<?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：vLLM</title>
  <link>https://aiwulin.itsmygo.uk/tools/vllm/</link>
  <description>AI 武林收錄的內容裡，最新提到 vLLM 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/vllm/rss.xml" rel="self" type="application/rss+xml"/>
  <item>
    <title>LLM加速及快取管理平臺LMCache多行程模式曝重大RCE漏洞</title>
    <link>https://aiwulin.itsmygo.uk/c/6b5a474fa9/</link>
    <guid isPermaLink="false">aiwulin-6b5a474fa9</guid>
    <pubDate>Thu, 08 Oct 2026 00:00:00 +0800</pubDate>
    <dc:creator>iThome</dc:creator>
    <description>&lt;p&gt;JFrog 揭露大型語言模型推論加速平台 LMCache 存在嚴重遠端程式碼執行漏洞，CVSS 評分達 9.8。該漏洞出現在多行程模式的 ZeroMQ 服務，因缺乏身分驗證且直接還原 Python 物件，讓攻擊者無需登入即可執行任意程式碼。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.ithome.com.tw/news/179511&quot;&gt;https://www.ithome.com.tw/news/179511&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>大型語言模型入門</category>
    <category>推論與部署</category>
    <category>可觀測性與 LLMOps</category>
  </item>
  <item>
    <title>Is Speculative Decoding Worth It? Profiling vLLM on NVIDIA Blackwell — Akamai</title>
    <link>https://aiwulin.itsmygo.uk/c/45f2de7758/</link>
    <guid isPermaLink="false">aiwulin-45f2de7758</guid>
    <pubDate>Wed, 07 Oct 2026 00:00:00 +0800</pubDate>
    <dc:creator>AI Engineer</dc:creator>
    <description>&lt;p&gt;Sheilah Kirui 介紹 Speculative Decoding 技術，利用小模型預測多個 token 供大模型驗證以加速推理。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=XTpyNrEgJQ4&quot;&gt;https://www.youtube.com/watch?v=XTpyNrEgJQ4&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>推論與部署</category>
  </item>
  <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>Unsloth Studio存在任意程式碼執行漏洞，檢查模型資訊時就可能觸發</title>
    <link>https://aiwulin.itsmygo.uk/c/a6d4469ec3/</link>
    <guid isPermaLink="false">aiwulin-a6d4469ec3</guid>
    <pubDate>Tue, 06 Oct 2026 00:00:00 +0800</pubDate>
    <dc:creator>iThome</dc:creator>
    <description>&lt;p&gt;Pillar Security 揭露 Unsloth Studio 存在任意程式碼執行漏洞，檢查模型資訊時即可觸發。問題源於預設啟用 trust_remote_code，攻擊者可透過自訂模型儲存庫注入惡意程式碼。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.ithome.com.tw/news/179390&quot;&gt;https://www.ithome.com.tw/news/179390&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>AI 資安</category>
  </item>
  <item>
    <title>Lessons from Generating 12 Trillion Synthetic Tokens — Bogdan Gaza, DatologyAI</title>
    <link>https://aiwulin.itsmygo.uk/c/0da6961f4e/</link>
    <guid isPermaLink="false">aiwulin-0da6961f4e</guid>
    <pubDate>Sat, 03 Oct 2026 00:00:00 +0800</pubDate>
    <dc:creator>AI Engineer</dc:creator>
    <description>&lt;p&gt;Bogdan Gaza 分享 DatologyAI 如何透過「種子重寫」技術，將合成資料規模從 300 億擴展至約 1200 億 token。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=FQwTqUmcbRg&quot;&gt;https://www.youtube.com/watch?v=FQwTqUmcbRg&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>訓練資料與合成資料</category>
  </item>
  <item>
    <title>AI Engineer Paris 2026 Main Stage: Google DeepMind, ElevenLabs, Hugging Face &amp; Stripe | Day 2</title>
    <link>https://aiwulin.itsmygo.uk/c/8a92b9fd44/</link>
    <guid isPermaLink="false">aiwulin-8a92b9fd44</guid>
    <pubDate>Fri, 25 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>AI Engineer</dc:creator>
