<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
<channel>
  <title>AI 武林：PyTorch</title>
  <link>https://aiwulin.itsmygo.uk/tools/pytorch/</link>
  <description>AI 武林收錄的內容裡，最新提到 PyTorch 的 30 筆，每筆附中文摘要。</description>
  <language>zh-TW</language>
  <lastBuildDate>Thu, 08 Oct 2026 17:22:41 GMT</lastBuildDate>
  <atom:link href="https://aiwulin.itsmygo.uk/tools/pytorch/rss.xml" rel="self" type="application/rss+xml"/>
  <item>
    <title>What Is an Inference Engine, Anyway? — Charles Frye, Modal</title>
    <link>https://aiwulin.itsmygo.uk/c/9380785f9e/</link>
    <guid isPermaLink="false">aiwulin-9380785f9e</guid>
    <pubDate>Tue, 06 Oct 2026 00:00:00 +0800</pubDate>
    <dc:creator>AI Engineer</dc:creator>
    <description>&lt;p&gt;Charles Frye 解析推理引擎如何處理從 API 請求到模型輸出的完整流程，涵蓋分詞、排程與 GPU 執行等關鍵環節。影片說明不同工作負載（如聊天機器人、背景代理）對延遲與吞吐量的影響，並探討 KV 快取與推測解碼等最佳化技術。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=woIYJYd_etI&quot;&gt;https://www.youtube.com/watch?v=woIYJYd_etI&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>推論與部署</category>
  </item>
  <item>
    <title>Framework Desktop 擴充 NVIDIA GPU：用 OCuLink 外接 RTX 5070 Ti</title>
    <link>https://aiwulin.itsmygo.uk/c/bfaf6fbe4d/</link>
    <guid isPermaLink="false">aiwulin-bfaf6fbe4d</guid>
    <pubDate>Sun, 04 Oct 2026 00:00:00 +0800</pubDate>
    <dc:creator>ihower（張文鈿）</dc:creator>
    <description>&lt;p&gt;分享如何透過 OCuLink 技術，將 NVIDIA RTX 5070 Ti 外接到 Framework Desktop，解決 AMD 平台跑 PyTorch 專案的 CUDA 支援問題。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://ihower.tw/blog/13789-framework-desktop-nvidia-gpu-oculink&quot;&gt;https://ihower.tw/blog/13789-framework-desktop-nvidia-gpu-oculink&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>AI 晶片與硬體</category>
  </item>
  <item>
    <title>Build A Reasoning Model From Scratch 5: Inference Scaling 2 (Logprob Scoring, Self-Refinement)</title>
    <link>https://aiwulin.itsmygo.uk/c/06606e0223/</link>
    <guid isPermaLink="false">aiwulin-06606e0223</guid>
    <pubDate>Sat, 26 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>Sebastian Raschka</dc:creator>
    <description>&lt;p&gt;介紹推理模型推論時的評分函式與自我改進技術，包含使用對數機率分數解決平手情況及實作自我修正迴圈。觀眾可學習如何計算 token 機率、評估模型答案並最佳化推理過程。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=TVMyOJ_3Gxo&quot;&gt;https://www.youtube.com/watch?v=TVMyOJ_3Gxo&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>推論與部署</category>
    <category>研究前沿</category>
    <category>從零打造語言模型</category>
  </item>
  <item>
    <title>Transformers now runs llama.cpp quants</title>
    <link>https://aiwulin.itsmygo.uk/c/98acfe91fd/</link>
    <guid isPermaLink="false">aiwulin-98acfe91fd</guid>
    <pubDate>Tue, 22 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>Hugging Face Blog</dc:creator>
    <description>&lt;p&gt;Hugging Face 將 llama.cpp 的 GGUF 量化格式整合進 transformers，透過 ggml 核心庫讓 Apple Silicon 裝置能以 Python 高效執行本地 AI 模型。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://huggingface.co/blog/transformers-llama-cpp-quants&quot;&gt;https://huggingface.co/blog/transformers-llama-cpp-quants&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>本機跑模型</category>
    <category>開源模型</category>
    <category>Transformer 原理</category>
