<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[joyce_peng的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/joyce_peng</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; joyce_peng]]></copyright><item><title><![CDATA[pip install -r跳过失败的包]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/143370305</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/143370305</guid><author>joyce_peng</author><pubDate>Wed, 30 Oct 2024 17:19:46 +0800</pubDate><description><![CDATA[pip]]></description><category></category></item><item><title><![CDATA[【分享笔记】符尧：预训练、指令微调、对齐、专业化——论大语言模型能力的来源]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/139008004</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/139008004</guid><author>joyce_peng</author><pubDate>Fri, 17 May 2024 17:24:38 +0800</pubDate><description><![CDATA[分享时间：2023.2。]]></description><category></category></item><item><title><![CDATA[transformer文章翻译【Attention is all you need】]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/134920576</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/134920576</guid><author>joyce_peng</author><pubDate>Mon, 11 Dec 2023 10:40:32 +0800</pubDate><description><![CDATA[（现状）主流的序列转换模型都是基于复杂的循环神经网络或卷积神经网络，且都包含一个encoder和一个decoder。表现最好的模型还通过attention机制把encoder和decoder联接起来。（创新点）我们提出了一个新的、简单的网络架构，Transformer. 它只基于单独的attention机制，完全避免使用循环和卷积。（优势）在两个翻译任务上表明，我们的模型在质量上更好，同时具有更高的并行性，且训练所需要的时间更少。（具体效果）]]></description><category></category></item><item><title><![CDATA[classification_report加入topk计算]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/132737026</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/132737026</guid><author>joyce_peng</author><pubDate>Thu, 07 Sep 2023 14:23:01 +0800</pubDate><description><![CDATA[官方文档：https://github.com/scikit-learn/scikit-learn/blob/7f9bad99d6e0a3e8ddf92a7e5561245224dab102/sklearn/metrics/_classification.py#L1551。参考：https://blog.youkuaiyun.com/dipizhong7224/article/details/104579159。]]></description><category></category></item><item><title><![CDATA[huggingface 自定义模型finetune训练测试--bert多任务]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/132716304</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/132716304</guid><author>joyce_peng</author><pubDate>Thu, 07 Sep 2023 14:20:30 +0800</pubDate><description><![CDATA[需要将bert改为多任务，但是官方仅支持多分类、二分类，并不支持多任务。改为多任务时我们需要修改输出层、loss、评测等。如果需要在bert结尾添加fc等也可以参考该添加方式。]]></description><category></category></item><item><title><![CDATA[[conda报错]simplejson.errors.JSONDecodeError: Expecting value: line 1 column 1 (char 0)]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/132363727</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/132363727</guid><author>joyce_peng</author><pubDate>Fri, 18 Aug 2023 16:00:01 +0800</pubDate><description><![CDATA[发现并不需要删除或者修改.condarc文件。看起来这个问题是channels的索引json文件出错导致的，我用conda clean -i命令移除了index cache，然后正常了。答案来源：https://github.com/conda/conda/issues/9590。（这里如果还不正常，可以在clean一下看看是否真的成功移除index）当安装新环境和安装新包时，报错。]]></description><category></category></item><item><title><![CDATA[论文图表--pyecharts使用]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/132058461</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/132058461</guid><author>joyce_peng</author><pubDate>Wed, 09 Aug 2023 11:06:36 +0800</pubDate><description><![CDATA[pyecharts是echarts的python接口，echarts本身是基于javascript的，由于javascript搭建也比较麻烦，这里使用pyecharts来画图。echarts常用来ppt和论文画图。]]></description><category></category></item><item><title><![CDATA[slurm/sbatch/srun 多步骤串行运行多个依赖性任务]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/131954422</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/131954422</guid><author>joyce_peng</author><pubDate>Thu, 27 Jul 2023 10:49:07 +0800</pubDate><description><![CDATA[slurm/sbatch/srun 多步骤串行运行多个依赖性任务]]></description><category></category></item><item><title><![CDATA[VQA评测evaluation代码：gqa / aokvqa / vqav2 / scienceQA]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/131702572</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/131702572</guid><author>joyce_peng</author><pubDate>Thu, 13 Jul 2023 15:03:32 +0800</pubDate><description><![CDATA[vqa评测]]></description><category></category></item><item><title><![CDATA[登录github.com报错]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/128737756</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/128737756</guid><author>joyce_peng</author><pubDate>Thu, 19 Jan 2023 20:28:51 +0800</pubDate><description><![CDATA[打开github报错]]></description><category></category></item><item><title><![CDATA[/home/mobaxterm/anaconda3/conda.exe: cannot execute binary file]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/128468374</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/128468374</guid><author>joyce_peng</author><pubDate>Wed, 28 Dec 2022 14:47:27 +0800</pubDate><description><![CDATA[conda安装报错]]></description><category></category></item><item><title><![CDATA[【论文精读】Is Synthetic data from generative models ready for image recognition? 