    <description>&lt;p&gt;收錄 AI Engineer Paris 2026 第二天的主舞台演講，涵蓋從模型自訂到邊緣 AI 的實作經驗。觀眾可從 Google DeepMind、ElevenLabs 等講者處學習生產級 AI 系統的架構與最佳化技巧。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=g0vqT_wZtXA&quot;&gt;https://www.youtube.com/watch?v=g0vqT_wZtXA&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>AI 輔助軟體工程</category>
  </item>
  <item>
    <title>Haters think AI agents can't write GPU code? This'll ROCm</title>
    <link>https://aiwulin.itsmygo.uk/c/94dd70a7c5/</link>
    <guid isPermaLink="false">aiwulin-94dd70a7c5</guid>
    <pubDate>Tue, 22 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>The Stack Overflow Podcast</dc:creator>
    <description>&lt;p&gt;AMD 軟體主管 Anush Elangovan 訪談中，介紹 ROCm 如何透過開放原始碼與 AI 代理技術，大幅降低 GPU 程式設計門檻。內容涵蓋 ROCm 架構、SIMT 平行計算原理，以及 AI 如何協助最佳化硬體程式碼與效能。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://rss.art19.com/episodes/e3dd73f3-c12a-4053-b0bf-4550eb47f58a.mp3?rss_browser=BAhJIg9BSVd1bGluQm90BjoGRVQ%3D--40c69dfd25591c59cb985eaa38245c7694c51649&quot;&gt;https://rss.art19.com/episodes/e3dd73f3-c12a-4053-b0bf-4550eb47f58a.mp3?rss_browser=BAhJIg9BSVd1bGluQm90BjoGRVQ%3D--40c69dfd25591c59cb985eaa38245c7694c51649&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>AI 晶片與硬體</category>
    <category>AI Agent 基礎</category>
  </item>
  <item>
    <title>The Frontier AI Inference Cloud for Agents — Byung-Gon (Gon) Chun, FriendliAI</title>
    <link>https://aiwulin.itsmygo.uk/c/1f9310f708/</link>
    <guid isPermaLink="false">aiwulin-1f9310f708</guid>
    <pubDate>Sun, 20 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>AI Engineer</dc:creator>
    <description>&lt;p&gt;探討代理推理與聊天不同的架構需求，強調以任務完成時間為指標。透過開源模型與閉源模型比較，展示開源模型能以極低成本達成可用品質。FriendliAI 透過字首快取、分層 KV 快取管理、感知快取的路由及感知代理的排程，最佳化任務端到端延遲。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=Hvb2LfMH58c&quot;&gt;https://www.youtube.com/watch?v=Hvb2LfMH58c&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>推論與部署</category>
    <category>AI Agent 基礎</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>Two Bugs That Hid in Plain Sight: A vLLM Debugging Detective Story — Asaf Gardin &amp; Yuval Belfer</title>
    <link>https://aiwulin.itsmygo.uk/c/35ce1882b6/</link>
    <guid isPermaLink="false">aiwulin-35ce1882b6</guid>
    <pubDate>Sun, 20 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>AI Engineer</dc:creator>
    <description>&lt;p&gt;Asaf Gardin 與 Yuval Belfer 分享在 vLLM 部署 AI21 Jamba 模型時發現的兩個隱性錯誤。第一個導致偶爾輸出無意義內容，根源在於解碼階段早於預填充階段執行；第二個則由 32 位索引溢位引發。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=btxG75rNJC4&quot;&gt;https://www.youtube.com/watch?v=btxG75rNJC4&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>推論與部署</category>
  </item>
  <item>
    <title>Why Diffusion Will Win AI Inference with Inception Co-Founder and CEO Stefano Ermon</title>
    <link>https://aiwulin.itsmygo.uk/c/c6269e47b2/</link>
    <guid isPermaLink="false">aiwulin-c6269e47b2</guid>
    <pubDate>Fri, 18 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>No Priors</dc:creator>