  </item>
  <item>
    <title>Large clusters for small models — Daniel Svonava, Superlinked</title>
    <link>https://aiwulin.itsmygo.uk/c/594efa1a7a/</link>
    <guid isPermaLink="false">aiwulin-594efa1a7a</guid>
    <pubDate>Sun, 20 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>AI Engineer</dc:creator>
    <description>&lt;p&gt;Daniel Svonava 分享如何讓小型開源模型在生產環境中發揮 frontier 級效能，透過任務分割與專用模型組合降低成本。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=g4SsanB0gMc&quot;&gt;https://www.youtube.com/watch?v=g4SsanB0gMc&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>開源模型</category>
    <category>推論與部署</category>
  </item>
  <item>
    <title>1028: The Chip Built for Agentic AI Inference, with SambaNova's Anton McGonnell</title>
    <link>https://aiwulin.itsmygo.uk/c/7b0ae0fcad/</link>
    <guid isPermaLink="false">aiwulin-7b0ae0fcad</guid>
    <pubDate>Fri, 18 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>Super Data Science</dc:creator>
    <description>&lt;p&gt;探討代理型 AI 如何改變推理工作負載，指出傳統 GPU 在生成階段面臨記憶體瓶頸。SambaNova 的 RDU 架構將整個模型空間化佈局，實現線性擴展與高並行，解決速度與吞吐量取捨問題，並可部署於既有資料中心。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.podtrac.com/pts/redirect.mp3/chrt.fm/track/E581B9/arttrk.com/p/VI4CS/pscrb.fm/rss/p/traffic.megaphone.fm/SUPERDATASCIENCEPTYLTD8072554290.mp3&quot;&gt;https://www.podtrac.com/pts/redirect.mp3/chrt.fm/track/E581B9/arttrk.com/p/VI4CS/pscrb.fm/rss/p/traffic.megaphone.fm/SUPERDATASCIENCEPTYLTD8072554290.mp3&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>推論與部署</category>
    <category>AI Agent 基礎</category>
    <category>AI 晶片與硬體</category>
  </item>
  <item>
    <title>Hands-On Evolution of Deep Learning – Geoffrey Hinton’s AI Legacy</title>
    <link>https://aiwulin.itsmygo.uk/c/7e0c609efc/</link>
    <guid isPermaLink="false">aiwulin-7e0c609efc</guid>
    <pubDate>Thu, 17 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>freeCodeCamp.org</dc:creator>
    <description>&lt;p&gt;透過重現 Geoffrey Hinton 的經典論文，一步步用 PyTorch 實作從 Boltzmann Machine 到 Deep Belief Networks 的演算法。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=BMYvfVk8Ar0&quot;&gt;https://www.youtube.com/watch?v=BMYvfVk8Ar0&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>深度學習</category>
  </item>
  <item>
    <title>Essential Skills for Becoming an AI Engineer: RAG, AI Agents, &amp; More</title>
    <link>https://aiwulin.itsmygo.uk/c/a03fc34c23/</link>
    <guid isPermaLink="false">aiwulin-a03fc34c23</guid>
    <pubDate>Sun, 13 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>IBM Technology</dc:creator>
    <description>&lt;p&gt;介紹 AI 工程師的職涯路徑與所需技能，區分機器學習研究員與 AI 工程師的差異。內容涵蓋從 Python、Git、Linux 等基礎，到 RAG、向量搜尋與 AI Agent，最後是 Kubernetes 部署與可觀測性，並提供實作專案…&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=XN3xNJvWXsc&quot;&gt;https://www.youtube.com/watch?v=XN3xNJvWXsc&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>AI 輔助軟體工程</category>
    <category>AI Agent 基礎</category>
    <category>Agent Skills 與 Harness</category>
  </item>
  <item>
    <title>Tuning GPU Performance with AI Agents | AMD’s Anush Elangovan on ROCm 10</title>
    <link>https://aiwulin.itsmygo.uk/c/68df319d73/</link>
    <guid isPermaLink="false">aiwulin-68df319d73</guid>
