<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Lowin Li</title>
  
  <subtitle>大浪淘沙，风起帆扬</subtitle>
  <link href="https://lowin.li/atom.xml" rel="self"/>
  
  <link href="https://lowin.li/"/>
  <updated>2026-01-02T16:52:28.009Z</updated>
  <id>https://lowin.li/</id>
  
  <author>
    <name>Lowin Li</name>
    
  </author>
  
  <generator uri="https://hexo.io/">Hexo</generator>
  
  <entry>
    <title>IQuest-Coder-V1调研，NanobananaPro + Gemini3Pro生成</title>
    <link href="https://lowin.li/2026/01/03/iquest-coder-v1-diao-yan/"/>
    <id>https://lowin.li/2026/01/03/iquest-coder-v1-diao-yan/</id>
    <published>2026-01-02T16:00:00.000Z</published>
    <updated>2026-01-02T16:52:28.009Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;h2 id=&quot;&quot;&gt;&lt;a href=&quot;#&quot; class=&quot;headerlink&quot; title=&quot;&quot;&gt;&lt;/a&gt;&lt;img src=&quot;/image/iq-coder/1.svg&quot;&gt;&lt;/h2&gt;&lt;h2 id=&quot;-1&quot;&gt;&lt;a href=&quot;#-1&quot; class=&quot;headerlink&quot;</summary>
        
      
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="vibe-coding" scheme="https://lowin.li/tags/vibe-coding/"/>
    
  </entry>
  
  <entry>
    <title>Vibecoding 时代，程序员会消失吗？——从“全自动”到“半自动”的冷思考</title>
    <link href="https://lowin.li/2025/12/18/vibecoding-shi-dai-cheng-xu-yuan-hui-xiao-shi-ma-cong-quan-zi-dong-dao-ban-zi-dong-de-leng-si-kao/"/>
    <id>https://lowin.li/2025/12/18/vibecoding-shi-dai-cheng-xu-yuan-hui-xiao-shi-ma-cong-quan-zi-dong-dao-ban-zi-dong-de-leng-si-kao/</id>
    <published>2025-12-17T16:00:00.000Z</published>
    <updated>2025-12-18T07:45:25.547Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;今年，“Vibecoding” 的概念席卷了技术圈。像 ClaudeCode、Lovable 这类产品，号称只需一句自然语言就能生成整套应用；CodingAgent 更是通过不断的 Action-Observation</summary>
        
      
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="vibe-coding" scheme="https://lowin.li/tags/vibe-coding/"/>
    
    <category term="Copilot" scheme="https://lowin.li/tags/Copilot/"/>
    
  </entry>
  
  <entry>
    <title>AzureOpenAI vs OpenAI</title>
    <link href="https://lowin.li/2023/02/17/azureopenaivsopenai/"/>
    <id>https://lowin.li/2023/02/17/azureopenaivsopenai/</id>
    <published>2023-02-17T07:00:00.000Z</published>
    <updated>2023-10-06T11:29:49.000Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;OpenAI是一家人工智能研究机构，最近几个月发布的ChatGPT火遍全球。OpenAI官方提供了API接口，可以帮助开发者轻松地介入Ada、Babbage、Curie、Davinci等模型，尤其是OpenAI发布的text-davinci-003模型，它的通用能力，让大家</summary>
        
      
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="GPT" scheme="https://lowin.li/tags/GPT/"/>
    
    <category term="AIGC" scheme="https://lowin.li/tags/AIGC/"/>
    
    <category term="OpenAI" scheme="https://lowin.li/tags/OpenAI/"/>
    
    <category term="Azure" scheme="https://lowin.li/tags/Azure/"/>
    
  </entry>
  
  <entry>
    <title>ChatGPT出圈的秘诀</title>
    <link href="https://lowin.li/2023/01/29/chatgpt-chu-quan-de-mi-jue/"/>
    <id>https://lowin.li/2023/01/29/chatgpt-chu-quan-de-mi-jue/</id>
    <published>2023-01-29T07:00:00.000Z</published>
    <updated>2023-10-06T11:29:49.000Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;通篇翻译自&lt;br&gt;&lt;a href=&quot;https://huggingface.co/blog/dialog-agents&quot;&gt;Rajani et al., “What Makes a Dialog Agent Useful?”, Hugging Face Blog,</summary>
        