    <description>&lt;p&gt;訪談 Inception 創辦人 Stefano Ermon，探討他如何將擴散模型應用於文字與程式碼生成，並解釋為何擴散架構在推理速度上優於自迴歸模型。內容涵蓋從學術研究到商業化部署的經驗，以及未來 AI 競爭將由效率決定的觀點。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://traffic.megaphone.fm/PDP7720707490.mp3&quot;&gt;https://traffic.megaphone.fm/PDP7720707490.mp3&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>推論與部署</category>
    <category>圖像生成</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>not much happened today</title>
    <link>https://aiwulin.itsmygo.uk/c/4fe1d45420/</link>
    <guid isPermaLink="false">aiwulin-4fe1d45420</guid>
    <pubDate>Thu, 10 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>AINews（Latent Space／smol.ai）</dc:creator>
    <description>&lt;p&gt;DeepSeek 推出 V4.1-Flash，採用新型因果編碼器 - 解碼器架構，大幅降低推理成本與記憶體佔用，在人工分析指數上獲 40 分，表現優於前代且僅次於 GLM-5.3-Flash。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://news.smol.ai/issues/26-09-10-not-much&quot;&gt;https://news.smol.ai/issues/26-09-10-not-much&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>開源模型</category>
    <category>推論與部署</category>
    <category>本機跑模型</category>
  </item>
  <item>
    <title>Async GRPO with LoRA across HF Jobs: a bucket, a proxy, and no NCCL</title>
    <link>https://aiwulin.itsmygo.uk/c/d4e55a9164/</link>
    <guid isPermaLink="false">aiwulin-d4e55a9164</guid>
    <pubDate>Thu, 10 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>Hugging Face Blog</dc:creator>
    <description>&lt;p&gt;介紹如何利用 Hugging Face Jobs 與 Storage Bucket，搭配 AsyncGRPOTrainer 與 LoRA，實現訓練與推論分離的加速方案。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://huggingface.co/blog/asyncgrpo-lora-hfjobs&quot;&gt;https://huggingface.co/blog/asyncgrpo-lora-hfjobs&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>微調</category>
    <category>推論與部署</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>OpenAI reports Navier-Stokes singularity find, a contender for second ever Millenium Prize awarded, overshadowing Cognition's $48B Series E, Mistral's $24B Series D, Meta's Muse agent, and GPT Image 2.5</title>
    <link>https://aiwulin.itsmygo.uk/c/febed0de22/</link>
    <guid isPermaLink="false">aiwulin-febed0de22</guid>
    <pubDate>Tue, 08 Sep 2026 00:00:00 +0800</pubDate>
    <dc:creator>AINews（Latent Space／smol.ai）</dc:creator>
    <description>&lt;p&gt;OpenAI 宣佈內部模型利用 10,000 個代理在 88 小時內完成納維 - 斯托克斯方程的證明，挑戰百萬美元獎金，引發對測試時間運算規模的討論。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://news.smol.ai/issues/26-09-08-navier-stokes&quot;&gt;https://news.smol.ai/issues/26-09-08-navier-stokes&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>開源模型</category>
    <category>AGI 與長期趨勢</category>
    <category>AI Agent 基礎</category>
  </item>
  <item>
    <title>DeepSeek竟然也遭遇斩杀线 | Qwen3.8-Flash Next | GLM-5.3-Flash | 大模型价格战 | 智谱 | 千问 | MoE架构 | 稀疏注意力 | Qwen4架构</title>
    <link>https://aiwulin.itsmygo.uk/c/9c525c26d3/</link>
    <guid isPermaLink="false">aiwulin-9c525c26d3</guid>
    <pubDate>Fri, 28 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>Best Partners TV</dc:creator>
    <description>&lt;p&gt;拆解阿里 Qwen3.8-Flash 與智譜 GLM-5.3-Flash 如何在 DeepSeek 漲價後，透過混合注意力、門控殘差及 N-gram 嵌入等架構創新大幅降低訓練與推理成本。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=VTBmsvOFlto&quot;&gt;https://www.youtube.com/watch?v=VTBmsvOFlto&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>開源模型</category>