    <pubDate>Wed, 09 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>Chain of Thought</dc:creator>
    <description>&lt;p&gt;AMD 企業 VP Anush Elangovan 分享 ROCm 10 如何透過 AI Agent 自動安裝、除錯與最佳化 GPU 工作負載。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://chainofthought.show/podcast/72-tuning-gpu-performance-with-ai-agents-amds-anush-elangovan-on-rocm-10&quot;&gt;https://chainofthought.show/podcast/72-tuning-gpu-performance-with-ai-agents-amds-anush-elangovan-on-rocm-10&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>AI 晶片與硬體</category>
    <category>AI Agent 基礎</category>
    <category>Coding Agent</category>
  </item>
  <item>
    <title>Build A Reasoning Model From Scratch 2: Loading a Base Model, Text Generation, and KV Caching</title>
    <link>https://aiwulin.itsmygo.uk/c/b40f6ad30e/</link>
    <guid isPermaLink="false">aiwulin-b40f6ad30e</guid>
    <pubDate>Sun, 06 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>Sebastian Raschka</dc:creator>
    <description>&lt;p&gt;介紹如何從零開始載入預訓練語言模型，解釋大型語言模型的文字生成機制，並說明使用 KV 快取提升效率的方法。觀眾可以學習基礎的模型載入與生成原理，為後續深入學習打下基礎。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=BJua0yjO5dk&quot;&gt;https://www.youtube.com/watch?v=BJua0yjO5dk&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>大型語言模型入門</category>
    <category>從零打造語言模型</category>
    <category>研究前沿</category>
  </item>
  <item>
    <title>Build A Reasoning Model From Scratch 1: Motivation &amp; Code Setup</title>
    <link>https://aiwulin.itsmygo.uk/c/8bee8c9872/</link>
    <guid isPermaLink="false">aiwulin-8bee8c9872</guid>
    <pubDate>Sun, 30 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>Sebastian Raschka</dc:creator>
    <description>&lt;p&gt;介紹從零開始建立推理模型的動機與程式環境設定，涵蓋從傳統大型語言模型到推理模型的差異，並指導使用 uv 安裝 PyTorch 依賴。觀眾可學習如何配置開發環境並理解自製模型的基礎概念。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=Kh9mqTzjuEQ&quot;&gt;https://www.youtube.com/watch?v=Kh9mqTzjuEQ&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>從零打造語言模型</category>
    <category>研究前沿</category>
  </item>
  <item>
    <title>🔬“We have foundation models for language, not for physics” — Anima Anandkumar, Bren Professor of Computing</title>
    <link>https://aiwulin.itsmygo.uk/c/48713ac2d4/</link>
    <guid isPermaLink="false">aiwulin-48713ac2d4</guid>
    <pubDate>Wed, 26 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>Latent Space</dc:creator>
    <description>&lt;p&gt;Anima Anandkumar 分享開發 FourCastNet 的經驗，用神經運算子將 AI 應用於天氣預測，在消費級 GPU 上實現比傳統模擬快數萬倍且準確的短期天氣預報。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.latent.space/p/anima&quot;&gt;https://www.latent.space/p/anima&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>AI 與科學研究</category>
    <category>AI 晶片與硬體</category>
  </item>
  <item>
    <title>Building a distributed training framework from first principles</title>
    <link>https://aiwulin.itsmygo.uk/c/4d698436a8/</link>
    <guid isPermaLink="false">aiwulin-4d698436a8</guid>
    <pubDate>Tue, 18 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>Umar Jamil</dc:creator>
    <description>&lt;p&gt;從第一原理出發，使用 PyTorch 建立分散式訓練框架，涵蓋數學原理與並行技術實作。觀眾能學習如何將管道、資料、張量與專家並行整合成單一工作系統。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=XoGvCBRnwLs&quot;&gt;https://www.youtube.com/watch?v=XoGvCBRnwLs&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>開發框架</category>
