[{"data":1,"prerenderedAt":205},["ShallowReactive",2],{"DlFXI4Eibt_Bn9lrEZz1TYbHCWFZj3IvqwHQSEW-Exc":3,"7B0hjmMQ31LwqHOfIMCcwNjzCXrIhhQVaSMUnGPxwWA":194},{"code":4,"msg":5,"data":6},0,"",{"category":7,"tag":11,"hot":39,"new":78,"banner":118,"data":143,"cache":193},[8,9,10],"Agent","OpenAI","LLM",[12,14,17,20,23,25,27,30,33,36],{"title":8,"total":13},39,{"title":15,"total":16},"Google",44,{"title":18,"total":19},"Nvidia",13,{"title":21,"total":22},"Claude",11,{"title":9,"total":24},35,{"title":10,"total":26},85,{"title":28,"total":29},"DeepSeek",9,{"title":31,"total":32},"OCR",1,{"title":34,"total":35},"Chat",7,{"title":37,"total":38},"Generator",116,[40,48,55,64,71],{"id":41,"publish_date":42,"is_original":4,"collection":5,"cover_url":43,"cover_url_1_1":44,"title":45,"summary":46,"author":47},557,"2022-04-29","article_res/cover/7a9b1375ed9bb298154981bae42b794d.jpeg","article_res/cover/afa281dd52bc0454e6735daa8e6b0706.jpeg","Translation and summary of Messari Report [2.8 Kristin Smith, Blockchain Association and Katie Haun, a16z]","We need unity and speed right now.","Translation",{"id":49,"publish_date":50,"is_original":4,"collection":5,"cover_url":51,"cover_url_1_1":52,"title":53,"summary":54,"author":47},531,"2022-05-25","article_res/cover/e8362057f8fa189594c60afdfaaeb6e5.jpeg","article_res/cover/8ea08d0d6fa7eee6b57ed4ec61b61ad6.jpeg","Decentralized Society: Finding Web3’s Soul / Decentralized Society: Finding the Soul of Web3 -7","Decentralization through Pluralism When analyzing ecosystems, it's desirable to measure how decentralized it is.",{"id":56,"publish_date":57,"is_original":32,"collection":58,"cover_url":59,"cover_url_1_1":60,"title":61,"summary":62,"author":63},127,"2024-11-14","#Google #AI Game #World Model #AI Story","article_res/cover/0233a875b7ec2debf59779e311547569.jpeg","article_res/cover/6ffddb6ae4914b3c699493311aa9f198.jpeg","Google Launches \"Unbounded\": A Generative Infinite Character Life Simulation Game","Unbounded: A Generative Infinite Game of Character Life Simulation","Renee's Entrepreneurial Journey",{"id":13,"publish_date":65,"is_original":32,"collection":66,"cover_url":67,"cover_url_1_1":68,"title":69,"summary":70,"author":63},"2025-02-14","#Deep Dive into LLMs #Andrej Karpathy #LLM #Tool Use #Hallucination","article_res/cover/11e858ad6b74dfa80f923d549b62855c.jpeg","article_res/cover/615e1b320f1fc163edc1d2d154a6de33.jpeg","Andrej Karpathy's in-depth explanation of LLM (Part 4): Hallucinations","hallucinations, tool use, knowledge/working memory",{"id":72,"publish_date":73,"is_original":4,"collection":5,"cover_url":74,"cover_url_1_1":75,"title":76,"summary":77,"author":47},579,"2022-04-07","article_res/cover/39387376ba28447af1eb40576b9df215.jpeg","article_res/cover/02727ede8551ed49901d0abe6d6305b7.jpeg","Messari Report Translation and Summary 【1-7 Surviving the Winter】","I’d be more cautious here: 10 year and 10 hour thinking only.",[79,87,95,103,111],{"id":80,"publish_date":81,"is_original":32,"collection":82,"cover_url":83,"cover_url_1_1":84,"title":85,"summary":86,"author":63},627,"2025-03-20","#AI Avatar #AI Video Generation","article_res/cover/d95481358f73924989f8c4ee9c75d1c8.jpeg","article_res/cover/b74bc0fab01f8b6a6aa87696c0c3ed8b.jpeg","DisPose: Generating Animated Videos by Driving Video with Reference Images","DisPose is a controllable human image animation method that enhances video generation.",{"id":88,"publish_date":89,"is_original":32,"collection":90,"cover_url":91,"cover_url_1_1":92,"title":93,"summary":94,"author":63},626,"2025-03-21","#Deep Dive into LLMs #LLM #RL #Andrej Karpathy #AlphaGo","article_res/cover/446553a5c8f8f2f07d97b20eaee84e56.jpeg","article_res/cover/e6c2823409c9b34624064b9acbaca6f1.jpeg","AlphaGo and the Power of Reinforcement Learning - Andrej Karpathy's Deep Dive on LLMs (Part 9)","Simply learning from humans will never surpass human capabilities.",{"id":96,"publish_date":97,"is_original":32,"collection":98,"cover_url":99,"cover_url_1_1":100,"title":101,"summary":102,"author":63},625,"2025-03-22","#Deep Dive into LLMs #LLM #RL #RLHF #Andrej Karpathy","article_res/cover/8da81d38b1e5cf558a164710fd8a5389.jpeg","article_res/cover/96f028d76c362a99a0dd56389e8f7a9b.jpeg","Reinforcement Learning from Human Feedback (RLHF) - Andrej Karpathy's Deep Dive on LLMs (Part 10)","Fine-Tuning Language Models from Human Preferences",{"id":104,"publish_date":105,"is_original":32,"collection":106,"cover_url":107,"cover_url_1_1":108,"title":109,"summary":110,"author":63},624,"2025-03-23","#Deep Dive into LLMs #LLM #Andrej Karpathy #AI Agent #MMM","article_res/cover/a5e7c3d48bb09109684d6513287c661d.jpeg","article_res/cover/d3f22b7c0ab8d82fd2da457a299e0773.jpeg","The Future of Large Language Models - Andrej Karpathy's In-Depth Explanation of LLM (Part 11)","preview of things to come",{"id":112,"publish_date":105,"is_original":32,"collection":113,"cover_url":114,"cover_url_1_1":115,"title":116,"summary":117,"author":63},623,"#Google #Voe #AI Video Generation","article_res/cover/c44062fea0f336c2b96b3928292392c2.jpeg","article_res/cover/a041041c69092ad3db191c5bf3ff981b.jpeg","Trial of Google's video generation model VOE2","Our state-of-the-art video generation model",[119,127,135],{"id":120,"publish_date":121,"is_original":32,"collection":122,"cover_url":123,"cover_url_1_1":124,"title":125,"summary":126,"author":63},160,"2024-10-04","#Philosophy","article_res/cover/496990c49211e8b7f996b7d39c18168e.jpeg","article_res/cover/14dbaa1ade9cb4316d5829423a900362.jpeg","Time","The fungus of the morning does not know the waxing and waning of the moon, and the cicada does not know the seasons; this is a short life. To the south of the state of Chu there is a dark spirit which regards five hundred years as spring and five hundred years as autumn. In ancient times there was a great tree called the Ming which regarded eight thousand years as spring and eight thousand years as autumn; this is a long life.",{"id":128,"publish_date":129,"is_original":32,"collection":130,"cover_url":131,"cover_url_1_1":132,"title":133,"summary":134,"author":63},98,"2024-12-17","#AI Video Generator #Sora #Pika","article_res/cover/3b86e85d03fff4f356a3e4cf2bb329c9.jpeg","article_res/cover/5fa5c20ad0b40f8f544d257c0ef02938.jpeg","Pika 2.0 video generation officially released: effect comparison with Sora","今天，我们推出了Pika 2.0模型。