      
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="RLHF" scheme="https://lowin.li/tags/RLHF/"/>
    
    <category term="ChatGPT" scheme="https://lowin.li/tags/ChatGPT/"/>
    
    <category term="AI" scheme="https://lowin.li/tags/AI/"/>
    
  </entry>
  
  <entry>
    <title>人工反馈的强化学习</title>
    <link href="https://lowin.li/2023/01/02/ren-gong-fan-kui-de-qiang-hua-xue-xi/"/>
    <id>https://lowin.li/2023/01/02/ren-gong-fan-kui-de-qiang-hua-xue-xi/</id>
    <published>2023-01-02T11:00:00.000Z</published>
    <updated>2023-10-06T11:29:49.000Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;h1 id=&quot;人工反馈的强化学习&quot;&gt;&lt;a href=&quot;#人工反馈的强化学习&quot; class=&quot;headerlink&quot; title=&quot;人工反馈的强化学习&quot;&gt;&lt;/a&gt;人工反馈的强化学习&lt;/h1&gt;&lt;ul&gt;
&lt;li&gt;翻译自&lt;a</summary>
        
      
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="NLP" scheme="https://lowin.li/tags/NLP/"/>
    
    <category term="transformers" scheme="https://lowin.li/tags/transformers/"/>
    
    <category term="RLHF" scheme="https://lowin.li/tags/RLHF/"/>
    
    <category term="强化学习" scheme="https://lowin.li/tags/%E5%BC%BA%E5%8C%96%E5%AD%A6%E4%B9%A0/"/>
    
  </entry>
  
  <entry>
    <title>Stable Diffusion的模型量化，降低内存75%、Streamlit的在线生成图片调试、docker服务部署</title>
    <link href="https://lowin.li/2022/10/09/stable-diffusion/"/>
    <id>https://lowin.li/2022/10/09/stable-diffusion/</id>
    <published>2022-10-09T14:00:00.000Z</published>
    <updated>2023-10-06T11:29:49.000Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;h2 id=&quot;摘要&quot;&gt;&lt;a href=&quot;#摘要&quot; class=&quot;headerlink&quot; title=&quot;摘要&quot;&gt;&lt;/a&gt;摘要&lt;/h2&gt;&lt;p&gt;最近几个月开源的&lt;a href=&quot;https://github.com/CompVis/stable-diffusion&quot;&gt;Stable</summary>
        
      
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="Quantization" scheme="https://lowin.li/tags/Quantization/"/>
    
    <category term="Stable-Diffusion" scheme="https://lowin.li/tags/Stable-Diffusion/"/>
    
    <category term="Streamlit" scheme="https://lowin.li/tags/Streamlit/"/>
    
    <category term="CV" scheme="https://lowin.li/tags/CV/"/>
    
    <category term="OnnxRuntime" scheme="https://lowin.li/tags/OnnxRuntime/"/>
    
  </entry>
  
  <entry>
    <title>训练一个SentenceTransformer模型</title>
    <link href="https://lowin.li/2022/09/12/xun-lian-yi-ge-sentencetransformer-mo-xing/"/>
    <id>https://lowin.li/2022/09/12/xun-lian-yi-ge-sentencetransformer-mo-xing/</id>
    <published>2022-09-12T12:00:00.000Z</published>
    <updated>2022-09-13T02:44:10.000Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;p&gt;原&lt;a href=&quot;https://huggingface.co/blog/how-to-train-sentence-transformers&quot;&gt;博客&lt;/a&gt;&lt;br&gt;完整notebook代码&lt;/p&gt;
&lt;p&gt;&lt;a</summary>
        
      
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="NLP" scheme="https://lowin.li/tags/NLP/"/>
    
    <category term="transformers" scheme="https://lowin.li/tags/transformers/"/>
    
    <category term="SentenceTransformers" scheme="https://lowin.li/tags/SentenceTransformers/"/>
    
    <category term="Semantic Search" scheme="https://lowin.li/tags/Semantic-Search/"/>
    