  </item>
  <item>
    <title>Record, train, and deploy from one place with Strands Agents, LeRobot, and Hugging Face Storage Buckets</title>
    <link>https://aiwulin.itsmygo.uk/c/c1c9885561/</link>
    <guid isPermaLink="false">aiwulin-c1c9885561</guid>
    <pubDate>Fri, 14 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>Hugging Face Blog</dc:creator>
    <description>&lt;p&gt;介紹如何使用 Strands Agents、LeRobot 與 Hugging Face Storage Buckets 建立一個完整的「記錄、訓練、部署」自動化迴圈。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://huggingface.co/blog/amazon/strands-lerobot-streaming-data-loop&quot;&gt;https://huggingface.co/blog/amazon/strands-lerobot-streaming-data-loop&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>AI Agent 基礎</category>
    <category>機器人與具身智慧</category>
    <category>開源模型</category>
  </item>
  <item>
    <title>Making Knowledge Distillation Cheap Enough to Run at Scale</title>
    <link>https://aiwulin.itsmygo.uk/c/90d59cd2b4/</link>
    <guid isPermaLink="false">aiwulin-90d59cd2b4</guid>
    <pubDate>Mon, 10 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>Hugging Face Blog</dc:creator>
    <description>&lt;p&gt;介紹如何透過離線頂-K 機率分佈與融合分塊KL損失，大幅降低知識蒸餾所需的視訊記憶體與成本。透過將教師模型輸出快取並分塊計算損失，可避免建立龐大矩陣，使單張GPU即可進行長上下文蒸餾，並公開相關實作程式碼。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://huggingface.co/blog/MultiverseComputingCAI/efficient-knowledge-distillation&quot;&gt;https://huggingface.co/blog/MultiverseComputingCAI/efficient-knowledge-distillation&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>量化與蒸餾</category>
  </item>
  <item>
    <title>If Developers Build on Chinese Open-Weight Models, Who Leads AI?</title>
    <link>https://aiwulin.itsmygo.uk/c/8e3c6f2044/</link>
    <guid isPermaLink="false">aiwulin-8e3c6f2044</guid>
    <pubDate>Mon, 03 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>Vanishing Gradients</dc:creator>
    <description>&lt;p&gt;Hugo Bowne-Anderson 與 Sebastian Raschka 探討開放權重模型的重要性，指出其能避免對單一廠商的依賴並促進競爭。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://hugobowne.substack.com/p/open-weight-ai-is-becoming-infrastructure&quot;&gt;https://hugobowne.substack.com/p/open-weight-ai-is-becoming-infrastructure&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>開源模型</category>
    <category>本機跑模型</category>
  </item>
  <item>
    <title>40 Trillion Tokens a Day (Yes, More Than OpenAI) | Lin Qiao, CEO of Fireworks</title>
    <link>https://aiwulin.itsmygo.uk/c/5c5eb04afa/</link>
    <guid isPermaLink="false">aiwulin-5c5eb04afa</guid>
    <pubDate>Mon, 03 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>Gradient Dissent</dc:creator>
    <description>&lt;p&gt;Lin Qiao 與 Lukas Biewald 探討 Fireworks 每日處理超過 40 兆 token 的規模，並主張未來應發展基於企業私有資料的專業模型，而非通用大模型。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://episodes.captivate.fm/episode/1a7e1d71-bca4-4edc-b9f8-70246d07ce65.mp3&quot;&gt;https://episodes.captivate.fm/episode/1a7e1d71-bca4-4edc-b9f8-70246d07ce65.mp3&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>開源模型</category>
    <category>大型語言模型入門</category>
