<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[python_innocent的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/python_innocent</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; python_innocent]]></copyright><item><title><![CDATA[cursor配置deepseek遇到的坑]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/142305984</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/142305984</guid><author>python_innocent</author><pubDate>Mon, 16 Sep 2024 19:23:59 +0800</pubDate><description><![CDATA[第一步，URL设置，第二步设置api key,第三步，点击verify，第四步点击右上角的按钮(verify之前按钮不会出现)这是cursor 0.40.4版本。]]></description><category></category></item><item><title><![CDATA[RuntimeError: Sizes of tensors must match except in dimension 3. Got 21 and 22 (The offending index]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/142002251</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/142002251</guid><author>python_innocent</author><pubDate>Sat, 07 Sep 2024 20:29:02 +0800</pubDate><description><![CDATA[本问题具体体现在步长为2的Conv，卷积后得到的尺寸，经过upsampling后无法回到原来的尺寸。x = m(x)，观察哪一行的x进行的错误，找到报错模块。所以更改Concat就可以了，找到common.py。]]></description><category></category></item><item><title><![CDATA[MMCV\MM系列 报错:ImportError: libcudart.so.12: cannot open shared object file: No such file or directory]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/134505057</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/134505057</guid><author>python_innocent</author><pubDate>Mon, 20 Nov 2023 12:37:08 +0800</pubDate><description><![CDATA[python版本太低了，通常是torch的版本太高，比如2.0，只能支持CUDA11.7，但此时3.8的python版本就会报这个错。这个时候只能重新建一个环境3.9的python环境了。]]></description><category></category></item><item><title><![CDATA[从零实现NMS]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/131134048</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/131134048</guid><author>python_innocent</author><pubDate>Fri, 09 Jun 2023 19:58:54 +0800</pubDate><description><![CDATA[说多说少都是泪，被问到NMS如何实现的时候侃侃而谈，但是一旦从零手写就捉襟见肘，往后再学习的时候一定作到从零书写每一块代码。]]></description><category></category></item><item><title><![CDATA[Github提交PR与修改pr]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/130560871</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/130560871</guid><author>python_innocent</author><pubDate>Mon, 08 May 2023 16:01:47 +0800</pubDate><description><![CDATA[如果已经在fork的仓库中更新了文件，想要想原始仓库提交pr：1.打开fork仓库，在页面顶部找到Pull requests标签，点击。2.在右上角点击New pull request按钮。3.在Comparing changes页面，会看到两个下拉菜单，左侧显示的是你fork仓库，右边是原始仓库，确保选择了正确的仓库，并选择要提交的分支(通常是master).4.点击Create pull request按钮，并填写一个简短的标题和描述，描述修改。]]></description><category></category></item><item><title><![CDATA[d2l BERT预训练(model+dataset*+train)]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/130404511</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/130404511</guid><author>python_innocent</author><pubDate>Thu, 27 Apr 2023 14:19:08 +0800</pubDate><description><![CDATA[## 构造辅助损失函数，用于计算mlm和nsp的损失值，相加### mlm_Y_hat为为对应Tmask预测的各个类分数;nsp_Y_hat为(bs,2)，为对应句子对是否为相连的。#@save# 前向传播# 计算遮蔽语⾔模型损失# 计算下⼀句⼦预测任务的损失这里计算损失就不要encoder_out了。注意mlm_l计算式，除weights是<pad>不算loss，最后再除weights数表示对每个预测取平均值。]]></description><category></category></item><item><title><![CDATA[d2l Transformer]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/130328759</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/130328759</guid><author>python_innocent</author><pubDate>Sun, 23 Apr 2023 19:59:35 +0800</pubDate><description><![CDATA[终于到变形金刚了，他的主要特征在于多头自注意力的使用，以及摒弃了rnn的操作。]]></description><category></category></item><item><title><![CDATA[d2l 使用attention的seq2seq]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/130311330</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/130311330</guid><author>python_innocent</author><pubDate>Sat, 22 Apr 2023 22:04:45 +0800</pubDate><description><![CDATA[这一章节与前面写好的function关联太大，建议看书P291.这章节主要讲述了添加attention的seq2seq,且只在decoder里面添加，所以全文都在讲这个decoter。]]></description><category></category></item><item><title><![CDATA[d2l 注意力评分函数 --附加mask_softmax讲解]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/130297708</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/130297708</guid><author>python_innocent</author><pubDate>Fri, 21 Apr 2023 22:03:43 +0800</pubDate><description><![CDATA[本章节tensor处理操作也不少，逐个讲解下：]]></description><category></category></item><item><title><![CDATA[d2l Nadaraya-Waston核回归]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/130241249</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/130241249</guid><author>python_innocent</author><pubDate>Wed, 19 Apr 2023 17:04:55 +0800</pubDate><description><![CDATA[注意力机制里面的非参数注意力汇聚。]]