  </entry>
  
  <entry>
    <title>8位混合精度矩阵乘法，小硬件跑大模型</title>
    <link href="https://lowin.li/2022/09/04/8-wei-hun-he-jing-du-ju-zhen-cheng-fa-xiao-ying-jian-pao-da-mo-xing/"/>
    <id>https://lowin.li/2022/09/04/8-wei-hun-he-jing-du-ju-zhen-cheng-fa-xiao-ying-jian-pao-da-mo-xing/</id>
    <published>2022-09-04T07:00:00.000Z</published>
    <updated>2022-09-04T06:02:10.000Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;ul&gt;
&lt;li&gt;原论文：&lt;a href=&quot;https://arxiv.org/pdf/2208.07339.pdf&quot;&gt;https://arxiv.org/pdf/2208.07339.pdf&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;原博客：&lt;a</summary>
        
      
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="NLP" scheme="https://lowin.li/tags/NLP/"/>
    
    <category term="transformers" scheme="https://lowin.li/tags/transformers/"/>
    
    <category term="Quantization" scheme="https://lowin.li/tags/Quantization/"/>
    
    <category term="GPT" scheme="https://lowin.li/tags/GPT/"/>
    
    <category term="GPU" scheme="https://lowin.li/tags/GPU/"/>
    
  </entry>
  
  <entry>
    <title>Constrained Beam Search</title>
    <link href="https://lowin.li/2022/07/03/shi-yong-transformers-zuo-xian-zhi-ji-shu-sou-suo-constrained-beam-search-de-wen-ben-sheng-cheng/"/>
    <id>https://lowin.li/2022/07/03/shi-yong-transformers-zuo-xian-zhi-ji-shu-sou-suo-constrained-beam-search-de-wen-ben-sheng-cheng/</id>
    <published>2022-07-03T14:00:00.000Z</published>
    <updated>2022-07-03T22:42:12.000Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;h1 id=&quot;使用Transformers做限制集束搜索（Constrained-Beam-Search）的文本生成&quot;&gt;&lt;a href=&quot;#使用Transformers做限制集束搜索（Constrained-Beam-Search）的文本生成&quot;</summary>
        
      
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="transformers" scheme="https://lowin.li/tags/transformers/"/>
    
    <category term="nlp" scheme="https://lowin.li/tags/nlp/"/>
    
    <category term="beam search" scheme="https://lowin.li/tags/beam-search/"/>
    
    <category term="text generation" scheme="https://lowin.li/tags/text-generation/"/>
    
  </entry>
  
  <entry>
    <title>盘点开源“Copilot”，do it yourself</title>
    <link href="https://lowin.li/2022/06/27/pan-dian-kai-yuan-copilot/"/>
    <id>https://lowin.li/2022/06/27/pan-dian-kai-yuan-copilot/</id>
    <published>2022-06-27T15:00:00.000Z</published>
    <updated>2022-06-28T04:52:57.000Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;h2 id=&quot;目录&quot;&gt;&lt;a href=&quot;#目录&quot; class=&quot;headerlink&quot; title=&quot;目录&quot;&gt;&lt;/a&gt;目录&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;&lt;a</summary>
        
      
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="transformers" scheme="https://lowin.li/tags/transformers/"/>
    
    <category term="nlp" scheme="https://lowin.li/tags/nlp/"/>
    
    <category term="text generation" scheme="https://lowin.li/tags/text-generation/"/>
    
    <category term="github" scheme="https://lowin.li/tags/github/"/>
    
    <category term="gpt" scheme="https://lowin.li/tags/gpt/"/>
    
    <category term="onnxruntime" scheme="https://lowin.li/tags/onnxruntime/"/>
    
    <category term="codegen" scheme="https://lowin.li/tags/codegen/"/>
    
    <category term="vscode" scheme="https://lowin.li/tags/vscode/"/>
    
    <category term="fastgpt" scheme="https://lowin.li/tags/fastgpt/"/>
    
  </entry>
  
  <entry>
    <title>使用fastgpt提速huggingface的GPT文本生成模型</title>
    <link href="https://lowin.li/2022/06/25/shi-yong-fastgpt-ti-su-huggingface-de-gpt-wen-ben-sheng-cheng-mo-xing/"/>
    <id>https://lowin.li/2022/06/25/shi-yong-fastgpt-ti-su-huggingface-de-gpt-wen-ben-sheng-cheng-mo-xing/</id>
    <published>2022-06-25T04:00:00.000Z</published>
    <updated>2022-06-25T05:47:52.000Z</updated>
    
    
      
      
        
