
  <rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
    <channel>
      <title>NLP Journey</title>
      <link>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog</link>
      <description>AI 算法工程师的公开研究笔记：记录大模型、系统工程与可复现实验，持续把复杂技术转化为可落地的产品。</description>
      <language>zh</language>
      <managingEditor>1395291968@qq.com (qiaoshilei)</managingEditor>
      <webMaster>1395291968@qq.com (qiaoshilei)</webMaster>
      <lastBuildDate>Thu, 06 Aug 2026 00:00:00 GMT</lastBuildDate>
      <atom:link href="https://tailwind-nextjs-starter-blog-six-theta.vercel.app/feed.xml" rel="self" type="application/rss+xml"/>
      
  <item>
    <guid>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog/2026-08-06-production-llm-subtitle-translation-en</guid>
    <title>Production LLM Subtitle Translation Is Much More Than a Good Prompt</title>
    <link>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog/2026-08-06-production-llm-subtitle-translation-en</link>
    <description>Lessons from three months of building a multilingual subtitle translation workflow across 16 languages: reliable quality requires translation, validation, targeted revision, human review, and a feedback loop.</description>
    <pubDate>Thu, 06 Aug 2026 00:00:00 GMT</pubDate>
    <author>1395291968@qq.com (qiaoshilei)</author>
    <category>LLM</category><category>Subtitle Translation</category><category>Multilingual</category><category>Prompt Engineering</category><category>Quality Evaluation</category><category>AI Engineering</category>
  </item>

  <item>
    <guid>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog/2026-08-04-production-llm-subtitle-translation</guid>
    <title>做了三个月 16 语种字幕翻译后，我发现生产级 LLM 翻译远不只是 Prompt</title>
    <link>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog/2026-08-04-production-llm-subtitle-translation</link>
    <description>结合三个月、16 种语言的视频字幕翻译实践，梳理生产级 LLM 翻译从初译、校验、定向修正到人工反馈与数据回流的完整质量体系。</description>
    <pubDate>Tue, 04 Aug 2026 00:00:00 GMT</pubDate>
    <author>1395291968@qq.com (qiaoshilei)</author>
    <category>LLM</category><category>字幕翻译</category><category>多语言</category><category>Prompt 工程</category><category>质量评估</category><category>AI 工程</category>
  </item>

  <item>
    <guid>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog/webnovel-title-localization-lab</guid>
    <title>WebNovel Title Localization Lab：把中文网文标题翻成真正适合英语读者的名字</title>
    <link>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog/webnovel-title-localization-lab</link>
    <description>一个仍在持续开发的标题本地化研究项目：先生成冻结候选集，再以可复现的八维评估比较规则、LLM 与学习排序方法。</description>
    <pubDate>Mon, 03 Aug 2026 00:00:00 GMT</pubDate>
    <author>1395291968@qq.com (qiaoshilei)</author>
    <category>NLP</category><category>LLM</category><category>Ranking</category><category>FastAPI</category><category>实验项目</category>
  </item>

  <item>
    <guid>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog/project-parliament</guid>
    <title>Project Parliament：让多个模型为开源方向辩论</title>
    <link>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog/project-parliament</link>
    <description>一个用两个周末完成的多模型工作流实验：让 AI 先提出候选开源方向，再通过结构化辩论收敛出真正适合开发者的行动路线。</description>
    <pubDate>Sun, 12 Apr 2026 00:00:00 GMT</pubDate>
    <author>1395291968@qq.com (qiaoshilei)</author>
    <category>AI Agents</category><category>Multi-model</category><category>Open Source</category><category>FastAPI</category><category>实验项目</category>
  </item>

  <item>
    <guid>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog/funasr-deployment-pitfall-guide-cpu-thread-explosion-memory-overflow</guid>
    <title>FunASR部署避坑指南：多进程下CPU线程爆炸导致内存溢出卡死的排查与解决</title>
    <link>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog/funasr-deployment-pitfall-guide-cpu-thread-explosion-memory-overflow</link>
    <description>本文分享了在部署FunASR语音识别服务时，遇到多进程下CPU线程爆炸导致内存溢出卡死的问题排查与解决方案。</description>
    <pubDate>Sun, 11 Jan 2026 00:00:00 GMT</pubDate>
    <author>1395291968@qq.com (qiaoshilei)</author>
    <category>FunASR</category><category>多进程</category><category>CPU优化</category><category>性能调优</category><category>深度学习部署</category>
  </item>

  <item>
    <guid>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog/2025-12-14-funasr-performance-optimization-shared-memory</guid>
    <title>FunASR 极致性能优化：从 Queue 到 SharedMemory 与全链路并发控制</title>
    <link>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog/2025-12-14-funasr-performance-optimization-shared-memory</link>
    <description>在上一篇多进程部署的基础上，通过引入 SharedMemory 零拷贝、AnyIO 流式管道以及精细的 Asyncio 锁机制，成功将 50 路并发下的 RTF 从 0.61 压至 0.23，实现吞吐量翻倍。</description>
    <pubDate>Sun, 14 Dec 2025 00:00:00 GMT</pubDate>
    <author>1395291968@qq.com (qiaoshilei)</author>
    <category>Python</category><category>SharedMemory</category><category>Asyncio</category><category>FunASR</category><category>性能优化</category><category>并发编程</category><category>系统工程与部署能力</category>
  </item>

  <item>
    <guid>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog/recommendation-system-analysis-from-recall-to-ranking-optimization</guid>
    <title>推荐系统全解析：从召回到精排，如何逐层优化与提升效果？</title>
    <link>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog/recommendation-system-analysis-from-recall-to-ranking-optimization</link>
    
    <pubDate>Sun, 07 Dec 2025 00:00:00 GMT</pubDate>
    <author>1395291968@qq.com (qiaoshilei)</author>
    <category>推荐系统</category><category>召回层</category><category>粗排序</category><category>精排序</category><category>CTR模型</category><category>产品导向与场景思维</category>
  </item>

  <item>
    <guid>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog/python-multiprocessing-multi-model-deploy</guid>
    <title>单 Python 文件搞定多模型部署：一次 multiprocessing 实战记录</title>
    <link>https://tailwind-nextjs-starter-blog-six-theta.vercel.app/blog/python-multiprocessing-multi-model-deploy</link>
    <description>在一个 Python 文件中，通过 multiprocessing 同时跑多个模型实例，实现多 GPU 绑定与并发任务调度的完整实战记录。</description>
    <pubDate>Tue, 25 Nov 2025 00:00:00 GMT</pubDate>
    <author>1395291968@qq.com (qiaoshilei)</author>
    <category>Python</category><category>multiprocessing</category><category>多进程</category><category>模型部署</category><category>ASR</category><category>系统工程与部署能力</category>
  </item>

    </channel>
  </rss>