生成数据对图像识别的影响]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/127826109</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/127826109</guid><author>joyce_peng</author><pubDate>Sat, 12 Nov 2022 22:14:12 +0800</pubDate><description><![CDATA[扩散模型生成的数据对图像识别的影响]]></description><category></category></item><item><title><![CDATA[【论文精读】The Missing Link: Finding label relations across datasets]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/125958161</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/125958161</guid><author>joyce_peng</author><pubDate>Mon, 15 Aug 2022 18:56:19 +0800</pubDate><description><![CDATA[探索不同数据集间标签关系，missing link]]></description><category></category></item><item><title><![CDATA[【论文精读】Grounded Language-Image Pre-training（GLIP）]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/125786938</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/125786938</guid><author>joyce_peng</author><pubDate>Sun, 24 Jul 2022 00:45:43 +0800</pubDate><description><![CDATA[glip（Grounded Language-Image Pre-training），可以做visual grounding和目标检测任务]]></description><category></category></item><item><title><![CDATA[【gitlab项目迁移】]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/125483020</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/125483020</guid><author>joyce_peng</author><pubDate>Mon, 27 Jun 2022 14:37:33 +0800</pubDate><description><![CDATA[gitlab项目迁移]]></description><category></category></item><item><title><![CDATA[【多模态】UNIMO]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/125388686</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/125388686</guid><author>joyce_peng</author><pubDate>Tue, 21 Jun 2022 13:55:19 +0800</pubDate><description><![CDATA[unimo模型]]></description><category></category></item><item><title><![CDATA[【论文摘要】标签合并Bamboo: Building Mega-Scale Vision Dataset Continually with Human-Machine Synergy]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/124667824</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/124667824</guid><author>joyce_peng</author><pubDate>Tue, 10 May 2022 14:35:45 +0800</pubDate><description><![CDATA[Bamboo: Building Mega-Scale Vision Dataset Continually with Human-Machine Synergy
前言：
论文链接：https://arxiv.org/pdf/2203.07845.pdf
github链接：https://github.com/Davidzhangyuanhan/Bamboo
商汤2022.3发表，主要内容为如何对开源数据集进行合并
论文分为摘要、背景介绍、相关工作、标签系统、Bamboo数据集构建、数据集数据、验证集实验
]]></description><category></category></item><item><title><![CDATA[【论文摘要】ImageNet：A Large-Scale Hierarchical Image Database]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/124544370</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/124544370</guid><author>joyce_peng</author><pubDate>Mon, 02 May 2022 21:45:08 +0800</pubDate><description><![CDATA[imagenet论文解读博客]]></description><category></category></item><item><title><![CDATA[【论文笔记】对比学习综述]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/124079844</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/124079844</guid><author>joyce_peng</author><pubDate>Sun, 10 Apr 2022 16:00:06 +0800</pubDate><description><![CDATA[对比学习综述；]]></description><category></category></item><item><title><![CDATA[【论文笔记】Model soups模型汤google]]></title><link>https://blog.youkuaiyun.com/joyce_peng/article/details/123777645</link><guid>https://blog.youkuaiyun.com/joyce_peng/article/details/123777645</guid><author>joyce_peng</author><pubDate>Sun, 27 Mar 2022 21:33:09 +0800</pubDate><description><![CDATA[论文地址：https://arxiv.org/pdf/2203.05482.pdf
论文相关讲解：https://baijiahao.baidu.com/s?id=1727838067837303006&amp;wfr=spider&amp;for=pc
论文贡献：

在imagenet上通过该方法对于ViT-G模型，达到sota 90.94%的效果
模型汤方法可以用在图像分类和NLP领域，提高下游的out-of-distribution和zero-shot效果
Finally, we analytical]]></description><category></category></item></channel></rss>