        
    <summary type="html">&lt;h1 id=&quot;使用fastgpt提速huggingface的GPT文本生成模型&quot;&gt;&lt;a href=&quot;#使用fastgpt提速huggingface的GPT文本生成模型&quot; class=&quot;headerlink&quot;</summary>
        
      
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="transformers" scheme="https://lowin.li/tags/transformers/"/>
    
    <category term="nlp" scheme="https://lowin.li/tags/nlp/"/>
    
    <category term="text generation" scheme="https://lowin.li/tags/text-generation/"/>
    
    <category term="github" scheme="https://lowin.li/tags/github/"/>
    
    <category term="gpt" scheme="https://lowin.li/tags/gpt/"/>
    
    <category term="onnx" scheme="https://lowin.li/tags/onnx/"/>
    
  </entry>
  
  <entry>
    <title>docker启devpi服务</title>
    <link href="https://lowin.li/2022/03/05/devpi/"/>
    <id>https://lowin.li/2022/03/05/devpi/</id>
    <published>2022-03-04T16:00:00.000Z</published>
    <updated>2023-10-06T11:29:49.000Z</updated>
    
    
    <summary type="html">&lt;h2 id=&quot;相关链接：&quot;&gt;&lt;a href=&quot;#相关链接：&quot; class=&quot;headerlink&quot; title=&quot;相关链接：&quot;&gt;&lt;/a&gt;相关链接：&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;&lt;a href=&quot;https://github.com/LowinLi/devpi-docker&quot;&gt;github&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href=&quot;https://hub.docker.com/repository/docker/lowinli98/devpi&quot;&gt;dockerhub&lt;/a&gt;&lt;h2 id=&quot;简要&quot;&gt;&lt;a href=&quot;#简要&quot; class=&quot;headerlink&quot; title=&quot;简要&quot;&gt;&lt;/a&gt;简要&lt;/h2&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href=&quot;https://devpi.net/docs/devpi/devpi/stable/%2Bd/index.html&quot;&gt;devpi工具&lt;/a&gt;相比其他pypi源工具，有如下特点：&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;节省硬盘&lt;/strong&gt;：不必完全同步下来公开源的所有包，仅在第一次pip安装时从公开源下载和缓存。&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;支持上传接口文档&lt;/strong&gt;：上传自己开发pip库时，可以把接口文档也上传到devpi。&lt;/li&gt;
&lt;/ol&gt;
&lt;/li&gt;
&lt;li&gt;本项目旨在用docker容器启动devpi服务。&lt;/li&gt;&lt;/ul&gt;</summary>
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="docker" scheme="https://lowin.li/tags/docker/"/>
    
    <category term="pip" scheme="https://lowin.li/tags/pip/"/>
    
    <category term="devpi" scheme="https://lowin.li/tags/devpi/"/>
    
  </entry>
  
  <entry>
    <title>DataMeasurementsTool介绍</title>
    <link href="https://lowin.li/2022/02/05/data-measurements-tool-jie-shao/"/>
    <id>https://lowin.li/2022/02/05/data-measurements-tool-jie-shao/</id>
    <published>2022-02-04T16:00:00.000Z</published>
    <updated>2023-10-06T11:29:49.000Z</updated>
    
    
    <summary type="html">&lt;h2 id=&quot;资源&quot;&gt;&lt;a href=&quot;#资源&quot; class=&quot;headerlink&quot; title=&quot;资源&quot;&gt;&lt;/a&gt;资源&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;翻译自 &lt;a href=&quot;https://huggingface.co/blog/data-measurements-tool&quot;&gt;Huggingface Blog&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href=&quot;https://huggingface.co/spaces/huggingface/data-measurements-tool&quot;&gt;在线工具&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href=&quot;https://github.com/huggingface/data-measurements-tool&quot;&gt;GitHub&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id=&quot;引子&quot;&gt;&lt;a href=&quot;#引子&quot; class=&quot;headerlink&quot; title=&quot;引子&quot;&gt;&lt;/a&gt;引子&lt;/h2&gt;&lt;p&gt;随着机器学习数据集统一平台的快速发展(&lt;a href=&quot;https://arxiv.org/abs/2109.02846&quot;&gt;Lhoest et al. 2021&lt;/a&gt;)，HuggingFace&lt;a href=&quot;https://huggingface.co/huggingface&quot;&gt;团队&lt;/a&gt;开始探索如何管理数据集文档(&lt;a href=&quot;https://arxiv.org/abs/2108.07374&quot;&gt;McMillan-Major et al., 2021&lt;/a&gt;)。文档是认识数据集必要的第一步，通过文档我们知道如何统计和查看这份数据集，动态观察数据集的不同角度。&lt;/p&gt;
&lt;p&gt;在这里，我们介绍一个开源Python库和零代码界面，名为&lt;a href=&quot;https://huggingface.co/spaces/huggingface/data-measurements-tool&quot;&gt;Data Measurements Tool&lt;/a&gt;。通过&lt;a href=&quot;https://huggingface.co/datasets&quot;&gt;Dataset&lt;/a&gt;和&lt;a href=&quot;https://huggingface.co/spaces/launch&quot;&gt;Spaces&lt;/a&gt;社区，搭配&lt;a href=&quot;https://streamlit.io/&quot;&gt;Streamlit tool&lt;/a&gt;工具，它可以用来帮助理解、构建、洞察和比较数据集。&lt;/p&gt;</summary>
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="transformers" scheme="https://lowin.li/tags/transformers/"/>
    
