如何计算的]]></title><link>https://blog.youkuaiyun.com/m0_62965652/article/details/138120315</link><guid>https://blog.youkuaiyun.com/m0_62965652/article/details/138120315</guid><author>m0_62965652</author><pubDate>Mon, 03 Jun 2024 21:34:50 +0800</pubDate><description><![CDATA[seq2seq模型loss使用交叉熵是如何计算的]]></description><category></category></item><item><title><![CDATA[nn.RNN的输入输出及其内部结构说明]]></title><link>https://blog.youkuaiyun.com/m0_62965652/article/details/138140007</link><guid>https://blog.youkuaiyun.com/m0_62965652/article/details/138140007</guid><author>m0_62965652</author><pubDate>Wed, 24 Apr 2024 00:02:52 +0800</pubDate><description><![CDATA[原因是h_n只保留了最后一步的 hidden_state，但中间的 hidden_state 也有可能会参与计算，所以 pytorch 把中间每一步输出的 hidden_state 都放到output中（当然，只保留了 hidden_state 最后一层的输出），因此，你可以发现这个output的维度是。3.h_0(隐藏层)（h_0代表隐藏层的输入输出，在rnn网络中输入输出是格式是相同的）如果没有提供,默认为全0num_layers是RNN的层数。]]></description><category></category></item><item><title><![CDATA[NLP的第一步：如何将文本变为embedding输入向量[N,T]]]></title><link>https://blog.youkuaiyun.com/m0_62965652/article/details/138008506</link><guid>https://blog.youkuaiyun.com/m0_62965652/article/details/138008506</guid><author>m0_62965652</author><pubDate>Wed, 24 Apr 2024 00:02:32 +0800</pubDate><description><![CDATA[让我们来看一个具体的例子。]]></description><category></category></item><item><title><![CDATA[Optimizer优化器发展 从SGD到Adam(W)及其对比 (附Pytorch代码)]]></title><link>https://blog.youkuaiyun.com/m0_62965652/article/details/136969277</link><guid>https://blog.youkuaiyun.com/m0_62965652/article/details/136969277</guid><author>m0_62965652</author><pubDate>Fri, 05 Apr 2024 17:32:54 +0800</pubDate><description><![CDATA[这里将讲解从最初的梯度下降一步一步完善直到AdamW并附有代码]]></description><category></category></item><item><title><![CDATA[NLP基础_词嵌入word embedding模型合集(框架理解版)]]></title><link>https://blog.youkuaiyun.com/m0_62965652/article/details/135370268</link><guid>https://blog.youkuaiyun.com/m0_62965652/article/details/135370268</guid><author>m0_62965652</author><pubDate>Thu, 21 Mar 2024 16:38:32 +0800</pubDate><description><![CDATA[在分词之后，对于文本类型的特征属性，需要进行，也就是需要。因为神经网络的本质还是数学运算。所以我们第一步是将分词转化为数字符号进行表示。基础方式如下：序号化、哑编码(One-Hot)、词袋法(BOW/TF) TF-IDF(Term frequency-inverse document frequency)主题模型LSALDA等word embedding部分：Word2VecChar2VecDoc2Vec紧接上文。]]></description><category></category></item><item><title><![CDATA[nlp中将文本数字化的方法]]></title><link>https://blog.youkuaiyun.com/m0_62965652/article/details/136871424</link><guid>https://blog.youkuaiyun.com/m0_62965652/article/details/136871424</guid><author>m0_62965652</author><pubDate>Wed, 20 Mar 2024 11:40:56 +0800</pubDate><description><![CDATA[分词后的下一步]]></description><category></category></item><item><title><![CDATA[Word2vec详解（附Gensim代码）]]></title><link>https://blog.youkuaiyun.com/m0_62965652/article/details/136764709</link><guid>https://blog.youkuaiyun.com/m0_62965652/article/details/136764709</guid><author>m0_62965652</author><pubDate>Tue, 19 Mar 2024 10:37:02 +0800</pubDate><description><![CDATA[简单讲解word2vec及其代码]]></description><category></category></item><item><title><![CDATA[文本转向量过程中的矩阵变化示例]]></title><link>https://blog.youkuaiyun.com/m0_62965652/article/details/135496469</link><guid>https://blog.youkuaiyun.com/m0_62965652/article/details/135496469</guid><author>m0_62965652</author><pubDate>Wed, 10 Jan 2024 10:29:38 +0800</pubDate><description><![CDATA[Word2Vec中的矩阵变换示例]]></description><category></category></item><item><title><![CDATA[NLP基础_分词_jieba学习笔记]]></title><link>https://blog.youkuaiyun.com/m0_62965652/article/details/135127169</link><guid>https://blog.youkuaiyun.com/m0_62965652/article/details/135127169</guid><author>m0_62965652</author><pubDate>Fri, 22 Dec 2023 18:04:58 +0800</pubDate><description><![CDATA[﻿#自定义词典：一词占一行，每行分三个部分：词语，词频（可忽略），词性（可忽略）饿了么 2 nt美团 2 nr#加载词典word_list = jieba.cut('饿了么是你值得信赖的选择', HMM=True)print("【载入词典后】: {}".format('/'.join(word_list)))【载入词典后】: 饿了么/是/你/值得/信赖/的/选择可以在程序中动态修改词典#例如分词为 徐 狰狞 时，可以做以下操作。]]></description><category></category></item></channel></rss>