    <category>大型語言模型入門</category>
  </item>
  <item>
    <title>not much happened today</title>
    <link>https://aiwulin.itsmygo.uk/c/d6db472a0b/</link>
    <guid isPermaLink="false">aiwulin-d6db472a0b</guid>
    <pubDate>Fri, 28 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>AINews（Latent Space／smol.ai）</dc:creator>
    <description>&lt;p&gt;整理 2026 年 8 月 AI 領域動態，涵蓋 Z.ai GLM-5.3、Tencent Hy4 與 Qwen3.8 Flash 等開源模型發布，以及 vLLM 推導解碼、搜尋系統與智慧體效能評估等系統層面進展。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://news.smol.ai/issues/26-08-28-not-much&quot;&gt;https://news.smol.ai/issues/26-08-28-not-much&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>開源模型</category>
    <category>模型發布與實測</category>
    <category>AI 評測</category>
  </item>
  <item>
    <title>not much happened today</title>
    <link>https://aiwulin.itsmygo.uk/c/bbf190a9a8/</link>
    <guid isPermaLink="false">aiwulin-bbf190a9a8</guid>
    <pubDate>Thu, 27 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>AINews（Latent Space／smol.ai）</dc:creator>
    <description>&lt;p&gt;彙整了 2026 年 8 月下旬 AI 領域的動態，重點介紹了 Pollen Robotics 與 Hugging Face 合作推出的 $399 開源雙足機器人 Microduck，以及 Z.ai 揭曉的 GLM-5.3-Flash 大…&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://news.smol.ai/issues/26-08-27-not-much&quot;&gt;https://news.smol.ai/issues/26-08-27-not-much&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>開源模型</category>
    <category>機器人與具身智慧</category>
    <category>大型語言模型入門</category>
  </item>
  <item>
    <title>This Small AI Will Change Everything</title>
    <link>https://aiwulin.itsmygo.uk/c/105ab17fca/</link>
    <guid isPermaLink="false">aiwulin-105ab17fca</guid>
    <pubDate>Tue, 25 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>Two Minute Papers</dc:creator>
    <description>&lt;p&gt;介紹 Qwen3.8-27b 模型在有限資源下的表現，展示其高效能與應用潛力。看完能了解該模型在實際環境中的運作方式與優勢。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=wMl6c_r0ubw&quot;&gt;https://www.youtube.com/watch?v=wMl6c_r0ubw&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>開源模型</category>
  </item>
  <item>
    <title>Granite 4.2 LLMs: How They're Built</title>
    <link>https://aiwulin.itsmygo.uk/c/74043a2c13/</link>
    <guid isPermaLink="false">aiwulin-74043a2c13</guid>
    <pubDate>Tue, 25 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>Hugging Face Blog</dc:creator>
    <description>&lt;p&gt;Granite 4.2 系列是 IBM 推出的三檔（3B、8B、30B）推理型大型語言模型，透過多階段強化學習與工具呼叫訓練，具備思考/不思考模式切換能力。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://huggingface.co/blog/ibm-granite/granite-4-2&quot;&gt;https://huggingface.co/blog/ibm-granite/granite-4-2&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>大型語言模型入門</category>
  </item>
  <item>
    <title>Inside LinkedIn's cognitive memory agent for agentic personalization</title>
    <link>https://aiwulin.itsmygo.uk/c/32a4881df7/</link>
    <guid isPermaLink="false">aiwulin-32a4881df7</guid>
    <pubDate>Tue, 25 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>The Stack Overflow Podcast</dc:creator>