    <category term="数据集治理" scheme="https://lowin.li/tags/%E6%95%B0%E6%8D%AE%E9%9B%86%E6%B2%BB%E7%90%86/"/>
    
    <category term="nlp" scheme="https://lowin.li/tags/nlp/"/>
    
  </entry>
  
  <entry>
    <title>bigbird长文本预训练模型介绍</title>
    <link href="https://lowin.li/2021/12/12/bigbird-chang-wen-ben-yu-xun-lian-mo-xing-jie-shao/"/>
    <id>https://lowin.li/2021/12/12/bigbird-chang-wen-ben-yu-xun-lian-mo-xing-jie-shao/</id>
    <published>2021-12-11T16:00:00.000Z</published>
    <updated>2023-10-06T11:29:49.000Z</updated>
    
    
    <summary type="html">&lt;ul&gt;
&lt;li&gt;本博客翻译自&lt;a href=&quot;https://huggingface.co/blog/big-bird&quot;&gt;huggingface blog&lt;/a&gt;。&lt;/li&gt;
&lt;li&gt;文末有惊喜&lt;/li&gt;
&lt;/ul&gt;
&lt;h3 id=&quot;前言&quot;&gt;&lt;a href=&quot;#前言&quot; class=&quot;headerlink&quot; title=&quot;前言&quot;&gt;&lt;/a&gt;前言&lt;/h3&gt;&lt;p&gt;基于Transformer的模型已经被证明了在许多NLP任务中的价值，但这类模型的时间复杂度、内存使用复杂度都是$n^2$（n为序列长度），因此当序列长度超过常规的512时，模型对算力的要求将会大幅提高。最近的一些文章&lt;code&gt;Longformer&lt;/code&gt;, &lt;code&gt;Performer&lt;/code&gt;, &lt;code&gt;Reformer&lt;/code&gt;, &lt;code&gt;Clustered attention&lt;/code&gt;都试图通过近似全主力机制改善该问题。例如这个&lt;a href=&quot;https://huggingface.co/blog/long-range-transformers&quot;&gt;帖子&lt;/a&gt;就是介绍这些模型的。&lt;br&gt;&lt;code&gt;BigBird&lt;/code&gt;&lt;a href=&quot;https://arxiv.org/abs/2007.14062&quot;&gt;论文&lt;/a&gt;是处理这类问题最新模型的其中之一，它使用&lt;code&gt;block sparse attention&lt;/code&gt;替换了原类似Bert一样的全注意力机制，在与BERT一样的计算力情况下，可以处理的序列长度达到4096。它已经在很多长文本序列的任务上达到SOTA效果，例如长文本摘要、长文本问答。&lt;br&gt;BigBird RoBERTa模型现在已经可以在Transformers仓库中使用。这篇博客的目的是为了让读者深入理解big bird的运行机制，快速使用Transformers仓库上手BigBird模型。但在开始之前，我们需要知道，&lt;strong&gt;BigBird&lt;/strong&gt;的注意力机制是一个近似&lt;strong&gt;BERT&lt;/strong&gt;的全注意力机制，因此它不是说比&lt;strong&gt;BERT&lt;/strong&gt;的注意力机制效果更好，而是运行效率更高。&lt;strong&gt;BERT&lt;/strong&gt;的注意力机制的存储与序列长度是二次方关系，在长文本情况下的存储需求就已经开始令人难以忍受，而&lt;strong&gt;BigBird&lt;/strong&gt;的&lt;code&gt;block sparse attention&lt;/code&gt;就是为了解决这个问题。也就是说，在$\infty$长度序列上，计算$&amp;amp;\ \infty$次时，我们应该把BERT的全注意力机制换成&lt;strong&gt;block sparse attention&lt;/strong&gt;。&lt;br&gt;如果你想知道，为什么在计算长序列时，我们需要更多的算力，这篇博客正好适合你。&lt;/p&gt;</summary>
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="transformers" scheme="https://lowin.li/tags/transformers/"/>
    