卓越的文字对齐效果。惊人的视觉表现。还有✨场景成分✨",{"id":136,"publish_date":137,"is_original":32,"collection":138,"cover_url":139,"cover_url_1_1":140,"title":141,"summary":142,"author":63},71,"2025-01-14","#Nvidia #World Foundation Model #Cosmos #Physical AI #Embodied AI","article_res/cover/feddf8c832dfb45d28804291f6a42a9e.jpeg","article_res/cover/d6bc2f1186d96b78228c2283a17a3645.jpeg","NVIDIA's Cosmos World Model","Cosmos World Foundation Model Platform for Physical AI",[144,163,188],{"title":8,"items":145},[146,147,155],{"id":104,"publish_date":105,"is_original":32,"collection":106,"cover_url":107,"cover_url_1_1":108,"title":109,"summary":110,"author":63},{"id":148,"publish_date":149,"is_original":32,"collection":150,"cover_url":151,"cover_url_1_1":152,"title":153,"summary":154,"author":63},622,"2025-03-24","#OWL #AI Agent #MAS #MCP #CUA","article_res/cover/cb50ca7f2bf4d1ed50202d7406e1c19a.jpeg","article_res/cover/4aa7aa3badfacf3cc84121334f1050dd.jpeg","OWL: Multi-agent collaboration","OWL: Optimized Workforce Learning for General Multi-Agent Assistance in Real-World Task Automation",{"id":156,"publish_date":157,"is_original":32,"collection":158,"cover_url":159,"cover_url_1_1":160,"title":161,"summary":162,"author":63},620,"2025-03-26","#LLM #Google #Gemini #AI Agent","article_res/cover/53751a6dbbe990b1eb0b63f3b062aed4.jpeg","article_res/cover/031344981f0a212ff82d1f3a64aa5756.jpeg","Gemini 2.5 Pro, claimed to be far ahead of the competition, has been released with great fanfare: comprehensively surpassing other LLMs and topping the global rankings","Gemini 2.5: Our most intelligent AI model",{"title":9,"items":164},[165,172,180],{"id":166,"publish_date":157,"is_original":32,"collection":167,"cover_url":168,"cover_url_1_1":169,"title":170,"summary":171,"author":63},619,"#OpenAI #AI Image Generator #4o #MMM #AR Transformer","article_res/cover/2faffc97fcecf3151552cb0fd3206d89.jpeg","article_res/cover/1133cb4948af44cee2e7fbe79efb69e5.jpeg","The native image function of GPT-4o is officially launched","Introducing 4o Image Generation",{"id":173,"publish_date":174,"is_original":4,"collection":175,"cover_url":176,"cover_url_1_1":177,"title":178,"summary":179,"author":63},434,"2023-07-15","#Anthropic #OpenAI #Google #AI Code Generator #Claude","article_res/cover/e1b6f600a2b9f262a4392684e5f2ce25.jpeg","article_res/cover/6e1772e83f78f9a351ab23d3e414adee.jpeg","Latest Updates on Google Bard /Anthropic Claude2 / ChatGPT Code Interpreter","We want our models to use their programming skills to provide more natural interfaces to the basic functions of our computers.  \n - OpenAI",{"id":181,"publish_date":182,"is_original":4,"collection":183,"cover_url":184,"cover_url_1_1":185,"title":186,"summary":187,"author":63},417,"2023-08-24","#OpenAI","article_res/cover/bccf897d50a88b18364e35f7466387e0.jpeg","article_res/cover/2f871085c1073717c1703ae86e18056f.jpeg","The GPT-3.5 Turbo fine-tuning (fine-tuning function) has been released～","Developers can now bring their own data to customize GPT-3.5 Turbo for their use cases.",{"title":10,"items":189},[190,191,192],{"id":88,"publish_date":89,"is_original":32,"collection":90,"cover_url":91,"cover_url_1_1":92,"title":93,"summary":94,"author":63},{"id":96,"publish_date":97,"is_original":32,"collection":98,"cover_url":99,"cover_url_1_1":100,"title":101,"summary":102,"author":63},{"id":104,"publish_date":105,"is_original":32,"collection":106,"cover_url":107,"cover_url_1_1":108,"title":109,"summary":110,"author":63},true,{"code":4,"msg":5,"data":195},{"id":196,"publish_date":197,"is_original":32,"collection":198,"articles_id":199,"cover_url":200,"cover_url_1_1":201,"title":202,"summary":203,"author":63,"content":204},24,"2025-02-28","#Deepseek #AGI","AZlCGABXqV8k0M6newejwA","article_res/cover/5f7d58e411f4f45eb94fc458c2b7acc6.jpeg","article_res/cover/d4a0130cd0de159f3d77b43579691da2.jpeg","DeepSeek #OpenSourceWeek - Five Consecutive Releases","We're a tiny team @deepseek_ai exploring AGI.","\u003Cdiv class=\"rich_media_content js_underline_content\n                       autoTypeSetting24psection\n            \" id=\"js_content\">\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan leaf=\"\">DeepSeek 在 #OpenSourceWeek 中的五连弹展示了其在 AGI 研究中的最新开源进展。