  </item>
  <item>
    <title>Stanford CS229 Machine Learning | Spring 2026 | Lecture 8: Neural Networks 2 (Backprop)</title>
    <link>https://aiwulin.itsmygo.uk/c/826d17b0a3/</link>
    <guid isPermaLink="false">aiwulin-826d17b0a3</guid>
    <pubDate>Fri, 31 Jul 2026 00:00:00 +0800</pubDate>
    <dc:creator>Stanford Online</dc:creator>
    <description>&lt;p&gt;深入解析神經網路訓練的核心演算法後向傳播（Backpropagation），透過可微分電路的理論證明前向與後向計算時間同階。觀眾將學習如何利用鏈式法則高效計算損失函式對參數的梯度，並掌握處理矩陣乘法與激活函數的具體實作技巧。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=ne2ngVAoMG8&quot;&gt;https://www.youtube.com/watch?v=ne2ngVAoMG8&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>機器學習基礎</category>
  </item>
  <item>
    <title>【十字路口】快一点！再快一点！快到世界能实时生成｜和生数科技张金涛聊：Vidu S1、推理加速、实时交互视频【视频播客】</title>
    <link>https://aiwulin.itsmygo.uk/c/bc3728df24/</link>
    <guid isPermaLink="false">aiwulin-bc3728df24</guid>
    <pubDate>Mon, 20 Jul 2026 00:00:00 +0800</pubDate>
    <dc:creator>Koji杨远骋</dc:creator>
    <description>&lt;p&gt;訪談生數科技張金濤，深入探討 Vidu S1 實時互動影片模型的技術實現與推理加速策略。金濤分享了從 SageAttention 到 TurboDiffusion 的技術演進路徑，並解析了如何透過系統級最佳化讓生成速度超越播放速度，實現無…&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=EAbsBmcDMQE&quot;&gt;https://www.youtube.com/watch?v=EAbsBmcDMQE&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>影片生成</category>
    <category>圖像生成</category>
  </item>
  <item>
    <title>PyTorch 深度學習入門 - 簡介、安裝、快速開始 #ai #人工智慧 #機器學習</title>
    <link>https://aiwulin.itsmygo.uk/c/d0b5fd5b2e/</link>
    <guid isPermaLink="false">aiwulin-d0b5fd5b2e</guid>
    <pubDate>Thu, 09 Jul 2026 00:00:00 +0800</pubDate>
    <dc:creator>彭彭的課程</dc:creator>
    <description>&lt;p&gt;介紹 PyTorch 深度學習框架的安裝與快速開始，適合具備 Python 基礎的開發者。看完即可掌握環境設定與基本測試方法，進入深度學習實作。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=NJhTcGW7RS4&quot;&gt;https://www.youtube.com/watch?v=NJhTcGW7RS4&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>機器學習基礎</category>
  </item>
  <item>
    <title>Import AI 464: Fable writes GPU kernels; AI automation; and analog computation</title>
    <link>https://aiwulin.itsmygo.uk/c/27e5ff5b99/</link>
    <guid isPermaLink="false">aiwulin-27e5ff5b99</guid>
    <pubDate>Mon, 06 Jul 2026 00:00:00 +0800</pubDate>
    <dc:creator>Import AI（Jack Clark）</dc:creator>
    <description>&lt;p&gt;介紹 Fable 寫出高效 GPU 核碼的突破，分析 AI 自動化線上任務能力的提升，並探討 OSWORLD 2.0 benchmarks 顯示 AI 在複雜電腦操作上的進展。最後以虛構故事展望未來類比運算的應用。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://importai.substack.com/p/import-ai-464-fables-writes-gpu-kernels&quot;&gt;https://importai.substack.com/p/import-ai-464-fables-writes-gpu-kernels&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>AI 晶片與硬體</category>
    <category>工作流自動化</category>
    <category>瀏覽器與電腦操作</category>
  </item>
  <item>
    <title>Relational Foundation Models for Enterprise Data with Jure Leskovec - #768</title>
    <link>https://aiwulin.itsmygo.uk/c/a32adb713d/</link>
    <guid isPermaLink="false">aiwulin-a32adb713d</guid>
    <pubDate>Fri, 22 May 2026 00:00:00 +0800</pubDate>
    <dc:creator>The TWIML AI Podcast</dc:creator>