    <description>&lt;p&gt;Podcast 訪談 LinkedIn 資深 AI 研究員 Praveen Bodigutla，介紹其團隊為招聘助手開發的認知記憶代理系統。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://rss.art19.com/episodes/412026c4-49d2-4c2a-b12a-15ccebde0c45.mp3?rss_browser=BAhJIg9BSVd1bGluQm90BjoGRVQ%3D--40c69dfd25591c59cb985eaa38245c7694c51649&quot;&gt;https://rss.art19.com/episodes/412026c4-49d2-4c2a-b12a-15ccebde0c45.mp3?rss_browser=BAhJIg9BSVd1bGluQm90BjoGRVQ%3D--40c69dfd25591c59cb985eaa38245c7694c51649&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>AI Agent 基礎</category>
    <category>檢索增強生成</category>
  </item>
  <item>
    <title>not much happened today</title>
    <link>https://aiwulin.itsmygo.uk/c/6ed84c545f/</link>
    <guid isPermaLink="false">aiwulin-6ed84c545f</guid>
    <pubDate>Mon, 24 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>AINews（Latent Space／smol.ai）</dc:creator>
    <description>&lt;p&gt;彙整了 AI 領域的最新動態，涵蓋 NVIDIA 提出的「Skill Lift」評估指標、Headlong 與 exo 等持久化代理架構、Anthropic 的企業 MCP 聯結器，以及 Qwen3.8-27B 等模型的表現。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://news.smol.ai/issues/26-08-24-not-much&quot;&gt;https://news.smol.ai/issues/26-08-24-not-much&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>AI 評測</category>
    <category>AI 晶片與硬體</category>
    <category>機器人與具身智慧</category>
  </item>
  <item>
    <title>not much happened today</title>
    <link>https://aiwulin.itsmygo.uk/c/b7d1c07510/</link>
    <guid isPermaLink="false">aiwulin-b7d1c07510</guid>
    <pubDate>Wed, 19 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>AINews（Latent Space／smol.ai）</dc:creator>
    <description>&lt;p&gt;介紹 Ornith-1.5 新模型家族、壓縮技術突破與 Agent 運算架構演進，涵蓋 Qwen3.8、DeepSeek Harness 與 TrueForge 等工具。讀者可掌握最新開源模型規格、效能比較與部署策略。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://news.smol.ai/issues/26-08-19-not-much&quot;&gt;https://news.smol.ai/issues/26-08-19-not-much&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>開源模型</category>
    <category>模型發布與實測</category>
    <category>AI Agent 基礎</category>
  </item>
  <item>
    <title>not much happened today</title>
    <link>https://aiwulin.itsmygo.uk/c/987042bbd3/</link>
    <guid isPermaLink="false">aiwulin-987042bbd3</guid>
    <pubDate>Fri, 14 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>AINews（Latent Space／smol.ai）</dc:creator>
    <description>&lt;p&gt;彙整了 2026 年 8 月 AI 領域的最新動態，重點涵蓋 Z.ai 推出 GLM-5.3、Alibaba 發布 Qwen3.8-27B 以及 DeepSeek V4-Pro 的更新。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://news.smol.ai/issues/26-08-14-cursor-xai&quot;&gt;https://news.smol.ai/issues/26-08-14-cursor-xai&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>開源模型</category>
    <category>AI 編輯器與開發工具</category>
    <category>模型發布與實測</category>
  </item>
  <item>
    <title>Meta is back with Muse Glimmer: local, agentic, multimodal, and open source</title>
    <link>https://aiwulin.itsmygo.uk/c/0da6fb777e/</link>
    <guid isPermaLink="false">aiwulin-0da6fb777e</guid>
    <pubDate>Mon, 10 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>Hugging Face Blog</dc:creator>
    <description>&lt;p&gt;Meta 推出 Muse Glimmer，一款 30B 參數的多模態開源模型，支援圖片、影片與程式碼處理，並提供本地部署與推論加速方案。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://huggingface.co/blog/muse-glimmer&quot;&gt;https://huggingface.co/blog/muse-glimmer&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>電腦視覺與多模態模型</category>