以下是各天发布的内容概览：\u003C/span>\u003C/p>\u003Csection data-mpa-template=\"t\" mpa-from-tpl=\"t\" data-mpa-action-id=\"m7nbe5c219uh\">\u003Csection style=\"display: flex;flex-direction: column;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"display: flex;flex-direction: column;align-self: center;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"width: 143px;display: grid;grid-template-columns: 100% 100%;align-items: flex-start;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"display: flex;justify-content: center;align-items: center;width: 143px;height: 73px;\" data-mid=\"\" mpa-from-tpl=\"t\" nodeleaf=\"\">\u003Cimg class=\"rich_pages wxw-img\" data-ratio=\"0.5524475524475524\" data-w=\"286\" src=\"https://res.cooltool.vip/article_res/assets/17423769958750.18887793091881178.png\">\u003C/section>\u003Csection style=\"transform: translateX(-100.1%);padding: 19px 8px 0 8px;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"text-align: left;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Cp style=\"font-weight: bold;font-size: 14px;color: #2D2D2D;line-height: 15px;word-break: break-word;\" data-mid=\"\">\u003Cspan leaf=\"\">DAY 0\u003C/span>\u003C/p>\u003C/section>\u003Csection style=\"text-align: left;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Cp style=\"font-weight: bold;font-size: 17px;color: #2D2D2D;line-height: 21px;word-break: break-word;\" data-mid=\"\">\u003Cspan leaf=\"\">预热\u003C/span>\u003C/p>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan leaf=\"\">DeepSeek 团队宣布将开源 5 个仓库，并强调其透明化的研究方式。团队自称“小而精”，但希望通过社区合作加速 AGI 的发展。\u003C/span>\u003C/p>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/p>\u003Csection data-mpa-template=\"t\" mpa-from-tpl=\"t\" data-mpa-action-id=\"m7nbe5c219uh\">\u003Csection style=\"display: flex;flex-direction: column;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"display: flex;flex-direction: column;align-self: center;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"width: 143px;display: grid;grid-template-columns: 100% 100%;align-items: flex-start;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"display: flex;justify-content: center;align-items: center;width: 143px;height: 73px;\" data-mid=\"\" mpa-from-tpl=\"t\" nodeleaf=\"\">\u003Cimg class=\"rich_pages wxw-img\" data-ratio=\"0.5524475524475524\" data-w=\"286\" src=\"https://res.cooltool.vip/article_res/assets/17423769958750.4188425376202407.png\">\u003C/section>\u003Csection style=\"transform: translateX(-100.1%);padding: 19px 8px 0 8px;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"text-align: left;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Cp style=\"font-weight: bold;font-size: 14px;color: #2D2D2D;line-height: 15px;word-break: break-word;\" data-mid=\"\">\u003Cspan leaf=\"\">DAY 1\u003C/span>\u003C/p>\u003C/section>\u003Csection style=\"text-align: left;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Cp style=\"font-weight: bold;font-size: 17px;color: #2D2D2D;line-height: 21px;word-break: break-word;\" data-mid=\"\">\u003Cspan leaf=\"\">FlashMLA\u003C/span>\u003C/p>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003Ch4 style='box-sizing: border-box;margin: 30px 0px 15px;color: rgba(0, 0, 0, 0.85);font-weight: 500;;padding: 0px;display: block;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan style=\"box-sizing: border-box;;font-size: 18px;color: rgb(0, 0, 0);line-height: 1.5em;letter-spacing: 0em;text-align: left;font-weight: bold;display: block;\">\u003Cspan leaf=\"\">核心内容：\u003C/span>\u003C/span>\u003C/h4>\u003Cul style='box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;' class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">FlashMLA 是一个高效的 MLA（Memory Limited Attention）解码内核，专门为 Hopper GPU 进行优化，支持可变长度序列。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">技术亮点\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">：\u003C/span>\u003C/section>\u003C/li>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">支持 BF16（BFloat16）\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">采用 Paged KV 缓存（块大小 64）\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">3000 GB/s 内存带宽 & 580 TFLOPS 计算性能（H800）\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003C/ul>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">📥 GitHub:\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> https://github.com/deepseek-ai/FlashMLA\u003C/span>\u003C/p>\u003Ch4 style='box-sizing: border-box;margin: 30px 0px 15px;color: rgba(0, 0, 0, 0.85);font-weight: 500;;padding: 0px;display: block;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan style=\"box-sizing: border-box;;font-size: 18px;color: rgb(0, 0, 0);line-height: 1.5em;letter-spacing: 0em;text-align: left;font-weight: bold;display: block;\">\u003Cspan leaf=\"\">关键优势：\u003C/span>\u003C/span>\u003C/h4>\u003Csection style='box-sizing: border-box;;margin: 20px 0px;padding: 10px 20px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgb(250, 250, 250);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Csection style=\"box-sizing: border-box;;\">\u003Col style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: decimal;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">针对 Hopper GPU 的优化\u003C/span>\u003C/strong>\u003C/p>\u003C/section>\u003C/li>\u003C/ol>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">专为 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">NVIDIA Hopper GPU（H800）\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> 设计，充分利用 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">Tensor Cores 和 Transformer Engine\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">3000 GB/s 内存带宽\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，加速数据传输。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">580 TFLOPS 计算性能\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，提升 AI 计算吞吐量。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">支持可变长度序列\u003C/span>\u003C/strong>\u003C/p>\u003C/section>\u003C/li>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">适配 NLP 任务（如 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">聊天机器人、翻译、文本生成\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">），无需固定序列长度，提高计算效率。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">高效内存管理（Paged KV Cache）\u003C/span>\u003C/strong>\u003C/p>\u003C/section>\u003C/li>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">块大小 64\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，减少内存占用，提高 LLM 推理速度。