    <category term="nlp" scheme="https://lowin.li/tags/nlp/"/>
    
    <category term="bigbird" scheme="https://lowin.li/tags/bigbird/"/>
    
  </entry>
  
  <entry>
    <title>Transformers仓库做语言生成的解码方法介绍</title>
    <link href="https://lowin.li/2021/11/08/transformers-cang-ku-zuo-yu-yan-sheng-cheng-de-jie-ma-fang-fa-jie-shao/"/>
    <id>https://lowin.li/2021/11/08/transformers-cang-ku-zuo-yu-yan-sheng-cheng-de-jie-ma-fang-fa-jie-shao/</id>
    <published>2021-11-08T13:00:00.000Z</published>
    <updated>2023-10-06T11:29:49.000Z</updated>
    
    
    <summary type="html">&lt;ul&gt;
&lt;li&gt;本博客翻译自&lt;a href=&quot;https://huggingface.co/blog/how-to-generate&quot;&gt;huggingface blog&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id=&quot;简介&quot;&gt;&lt;a href=&quot;#简介&quot; class=&quot;headerlink&quot; title=&quot;简介&quot;&gt;&lt;/a&gt;简介&lt;/h2&gt;&lt;p&gt;最近几年，以OpenAI公司的GPT3为代表，基于transformer结构的大模型都已经开始在上百万级别的网页上面训练。因此大家对开放领域语言生成的期待值也越来越高。开放领域的条件语言生成效果也日新月异，例如&lt;a href=&quot;https://openai.com/blog/better-language-models/#samples&quot;&gt;GPT2&lt;/a&gt;、&lt;a href=&quot;https://amanrusia.medium.com/xlnet-speaks-comparison-to-gpt-2-ea1a4e9ba39e&quot;&gt;XLNet&lt;/a&gt;、&lt;a href=&quot;https://blog.einstein.ai/introducing-a-conditional-transformer-language-model-for-controllable-generation/&quot;&gt;CTRL&lt;/a&gt;。除了transformers结构和海量的无监督预训练数据，更好的解码方法也在其中扮演了重要角色。&lt;/p&gt;
&lt;p&gt;这篇博客简要回顾了多种解码策略，帮助你用transformers库实现他们。&lt;/p&gt;</summary>
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="transformers" scheme="https://lowin.li/tags/transformers/"/>
    
    <category term="nlp" scheme="https://lowin.li/tags/nlp/"/>
    
    <category term="nlg" scheme="https://lowin.li/tags/nlg/"/>
    
    <category term="decoder" scheme="https://lowin.li/tags/decoder/"/>
    
  </entry>
  
  <entry>
    <title>谁说torchtext不能做多标签任务</title>
    <link href="https://lowin.li/2021/10/24/shui-shuo-torchtext-bu-neng-zuo-duo-biao-qian-ren-wu/"/>
    <id>https://lowin.li/2021/10/24/shui-shuo-torchtext-bu-neng-zuo-duo-biao-qian-ren-wu/</id>
    <published>2021-10-24T14:00:00.000Z</published>
    <updated>2022-05-04T02:28:04.000Z</updated>
    