    <description>&lt;p&gt;訪談 Jure Leskovec 探討 AI for Science 與關係式深度學習。介紹 AI Virtual Cell 如何利用單細胞 RNA-seq 資料建立從蛋白質到患者的多尺度模型，並說明 Kumo 的 Relational…&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://twimlai.com/podcast/twimlai/relational-foundation-models-enterprise-data&quot;&gt;https://twimlai.com/podcast/twimlai/relational-foundation-models-enterprise-data&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>AI 與科學研究</category>
    <category>資料科學</category>
    <category>上下文工程</category>
  </item>
  <item>
    <title>What I Learned From Implementing LLM Architectures From Scratch (And How to Get Started)</title>
    <link>https://aiwulin.itsmygo.uk/c/23ac353119/</link>
    <guid isPermaLink="false">aiwulin-23ac353119</guid>
    <pubDate>Wed, 13 May 2026 00:00:00 +0800</pubDate>
    <dc:creator>Sebastian Raschka</dc:creator>
    <description>&lt;p&gt;Sebastian Raschka 分享從零用 Python 和 PyTorch 實作大型語言模型架構的經驗，說明實作細節如何幫助理解現代 LLM 運作原理。觀眾能學習如何閱讀模型卡片、比對參考實作，並掌握分析新架構的具體工作流。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=TXzQ7PGpO6w&quot;&gt;https://www.youtube.com/watch?v=TXzQ7PGpO6w&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>大型語言模型入門</category>
    <category>從零打造語言模型</category>
  </item>
  <item>
    <title>How to Engineer AI Inference Systems with Philip Kiely - #766</title>
    <link>https://aiwulin.itsmygo.uk/c/e6373909cc/</link>
    <guid isPermaLink="false">aiwulin-e6373909cc</guid>
    <pubDate>Fri, 01 May 2026 00:00:00 +0800</pubDate>
    <dc:creator>The TWIML AI Podcast</dc:creator>
    <description>&lt;p&gt;深入探討推論工程（Inference Engineering），解釋為何它是 AI 最關鍵且複雜的工作負載。內容涵蓋從研究到生產的快速週期、關鍵技術如量化與 KV Cache 重用的應用，以及業界從 API 呼叫到自建平台的演進路徑。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://twimlai.com/podcast/twimlai/how-engineer-ai-inference-systems&quot;&gt;https://twimlai.com/podcast/twimlai/how-engineer-ai-inference-systems&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>推論與部署</category>
  </item>
  <item>
    <title>E228｜谷歌TPU能撼动英伟达吗？前TPU工程师首次揭秘</title>
    <link>https://aiwulin.itsmygo.uk/c/3916aadb6c/</link>
    <guid isPermaLink="false">aiwulin-3916aadb6c</guid>
    <pubDate>Fri, 13 Mar 2026 00:00:00 +0800</pubDate>
    <dc:creator>硅谷101</dc:creator>
    <description>&lt;p&gt;前谷歌 TPU 工程師 Henry 深入解析 TPU 與 GPU 在架構、效能與生態上的差異，探討 TPU 如何透過客製化與供應鏈優勢挑戰輝達。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://sv101.fireside.fm/241&quot;&gt;https://sv101.fireside.fm/241&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>AI 晶片與硬體</category>
  </item>
  <item>
    <title>Kubernetes, Compliance, and Control: The Operational Backbone of AI Sovereignty</title>
    <link>https://aiwulin.itsmygo.uk/c/57848968c9/</link>
    <guid isPermaLink="false">aiwulin-57848968c9</guid>
    <pubDate>Wed, 25 Feb 2026 00:00:00 +0800</pubDate>
    <dc:creator>AI Engineering Podcast</dc:creator>
    <description>&lt;p&gt;探討企業與國家如何透過自管基礎設施與合規策略實現 AI 主權，涵蓋 Kubernetes、PyTorch 等工具在 LLM 服務中的應用。內容涵蓋從 GPU 存取難題到代理身份認證等實務挑戰，幫助讀者理解 AI 主權的技術與營運路徑。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.aiengineeringpodcast.com/ai-sovereignty-infrastructure-adaptation-episode-78&quot;&gt;https://www.aiengineeringpodcast.com/ai-sovereignty-infrastructure-adaptation-episode-78&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>AI 晶片與硬體</category>
    <category>大型語言模型入門</category>
    <category>推論與部署</category>