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">受 \"LLM in a Flash\" 研究启发，优化长序列推理。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">支持 BF16 精度\u003C/span>\u003C/strong>\u003C/p>\u003C/section>\u003C/li>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">比 FP32 更高效\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，降低存储需求，加速计算，同时保持 AI 任务的精度。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">可运行更大规模 AI 模型\u003C/span>\u003C/strong>\u003C/p>\u003C/section>\u003C/li>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">结合 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">Flash Memory & 高效数据传输\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，突破 GPU DRAM 限制。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">运行比 GPU 内存 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">大 2 倍的模型\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，提高 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">4-5 倍（CPU）/ 20-25 倍（GPU）\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> 性能。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">开源，推动 AI 社区创新\u003C/span>\u003C/strong>\u003C/p>\u003C/section>\u003C/li>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">GitHub 开源\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，全球开发者可自由使用、优化、集成，降低开发成本。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">已在生产环境使用\u003C/span>\u003C/strong>\u003C/p>\u003C/section>\u003C/li>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">经过测试，稳定可靠，可直接部署于 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">大规模 AI 任务\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">竞争优势\u003C/span>\u003C/strong>\u003C/p>\u003C/section>\u003C/li>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">对标 FlashAttention\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，深度优化 Hopper GPU，降低 AI 推理成本，提高性能。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003C/section>\u003C/section>\u003Ch4 style='box-sizing: border-box;margin: 30px 0px 15px;color: rgba(0, 0, 0, 0.85);font-weight: 500;;padding: 0px;display: block;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan style=\"box-sizing: border-box;;font-size: 18px;color: rgb(0, 0, 0);line-height: 1.5em;letter-spacing: 0em;text-align: left;font-weight: bold;display: block;\">\u003Cspan leaf=\"\">网友评论：\u003C/span>\u003C/span>\u003C/h4>\u003Cblockquote style='box-sizing: border-box;margin: 20px 0px;;padding: 10px 10px 10px 20px;border-style: none none none solid;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0.05);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;overflow: auto;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;font-weight: normal;\">\u003Cspan leaf=\"\">Wow, what an awesome start into the new week! Hope PyTorch adopts it in their next version release (like they integrated FlashAttention).\u003C/span>\u003C/p>\u003C/blockquote>\u003Cblockquote style='box-sizing: border-box;margin: 20px 0px;;padding: 10px 10px 10px 20px;border-style: none none none solid;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0.05);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;overflow: auto;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;font-weight: normal;\">\u003Cspan leaf=\"\">DeepSeek’s FlashMLA is so fast it’s practically doing stand-up comedy on Hopper GPUs—3000 GB/s and 580 TFLOPS, wow, what’s next, live conversations with your toaster, or full-on AGI by breakfast? 😂\u003C/span>\u003C/p>\u003C/blockquote>\u003Cblockquote style='box-sizing: border-box;margin: 20px 0px;;padding: 10px 10px 10px 20px;border-style: none none none solid;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0.05);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;overflow: auto;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;font-weight: normal;\">\u003Cspan leaf=\"\">FlashMLA's optimization for Hopper GPUs is set to give DeepSeek a serious performance boost. By enhancing decoding efficiency, it's tackling those pesky \"server busy\" issues head-on, especially at peak times. The secret sauce? Streamlining variable-length sequence handling.\u003C/span>\u003C/p>\u003C/blockquote>\u003Csection>\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/section>\u003Csection data-mpa-template=\"t\" mpa-from-tpl=\"t\" data-mpa-action-id=\"m7nbe5c219uh\">\u003Csection style=\"display: flex;flex-direction: column;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"display: flex;flex-direction: column;align-self: center;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"width: 143px;display: grid;grid-template-columns: 100% 100%;align-items: flex-start;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"display: flex;justify-content: center;align-items: center;width: 143px;height: 73px;\" data-mid=\"\" mpa-from-tpl=\"t\" nodeleaf=\"\">\u003Cimg class=\"rich_pages wxw-img\" data-ratio=\"0.5524475524475524\" data-w=\"286\" src=\"https://res.cooltool.vip/article_res/assets/17423769958830.2056293898811672.png\">\u003C/section>\u003Csection style=\"transform: translateX(-100.1%);padding: 19px 8px 0 8px;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"text-align: left;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Cp style=\"font-weight: bold;font-size: 14px;color: #2D2D2D;line-height: 15px;word-break: break-word;\" data-mid=\"\">\u003Cspan leaf=\"\">DAY 2\u003C/span>\u003C/p>\u003C/section>\u003Csection style=\"text-align: left;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Cp style=\"font-weight: bold;font-size: 17px;color: #2D2D2D;line-height: 21px;word-break: break-word;\" data-mid=\"\">\u003Cspan leaf=\"\">DeepEP\u003C/span>\u003C/p>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003Ch4 style='box-sizing: border-box;margin: 30px 0px 15px;color: rgba(0, 0, 0, 0.85);font-weight: 500;;padding: 0px;display: block;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan style=\"box-sizing: border-box;;font-size: 18px;color: rgb(0, 0, 0);line-height: 1.5em;letter-spacing: 0em;text-align: left;font-weight: bold;display: block;\">\u003Cspan leaf=\"\">核心内容：\u003C/span>\u003C/span>\u003C/h4>\u003Cul style='box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;' class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">DeepEP 是首个开源的 EP（Expert Parallel）通信库，优化了 Mixture of Experts（MoE）模型的训练与推理通信。