    
    <summary type="html">&lt;h2 id=&quot;背景&quot;&gt;&lt;a href=&quot;#背景&quot; class=&quot;headerlink&quot; title=&quot;背景&quot;&gt;&lt;/a&gt;背景&lt;/h2&gt;&lt;p&gt;最近刷到一篇&lt;a href=&quot;https://blog.csdn.net/weixin_33626609/article/details/112503080&quot;&gt;博客&lt;/a&gt;，吐槽&lt;code&gt;torchtext&lt;/code&gt;不能做多标签任务，特来为&lt;code&gt;torchtext&lt;/code&gt;鸣不平，看好，我要用&lt;code&gt;torchtext&lt;/code&gt;做多标签任务了。&lt;/p&gt;
&lt;h2 id=&quot;简要&quot;&gt;&lt;a href=&quot;#简要&quot; class=&quot;headerlink&quot; title=&quot;简要&quot;&gt;&lt;/a&gt;简要&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;&lt;a href=&quot;#%E8%A7%A3%E8%AF%BB&quot;&gt;解读&lt;/a&gt;&lt;ul&gt;
&lt;li&gt;torchtext库，做多标签任务&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;&lt;a href=&quot;#%E5%AE%9E%E8%B7%B5&quot;&gt;实践&lt;/a&gt;&lt;ul&gt;
&lt;li&gt;textcnn模型，跑&lt;a href=&quot;https://aistudio.baidu.com/aistudio/competition/detail/32/0/introduction&quot;&gt;百度事件多标签比赛&lt;/a&gt;，验证集准确率accuracy达到&lt;code&gt;86%&lt;/code&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;&lt;a href=&quot;#%E8%BF%90%E8%A1%8C&quot;&gt;运行&lt;/a&gt;&lt;ul&gt;
&lt;li&gt;&lt;code&gt;github&lt;/code&gt;的&lt;code&gt;action&lt;/code&gt;中，完成全程训练、批测，结果报告通过&lt;code&gt;cml工具&lt;/code&gt;发送至commit评论&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;/ul&gt;</summary>
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="nlp" scheme="https://lowin.li/tags/nlp/"/>
    
    <category term="github" scheme="https://lowin.li/tags/github/"/>
    
    <category term="onnx" scheme="https://lowin.li/tags/onnx/"/>
    
    <category term="cml" scheme="https://lowin.li/tags/cml/"/>
    
    <category term="mlops" scheme="https://lowin.li/tags/mlops/"/>
    
    <category term="多标签分类" scheme="https://lowin.li/tags/%E5%A4%9A%E6%A0%87%E7%AD%BE%E5%88%86%E7%B1%BB/"/>
    
  </entry>
  
  <entry>
    <title>转载：人工智能能否实现？</title>
    <link href="https://lowin.li/2021/10/18/zhuan-zai-ren-gong-zhi-neng-neng-fou-shi-xian/"/>
    <id>https://lowin.li/2021/10/18/zhuan-zai-ren-gong-zhi-neng-neng-fou-shi-xian/</id>
    <published>2021-10-17T16:00:00.000Z</published>
    <updated>2021-10-19T02:28:00.000Z</updated>
    
    
    <summary type="html">&lt;ul&gt;
&lt;li&gt;以下通篇转载自&lt;br&gt;&lt;a href=&quot;http://fancyerii.github.io/2019/03/14/philosophy/&quot;&gt;http://fancyerii.github.io/2019/03/14/philosophy/&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;hr&gt;
&lt;p&gt;本文讨论人工智能是否可以实现这个哲学问题。本文是《深度学习理论与实战：提高篇》的一章，更多内容请点击深度学习理论与实战：提高篇。&lt;br&gt;转载请联系作者(fancyerii at gmail dot com)！&lt;/p&gt;</summary>
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="nlp" scheme="https://lowin.li/tags/nlp/"/>
    
    <category term="哲学" scheme="https://lowin.li/tags/%E5%93%B2%E5%AD%A6/"/>
    
  </entry>
  
  <entry>
    <title>分享CML工具在github上的一个原创例子</title>
    <link href="https://lowin.li/2021/10/15/fen-xiang-cml-gong-ju-zai-github-shang-de-yi-ge-yuan-chuang-li-zi/"/>
    <id>https://lowin.li/2021/10/15/fen-xiang-cml-gong-ju-zai-github-shang-de-yi-ge-yuan-chuang-li-zi/</id>
    <published>2021-10-15T14:00:00.000Z</published>
    <updated>2022-05-04T02:29:58.000Z</updated>
    