></description><category></category></item><item><title><![CDATA[d2l解码编码器与seq2seq]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/130232040</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/130232040</guid><author>python_innocent</author><pubDate>Tue, 18 Apr 2023 22:18:08 +0800</pubDate><description><![CDATA[总结一下简介实现rnn的输入输出，输入为X(T,bs,emb);net = nn.rnn(input,h)改为nn.GRU\LSTM都一样，都为(input,h)。在数据处理后，input可以为emb，或者是上述的emb+h。通过nn的net处理后得到的都是(T,bs,h)。]]></description><category></category></item><item><title><![CDATA[d2l机器翻译数据集]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/130231778</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/130231778</guid><author>python_innocent</author><pubDate>Tue, 18 Apr 2023 21:40:24 +0800</pubDate><description><![CDATA[两组(bs,T)+(bs)；分别表示features与labels的bs个长为T的最终标号,与该bs对应的valid_lenprint('X的有效⻓度:', X_valid_len)print('Y的有效⻓度:', Y_valid_len)break'''X的有效⻓度: tensor([4, 4])Y的有效⻓度: tensor([3, 3])''']]></description><category></category></item><item><title><![CDATA[d2l 里面GRU与Lstm实现]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/130147438</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/130147438</guid><author>python_innocent</author><pubDate>Fri, 14 Apr 2023 10:33:53 +0800</pubDate><description><![CDATA[此二者的本质都是对rnn进行改良：关注当前多还是关注之前多。]]></description><category></category></item><item><title><![CDATA[d2l从零与简洁实现RNN]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/130135431</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/130135431</guid><author>python_innocent</author><pubDate>Thu, 13 Apr 2023 17:45:16 +0800</pubDate><description><![CDATA[一层循环神经网络的输出被用作下一层循环神经网络的输入''''''这里的X经过rnn得到的Y，输出的是(T,bs,hiddens),不涉及层的运算，指每个时间步的隐状态state尺寸为(隐藏层数,bs,hidden)''''''torch里面的rnnlayer只包括隐藏层，不包括输出层。]]></description><category></category></item><item><title><![CDATA[d2l语言模型--生成小批量序列]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/130094446</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/130094446</guid><author>python_innocent</author><pubDate>Tue, 11 Apr 2023 21:59:59 +0800</pubDate><description><![CDATA[对语言模型的数据集处理做以下汇总与总结。]]></description><category></category></item><item><title><![CDATA[d2l 文本预处理textDataset]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/130066329</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/130066329</guid><author>python_innocent</author><pubDate>Mon, 10 Apr 2023 19:03:37 +0800</pubDate><description><![CDATA[这一节极其重要，重要到本来是d2l的内容我也要归到pyhon封面，这里面class的操作很多，让我娓娓道来！]]></description><category></category></item><item><title><![CDATA[d2l Markov序列模型]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/130065931</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/130065931</guid><author>python_innocent</author><pubDate>Mon, 10 Apr 2023 18:06:03 +0800</pubDate><description><![CDATA[本节的任务是使用Markov模型对后续序列进行预测，使用sin函数＋噪声绘制1000个样本点，取tau为4，即利用后四个的信息预测第五个。]]></description><category></category></item><item><title><![CDATA[d2l FCN全连接网络 --children()调用部分原网络操作与分割任务loss计算流程]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/129989109</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/129989109</guid><author>python_innocent</author><pubDate>Thu, 06 Apr 2023 16:19:33 +0800</pubDate><description><![CDATA[在此讲解FCN全连接网络，主要介绍里面的children()拾取原网络中部分网络层并附加新网络层的操作，并对分割任务的loss计算流程进行了debug，为后续类似做法做参考。]]></description><category></category></item><item><title><![CDATA[d2l风格迁移--包含tensor与pil图片互换操作]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/129990830</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/129990830</guid><author>python_innocent</author><pubDate>Thu, 06 Apr 2023 15:42:02 +0800</pubDate><description><![CDATA[对13章风格迁移任务进行讲解，并对其中的部分操作如pil与tensor呼唤进行具体介绍，方便后续调用！]]></description><category></category></item><item><title><![CDATA[d2l图像分割数据集制作]]></title><link>https://blog.youkuaiyun.com/python_innocent/article/details/129978340</link><guid>https://blog.youkuaiyun.com/python_innocent/article/details/129978340</guid><author>python_innocent</author><pubDate>Wed, 05 Apr 2023 22:17:31 +0800</pubDate><description><![CDATA[本文主要讲述图像分割Segmentation数据集的制作。]]></description><category></category></item></channel></rss>