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">技术亮点\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">：\u003C/span>\u003C/section>\u003C/li>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">高效的 All-to-All 通信优化\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">支持 NVLink 和 RDMA 的节点内与节点间通信\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">预填充推理的高吞吐核函数\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">低延迟解码推理\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">原生 FP8 计算支持\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">GPU 资源灵活管理，实现计算-通信重叠\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003C/ul>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">📥 GitHub:\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> https://github.com/deepseek-ai/DeepEP\u003C/span>\u003C/p>\u003Ch4 style='box-sizing: border-box;margin: 30px 0px 15px;color: rgba(0, 0, 0, 0.85);font-weight: 500;;padding: 0px;display: block;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan style=\"box-sizing: border-box;;font-size: 18px;color: rgb(0, 0, 0);line-height: 1.5em;letter-spacing: 0em;text-align: left;font-weight: bold;display: block;\">\u003Cspan leaf=\"\">DeepEP说明\u003C/span>\u003C/span>\u003C/h4>\u003Csection style='box-sizing: border-box;;margin: 20px 0px;padding: 10px 20px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgb(250, 250, 250);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Csection style=\"box-sizing: border-box;;\">\u003Col style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: decimal;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">高效的 All-to-All 通信\u003C/span>\u003C/section>\u003C/li>\u003C/ol>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">在 MoE 模型 中，每个专家（Expert）都需要与其他专家 交换数据。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">DeepEP 让数据像 高速公路上的专用车道 一样，确保流畅传输，无通信瓶颈。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">影响：加快 MoE 模型的训练速度，减少数据交换延迟。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">📌 解读：传统 MoE 模型在多 GPU 之间交换数据时容易出现通信瓶颈，DeepEP 通过优化 All-to-All 通信，让数据高效流动，避免“塞车”。\u003C/span>\u003C/p>\u003Col style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: decimal;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">支持 Intranode & Internode 通信\u003C/span>\u003C/section>\u003C/li>\u003C/ol>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">Intranode 通信（同一芯片内）\u003C/span>\u003C/p>\u003C/section>\u003C/li>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">NVLink 技术（NVIDIA 高速互联技术）加速数据传输，让 GPU 之间高效沟通。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">比喻：像城市内部的高速公路，确保市内交通快速流动。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">Internode 通信（跨设备传输）\u003C/span>\u003C/p>\u003C/section>\u003C/li>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">采用 RDMA（远程直接内存访问），让数据直接在不同机器间传输，无需 CPU 介入。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">比喻：类似 城际高铁，让数据跨服务器流动时更快、更低延迟。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">📌 解读：DeepEP 同时优化 GPU 内部和跨 GPU 设备的通信，确保 AI 训练在单机和多机环境下都能高效运行。\u003C/span>\u003C/p>\u003Col style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: decimal;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">高吞吐 & 低延迟 GPU 内核\u003C/span>\u003C/section>\u003C/li>\u003C/ol>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">AI 训练的效率不仅取决于 GPU 计算能力，还取决于数据交换的速度。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">DeepEP 采用 专用 GPU 内核，减少数据交换等待时间，让 GPU 始终保持高效计算。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">影响：AI 训练更快，GPU 资源利用率更高。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">📌 解读：传统通信方式可能导致 GPU 等待数据，DeepEP 减少等待时间，提高吞吐量，让 GPU 计算能力得到充分发挥。\u003C/span>\u003C/p>\u003Col style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: decimal;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">灵活的资源控制\u003C/span>\u003C/section>\u003C/li>\u003C/ol>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">动态分配 GPU 计算任务，让 AI 训练 适配不同硬件环境。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">不同规模的 AI 训练任务都能高效运行，不会因为资源分配不均导致性能浪费。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">影响：适用于 多种 AI 训练场景，让开发者灵活优化计算资源。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">📌 解读：DeepEP 让 GPU 计算资源可以灵活适配不同的 AI 任务，即使硬件配置不同，也能保持高效运作。\u003C/span>\u003C/p>\u003C/section>\u003C/section>\u003Ch4 style='box-sizing: border-box;margin: 30px 0px 15px;color: rgba(0, 0, 0, 0.85);font-weight: 500;;padding: 0px;display: block;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan style=\"box-sizing: border-box;;font-size: 18px;color: rgb(0, 0, 0);line-height: 1.5em;letter-spacing: 0em;text-align: left;font-weight: bold;display: block;\">\u003Cspan leaf=\"\">网友评论：\u003C/span>\u003C/span>\u003C/h4>\u003Cblockquote style='box-sizing: border-box;margin: 20px 0px;;padding: 10px 10px 10px 20px;border-style: none none none solid;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0.05);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;overflow: auto;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cp>\u003Cspan leaf=\"\">昨天DeepSeek发布的DeepEP在文档中描述：\u003C/span>\u003C/p>\u003Cp>\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/p>\u003Cp>\u003Cspan leaf=\"\">“为了追求极致性能，我们发现并使用了一个文档之外的PTX指令：ld.global.nc.L1::no_allocate.L2::256B。