    
    <summary type="html">&lt;h2 id=&quot;标签&quot;&gt;&lt;a href=&quot;#标签&quot; class=&quot;headerlink&quot; title=&quot;标签&quot;&gt;&lt;/a&gt;标签&lt;/h2&gt;&lt;p&gt;&lt;strong&gt;MLOPS&lt;/strong&gt;,&lt;strong&gt;CML&lt;/strong&gt;,&lt;strong&gt;ONNX&lt;/strong&gt;,&lt;strong&gt;textcnn&lt;/strong&gt;,&lt;strong&gt;CLUE&lt;/strong&gt;,&lt;strong&gt;Continuous&lt;/strong&gt;&lt;/p&gt;
&lt;h2 id=&quot;简要&quot;&gt;&lt;a href=&quot;#简要&quot; class=&quot;headerlink&quot; title=&quot;简要&quot;&gt;&lt;/a&gt;简要&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;开源一个使用&lt;code&gt;CML&lt;/code&gt;工具的原创例子。&lt;/li&gt;
&lt;li&gt;在github的&lt;code&gt;actions&lt;/code&gt;中，训练和批测&lt;code&gt;iflytek&lt;/code&gt;数据集，批测准确率&lt;code&gt;55%&lt;/code&gt;，onnx加速后，在github的action分配的资源中，单核cpu单条预测&lt;code&gt;2-4ms&lt;/code&gt;。&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;&lt;a href=&quot;https://github.com/LowinLi/cml4textcnn&quot;&gt;开源地址&lt;/a&gt;&lt;/strong&gt;&lt;/p&gt;</summary>
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="nlp" scheme="https://lowin.li/tags/nlp/"/>
    
    <category term="github" scheme="https://lowin.li/tags/github/"/>
    
    <category term="onnx" scheme="https://lowin.li/tags/onnx/"/>
    
    <category term="cml" scheme="https://lowin.li/tags/cml/"/>
    
    <category term="mlops" scheme="https://lowin.li/tags/mlops/"/>
    
    <category term="CLUE" scheme="https://lowin.li/tags/CLUE/"/>
    
  </entry>
  
  <entry>
    <title>.li域名注册教程</title>
    <link href="https://lowin.li/2021/09/25/li-yu-ming-zhu-ce-jiao-cheng/"/>
    <id>https://lowin.li/2021/09/25/li-yu-ming-zhu-ce-jiao-cheng/</id>
    <published>2021-09-25T10:25:17.000Z</published>
    <updated>2023-10-06T11:29:49.000Z</updated>
    
    
    <summary type="html">&lt;h3 id=&quot;简要&quot;&gt;&lt;a href=&quot;#简要&quot; class=&quot;headerlink&quot; title=&quot;简要&quot;&gt;&lt;/a&gt;简要&lt;/h3&gt;&lt;p&gt;本文记录了在列支敦士登公国注册.li域名，用于个人博客的踩坑过程，仅供参考。&lt;/p&gt;</summary>
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="分享" scheme="https://lowin.li/tags/%E5%88%86%E4%BA%AB/"/>
    
    <category term="博客" scheme="https://lowin.li/tags/%E5%8D%9A%E5%AE%A2/"/>
    
  </entry>
  
  <entry>
    <title>Transformers仓库解读之一DataCollator</title>
    <link href="https://lowin.li/2021/09/25/transformers-yi-datacollator/"/>
    <id>https://lowin.li/2021/09/25/transformers-yi-datacollator/</id>
    <published>2021-09-24T16:00:00.000Z</published>
    <updated>2023-10-06T11:29:49.000Z</updated>
    
    
    <summary type="html">&lt;h3 id=&quot;简要&quot;&gt;&lt;a href=&quot;#简要&quot; class=&quot;headerlink&quot; title=&quot;简要&quot;&gt;&lt;/a&gt;简要&lt;/h3&gt;&lt;p&gt;上接&lt;a href=&quot;https://lowin.li/2021/09/18/transformers%E5%BA%8F/&quot;&gt;Transformers仓库解读之序&lt;/a&gt;,对transformers库中的DataCollator的子类进行调用介绍&lt;/p&gt;</summary>
    
    
    
    <category term="技术" scheme="https://lowin.li/categories/%E6%8A%80%E6%9C%AF/"/>
    
    
    <category term="transformers" scheme="https://lowin.li/tags/transformers/"/>
    
    <category term="分享" scheme="https://lowin.li/tags/%E5%88%86%E4%BA%AB/"/>
    
  </entry>
  
</feed>