    <category>開源模型</category>
    <category>本機跑模型</category>
  </item>
  <item>
    <title>not much happened today</title>
    <link>https://aiwulin.itsmygo.uk/c/767c77ad4c/</link>
    <guid isPermaLink="false">aiwulin-767c77ad4c</guid>
    <pubDate>Mon, 10 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>AINews（Latent Space／smol.ai）</dc:creator>
    <description>&lt;p&gt;Meta 推出 Muse Glimmer 30B 開源多模態代理模型，支援本地部署與量化加速；同時 Anthropic 的 Claude 變體在黎曼猜想研究中提升下界，OpenAI 則發布 GPT-5.6-Cyber 專注網路安全。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://news.smol.ai/issues/26-08-10-not-much&quot;&gt;https://news.smol.ai/issues/26-08-10-not-much&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>模型發布與實測</category>
    <category>開源模型</category>
    <category>本機跑模型</category>
  </item>
  <item>
    <title>Inside vLLM: The Engine Powering Open-Source AI</title>
    <link>https://aiwulin.itsmygo.uk/c/71b627b645/</link>
    <guid isPermaLink="false">aiwulin-71b627b645</guid>
    <pubDate>Thu, 06 Aug 2026 00:00:00 +0800</pubDate>
    <dc:creator>AI + a16z</dc:creator>
    <description>&lt;p&gt;探討 vLLM 如何從研究專案演變為關鍵基礎設施，並由 Simon Mo 與 Matt Bornstein 分析開源 AI 的經濟模式與未來。內容涵蓋開權重模型的興起、企業轉向開源的原因、模型授權機制，以及開源與閉源模型能力的差距縮小趨勢。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://ai-a16z.simplecast.com/episodes/inside-vllm-the-engine-powering-open-source-ai-etgKL1N9&quot;&gt;https://ai-a16z.simplecast.com/episodes/inside-vllm-the-engine-powering-open-source-ai-etgKL1N9&lt;/a&gt;&lt;/p&gt;</description>
    <category>Podcast</category>
    <category>推論與部署</category>
    <category>開源模型</category>
    <category>大型語言模型入門</category>
  </item>
  <item>
    <title>not much happened today</title>
    <link>https://aiwulin.itsmygo.uk/c/7d14e96bd7/</link>
    <guid isPermaLink="false">aiwulin-7d14e96bd7</guid>
    <pubDate>Wed, 29 Jul 2026 00:00:00 +0800</pubDate>
    <dc:creator>AINews（Latent Space／smol.ai）</dc:creator>
    <description>&lt;p&gt;彙整了 2026 年 7 月底 AI 領域動態，涵蓋 OpenAI 代理安全事件、GPT-5.6 效能最佳化、Kimi K3 模型細節及學術資源開放等議題。讀者可了解最新模型表現、安全治理爭議及開發工具更新。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://news.smol.ai/issues/26-07-29-not-much&quot;&gt;https://news.smol.ai/issues/26-07-29-not-much&lt;/a&gt;&lt;/p&gt;</description>
    <category>文章</category>
    <category>模型發布與實測</category>
    <category>開源模型</category>
    <category>AI Agent 基礎</category>
  </item>
  <item>
    <title>You Kaichao: vLLM, Open-Source Infra, Model Co-Design &amp; Journey from Community to Startup</title>
    <link>https://aiwulin.itsmygo.uk/c/deb198da21/</link>
    <guid isPermaLink="false">aiwulin-deb198da21</guid>
    <pubDate>Wed, 29 Jul 2026 00:00:00 +0800</pubDate>
    <dc:creator>Zhang Xiaojun Podcast</dc:creator>
    <description>&lt;p&gt;訪談遊凱超分享 vLLM 如何從校園專案演變為獲 1.5 億美元融資的創業公司，並探討 AI 基礎設施的多方協同設計。&lt;/p&gt;&lt;p&gt;原站：&lt;a href=&quot;https://www.youtube.com/watch?v=ifniRXf467I&quot;&gt;https://www.youtube.com/watch?v=ifniRXf467I&lt;/a&gt;&lt;/p&gt;</description>
    <category>影片</category>
    <category>推論與部署</category>
  </item>
</channel>
</rss>