这条指令会导致未定义行为：使用非一致性只读PTX修饰符.nc访问易失性GPU内存。但是，在Hopper架构上，经过测试，使用.L1::no_allocate可以保证正确性，并且性能会好得多。”\u003C/span>\u003C/p>\u003Cp>\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/p>\u003Cp>\u003Cspan leaf=\"\">但实际上no_allocate这个指令其实文档中是有的（详见文档214页），但是只是草草说了句用途，并没有详细解释能带来什么提升。\u003C/span>\u003C/p>\u003Cp>\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/p>\u003Cp>\u003Cspan leaf=\"\">可见DeepSeek 对CUDA的研究程度之深，以及在GPU领域的积累。\u003C/span>\u003C/p>\u003C/blockquote>\u003Cblockquote style='box-sizing: border-box;margin: 20px 0px;;padding: 10px 10px 10px 20px;border-style: none none none solid;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0.05);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;overflow: auto;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;font-weight: normal;\">\u003Cspan leaf=\"\">官网文档没说清楚，好多nv的哥们昨天都在吐槽，不过基本不太有人deep 到这个细节程度，可想而知他们是做了多少实验，确实挺狠。\u003C/span>\u003C/p>\u003C/blockquote>\u003Cblockquote style='box-sizing: border-box;margin: 20px 0px;;padding: 10px 10px 10px 20px;border-style: none none none solid;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0.05);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;overflow: auto;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;font-weight: normal;\">\u003Cspan leaf=\"\">工程师的极端优化思维[赞][赞][赞]\u003C/span>\u003C/p>\u003C/blockquote>\u003Csection>\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/section>\u003Csection data-mpa-template=\"t\" mpa-from-tpl=\"t\" data-mpa-action-id=\"m7nbe5c219uh\">\u003Csection style=\"display: flex;flex-direction: column;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"display: flex;flex-direction: column;align-self: center;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"width: 143px;display: grid;grid-template-columns: 100% 100%;align-items: flex-start;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"display: flex;justify-content: center;align-items: center;width: 143px;height: 73px;\" data-mid=\"\" mpa-from-tpl=\"t\" nodeleaf=\"\">\u003Cimg class=\"rich_pages wxw-img\" data-ratio=\"0.5524475524475524\" data-w=\"286\" src=\"https://res.cooltool.vip/article_res/assets/17423769958830.9929509932720362.png\">\u003C/section>\u003Csection style=\"transform: translateX(-100.1%);padding: 19px 8px 0 8px;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"text-align: left;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Cp style=\"font-weight: bold;font-size: 14px;color: #2D2D2D;line-height: 15px;word-break: break-word;\" data-mid=\"\">\u003Cspan leaf=\"\">DAY 3\u003C/span>\u003C/p>\u003C/section>\u003Csection style=\"text-align: left;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Cp style=\"font-weight: bold;font-size: 17px;color: #2D2D2D;line-height: 21px;word-break: break-word;\" data-mid=\"\">\u003Cspan leaf=\"\">DeepGEMM\u003C/span>\u003C/p>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003Ch4 style='box-sizing: border-box;margin: 30px 0px 15px;color: rgba(0, 0, 0, 0.85);font-weight: 500;;padding: 0px;display: block;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan style=\"box-sizing: border-box;;font-size: 18px;color: rgb(0, 0, 0);line-height: 1.5em;letter-spacing: 0em;text-align: left;font-weight: bold;display: block;\">\u003Cspan leaf=\"\">核心内容：\u003C/span>\u003C/span>\u003C/h4>\u003Cul style='box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;' class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">DeepGEMM 是一款 FP8 GEMM（General Matrix Multiplication）库，支持稠密和 MoE 计算，优化 V3/R1 训练与推理。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">技术亮点\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">：\u003C/span>\u003C/section>\u003C/li>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">Hopper GPU 上可达 1350+ FP8 TFLOPS\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">轻量级设计，无重依赖\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">采用 Just-In-Time（JIT）编译\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">核心代码仅 300 行，但超越专家调优的核函数\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">兼容稠密布局及两种 MoE 布局\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003C/ul>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">📥 GitHub:\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> https://github.com/deepseek-ai/DeepGEMM\u003C/span>\u003C/p>\u003Ch4 style='box-sizing: border-box;margin: 30px 0px 15px;color: rgba(0, 0, 0, 0.85);font-weight: 500;;padding: 0px;display: block;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan style=\"box-sizing: border-box;;font-size: 18px;color: rgb(0, 0, 0);line-height: 1.5em;letter-spacing: 0em;text-align: left;font-weight: bold;display: block;\">\u003Cspan leaf=\"\">解析：\u003C/span>\u003C/span>\u003C/h4>\u003Csection style='box-sizing: border-box;;margin: 20px 0px;padding: 10px 20px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgb(250, 250, 250);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Csection style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">\u003Cspan textstyle=\"\" style=\"font-weight: bold;\">AI 计算的“高速公路” \u003C/span>\u003C/span>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">DeepGEMM 类似于 AI 计算世界中的“超级引擎”，确保数据在神经网络之间高效流动，就像 未来城市中的高效交通系统，消除计算瓶颈，让 AI 训练和推理更快。\u003C/span>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/p>\u003C/section>\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">\u003Cspan textstyle=\"\" style=\"font-weight: bold;\">超高计算性能\u003C/span>\u003C/span>\u003C/p>\u003C/section>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">1350+ TFLOPS（每秒 1.35 千万亿次计算），专为 H800 GPU 设计。