  </item>
  <item>
    <title>Taming Voice Complexity with Dynamic Ensembles at Modulate</title>
    <link>https://aiwulin.itsmygo.uk/c/36588b3a12/</link>
    <guid isPermaLink="false">aiwulin-36588b3a12</guid>
    <pubDate>Mon, 09 Feb 2026 00:00:00 +0800</pubDate>
    <dc:creator>AI Engineering Podcast</dc:creator>
    <description>&lt;p&gt;探討 Modulate 公司如何透過動態叢集架構解決語音 AI 的低延遲與高準確度挑戰。講者 Carter Huffman 解釋了為何單純的轉寫再轉語音流程無法捕捉語氣與情緒，並介紹了利用小型專模型組成的叢集系統，以平衡成本、延遲與多變環…&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.aiengineeringpodcast.com/ensemble-listening-models-episode-76&quot;&gt;https://www.aiengineeringpodcast.com/ensemble-listening-models-episode-76&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>語音與音訊</category>
    <category>推論與部署</category>
  </item>
  <item>
    <title>The Future of Dev Experience: Spotify’s Playbook for Organization‑Scale AI</title>
    <link>https://aiwulin.itsmygo.uk/c/9cc45d94a2/</link>
    <guid isPermaLink="false">aiwulin-9cc45d94a2</guid>
    <pubDate>Tue, 20 Jan 2026 00:00:00 +0800</pubDate>
    <dc:creator>AI Engineering Podcast</dc:creator>
    <description>&lt;p&gt;Spotify 首席架構師 Niklas Gustavsson 分享如何透過標準化與 Backstage 平台，在分散式架構中加速 AI 工具（如 Copilot、Cursor）的採用。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.aiengineeringpodcast.com/spotify-agentic-developer-experience-episode-74&quot;&gt;https://www.aiengineeringpodcast.com/spotify-agentic-developer-experience-episode-74&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>AI 編輯器與開發工具</category>
    <category>Coding Agent</category>
    <category>AI Agent 基礎</category>
  </item>
  <item>
    <title>Beyond the Chatbot: Practical Frameworks for Agentic Capabilities in SaaS</title>
    <link>https://aiwulin.itsmygo.uk/c/e477e86525/</link>
    <guid isPermaLink="false">aiwulin-e477e86525</guid>
    <pubDate>Mon, 29 Dec 2025 00:00:00 +0800</pubDate>
    <dc:creator>AI Engineering Podcast</dc:creator>
    <description>&lt;p&gt;訪談 SaaS 產品與工程主管 Preeti Shukla，探討如何在多租戶軟體中安全地加入 AI 代理能力。內容涵蓋如何平衡延遲、成本與隱私，選擇合適的模型與架構，以及透過驗證與可觀測性系統來防止「自信錯誤」。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.aiengineeringpodcast.com/adding-agentic-behavior-to-saas-episode-72&quot;&gt;https://www.aiengineeringpodcast.com/adding-agentic-behavior-to-saas-episode-72&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>AI Agent 基礎</category>
  </item>
  <item>
    <title>Scaling Agentic Inference Across Heterogeneous Compute with Zain Asgar - #757</title>
    <link>https://aiwulin.itsmygo.uk/c/bb01d71e56/</link>
    <guid isPermaLink="false">aiwulin-bb01d71e56</guid>
    <pubDate>Wed, 03 Dec 2025 00:00:00 +0800</pubDate>
    <dc:creator>The TWIML AI Podcast</dc:creator>
    <description>&lt;p&gt;探討 Gimlet Labs 如何透過異質運算架構，將 AI 工作負載分散於不同硬體以提升效能與成本效益。講者介紹了將模型拆解、使用 LLM 自動最佳化運算核心等技術，並討論了從邊緣裝置到資料中心的應用挑戰。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://twimlai.com/podcast/twimlai/scaling-agentic-inference-across-heterogeneous-compute&quot;&gt;https://twimlai.com/podcast/twimlai/scaling-agentic-inference-across-heterogeneous-compute&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>推論與部署</category>
    <category>AI 晶片與硬體</category>
  </item>
</channel>
</rss>