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">支持 Dense 和 Mixture of Experts（MoE） 计算，可高效运行 DeepSeek-V3 和 R1 等大模型。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">仅 300 行核心代码，设计精简，但性能超越许多 专家级优化的传统库。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">无繁重依赖，完全 JIT（即时编译），像“超跑”一样轻量高效。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">意义：\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> 相比传统库，DeepGEMM 占用更少资源，计算速度更快，更适合 AI 训练和推理任务。\u003C/span>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/p>\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">\u003Cspan textstyle=\"\" style=\"font-weight: bold;\">FP8 精度，降低资源消耗\u003C/span>\u003C/span>\u003C/section>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">采用 FP8（8 位浮点数），比 FP32 和 BF16 占用更少内存，计算速度更快。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">适用于 大规模 AI 训练，即使模型参数达 数十亿 级别，也能高效运行。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">在 中国市场定制优化，让 AI 训练更加经济高效。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">意义\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">：FP8 降低了 AI 计算成本，同时保持足够的模型精度，使得更大规模的 AI 模型可以在有限的 GPU 资源上运行。\u003C/span>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/p>\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">\u003Cspan textstyle=\"\" style=\"font-weight: bold;\">适用于 Dense 和 MoE（专家网络）\u003C/span>\u003C/span>\u003C/section>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">Dense 计算：适用于标准 AI 任务，如 NLP 和 CV。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">MoE（Mixture of Experts）支持：\u003C/span>\u003C/section>\u003C/li>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">允许多个专家网络协同工作，提高大模型的计算效率。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">使 AI 能动态分配计算任务，适用于 对话 AI、科学计算等任务。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">意义\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">：DeepGEMM 让 MoE 模型更加高效，相比传统模型，计算速度更快、能耗更低。\u003C/span>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/p>\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">\u003Cspan textstyle=\"\" style=\"font-weight: bold;\">适合科研、企业和初创公司\u003C/span>\u003C/span>\u003C/section>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">提升 AI 训练效率：研究人员可以用更低的硬件成本，训练 高性能 AI 模型。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">降低 AI 计算门槛：适合 初创公司、学术研究者，无需昂贵硬件即可使用。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">即插即用（Plug & Play）：无复杂依赖，轻松集成到 AI 项目中。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">意义\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">：DeepGEMM 让 AI 开发更便宜、更快、更智能，特别适合资金有限的团队。\u003C/span>\u003C/p>\u003C/section>\u003C/section>\u003Ch4 style='box-sizing: border-box;margin: 30px 0px 15px;color: rgba(0, 0, 0, 0.85);font-weight: 500;;padding: 0px;display: block;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan style=\"box-sizing: border-box;;font-size: 18px;color: rgb(0, 0, 0);line-height: 1.5em;letter-spacing: 0em;text-align: left;font-weight: bold;display: block;\">\u003Cspan leaf=\"\">网友评论：\u003C/span>\u003C/span>\u003C/h4>\u003Cblockquote style='box-sizing: border-box;margin: 20px 0px;;padding: 10px 10px 10px 20px;border-style: none none none solid;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0.05);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;overflow: auto;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;font-weight: normal;\">\u003Cspan leaf=\"\">DeepGEMM 就像一个魔法玩具，让计算机以超快的速度完成数学运算！🚗💨\u003C/span>\u003C/p>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">速度\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">：就像 赛车 疾驰而过。🏎️\u003C/span>\u003C/p>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">简单\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">：使用起来轻松直观，就像在玩游戏。🎮\u003C/span>\u003C/p>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">高效\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">：节能省电，就像一盏明亮的灯泡。💡\u003C/span>\u003C/p>\u003C/section>\u003C/li>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;font-weight: normal;\">\u003Cspan leaf=\"\">DeepGEMM 让计算机更聪明、更快速地解决难题！🌟\u003C/span>\u003C/p>\u003C/blockquote>\u003Csection data-mpa-template=\"t\" mpa-from-tpl=\"t\" data-mpa-action-id=\"m7nbe5c219uh\">\u003Csection style=\"display: flex;flex-direction: column;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"display: flex;flex-direction: column;align-self: center;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"width: 143px;display: grid;grid-template-columns: 100% 100%;align-items: flex-start;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"display: flex;justify-content: center;align-items: center;width: 143px;height: 73px;\" data-mid=\"\" mpa-from-tpl=\"t\" nodeleaf=\"\">\u003Cimg class=\"rich_pages wxw-img\" data-ratio=\"0.5524475524475524\" data-w=\"286\" src=\"https://res.cooltool.vip/article_res/assets/17423769959160.8099276097183004.png\">\u003C/section>\u003Csection style=\"transform: translateX(-100.1%);padding: 19px 8px 0 8px;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"text-align: left;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Cp style=\"font-weight: bold;font-size: 14px;color: #2D2D2D;line-height: 15px;word-break: break-word;\" data-mid=\"\">\u003Cspan leaf=\"\">DAY 4\u003C/span>\u003C/p>\u003C/section>\u003Csection style=\"text-align: left;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Cp style=\"font-weight: bold;font-size: 17px;color: #2D2D2D;line-height: 21px;word-break: break-word;\" data-mid=\"\">\u003Cspan leaf=\"\">优化并行策略\u003C/span>\u003C/p>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003Ch4 style='box-sizing: border-box;margin: 30px 0px 15px;color: rgba(0, 0, 0, 0.85);font-weight: 500;;padding: 0px;display: block;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan style=\"box-sizing: border-box;;font-size: 18px;color: rgb(0, 0, 0);line-height: 1.5em;letter-spacing: 0em;text-align: left;font-weight: bold;display: block;\">\u003Cspan leaf=\"\">核心内容：\u003C/span>\u003C/span>\u003C/h4>\u003Cul style='box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;' class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">发布了两种并行优化策略及计算分析工具：\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Col style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: decimal;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">DualPipe\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> - 一种双向流水线并行算法，实现计算-通信重叠\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">EPLB\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> - 专为 V3/R1 设计的专家并行负载均衡器\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">V3/R1 计算-通信重叠分析工具\u003C/span>\u003C/strong>\u003C/section>\u003C/li>\u003C/ol>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">📥 GitHub:\u003C/span>\u003C/strong>\u003C/p>\u003Cul style='box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;' class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">DualPipe https://github.com/deepseek-ai/DualPipe\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">EPLB https://github.com/deepseek-ai/eplb\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">Profile Data https://github.com/deepseek-ai/profile-data)\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Ch4 style='box-sizing: border-box;margin: 30px 0px 15px;color: rgba(0, 0, 0, 0.85);font-weight: 500;;padding: 0px;display: block;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan style=\"box-sizing: border-box;;font-size: 18px;color: rgb(0, 0, 0);line-height: 1.5em;letter-spacing: 0em;text-align: left;font-weight: bold;display: block;\">\u003Cspan leaf=\"\">解析：\u003C/span>\u003C/span>\u003C/h4>\u003Csection style='box-sizing: border-box;;margin: 20px 0px;padding: 10px 20px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgb(250, 250, 250);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Csection style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">\u003Cspan textstyle=\"\" style=\"font-weight: bold;\">训练更快（Faster Training）\u003C/span>\u003C/span>\u003C/section>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">DualPipe 通过 计算与通信重叠，减少 GPU 等待时间，让训练更高效。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">比喻：就像接力赛跑，接力棒从不停止，计算任务与数据传输无缝衔接。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">影响：减少训练中的空闲时间，让模型训练比传统方式更快完成。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">📌 解读：在传统的 AI 训练中，GPU 计算和数据传输（通信）通常是分开执行的，导致 GPU 在等待数据时浪费大量算力。DualPipe 让这两个过程 并行执行，极大提高了计算效率。\u003C/span>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/p>\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">\u003Cspan textstyle=\"\" style=\"font-weight: bold;\">计算成本更低（Cost Savings）\u003C/span>\u003C/span>\u003C/section>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">优化计算效率，减少 GPU 需求，相比竞品 计算需求减少高达 11 倍。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">DeepSeek 仅使用 2,048 张 H800 GPU，而不是更昂贵的大规模集群。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">影响：减少硬件成本，降低 AI 训练的经济门槛。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">📌 解读：目前 OpenAI、Google、Meta 等公司依赖 H100 超大集群 进行 AI 训练，而 DeepSeek 通过 H800 + 计算优化，达成相似的训练性能，但 硬件成本大幅下降。这意味着 AI 训练变得更经济高效，让更多团队有能力训练大模型。\u003C/span>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/p>\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">\u003Cspan textstyle=\"\" style=\"font-weight: bold;\">可扩展性更强（Scalability）\u003C/span>\u003C/span>\u003C/section>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">DeepSeek V3 规模达到 6710 亿（671B）参数，依然能保持高效训练。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">模型变大，但训练时间和资源消耗不会爆炸式增长。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">影响：可以高效扩展超大规模 AI 模型，而不会因计算瓶颈而拖慢进度。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">📌 解读：一般来说，模型参数越大，训练时间和资源需求都会成倍增长。然而 DeepSeek 通过 DualPipe + EPLB，在扩展超大模型的同时，保持相对稳定的计算开销，解决了 AI 训练中的可扩展性问题。\u003C/span>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/p>\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">\u003Cspan textstyle=\"\" style=\"font-weight: bold;\">资源利用率更高（Resource Efficiency）\u003C/span>\u003C/span>\u003C/section>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">EPLB（专家并行负载均衡） 确保 每块 GPU 都在高效工作，避免闲置浪费。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">DualPipe 进一步 提升计算与通信的重叠度，最大化 GPU 计算能力。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">影响：减少 GPU 闲置，提高整体计算效率，提升 AI 训练吞吐量。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">📌 解读：很多 AI 训练任务在 GPU 资源调度不均衡时，可能导致部分 GPU 过载，而部分 GPU 长时间空闲。EPLB 通过 智能任务分配，确保 GPU 负载均衡，减少计算资源浪费。\u003C/span>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">与行业巨头的对比\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">（Comparison to Big Players）\u003C/span>\u003C/p>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">DeepSeek AI vs. OpenAI / Google / Meta：DeepSeek 采用 H800 + 高效优化，而其他巨头大多使用 H100 超级集群。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">强调“效率”而非“暴力计算”：DeepSeek 通过 DualPipe + EPLB，用更少的 GPU 资源训练同级别 AI 模型。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">比喻：类似 “大卫 vs. 歌利亚”，DeepSeek 通过 计算优化 取胜，而非仅靠硬件堆砌。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">📌 解读：目前大多数 AI 训练依赖 昂贵的 H100 超大集群，而 DeepSeek 选择了一条 低成本高效能 的道路，证明了优化算法可以弥补硬件差距。这使得 AI 训练更具普惠性，也让 计算资源更容易获取。\u003C/span>\u003C/p>\u003C/section>\u003C/section>\u003Ch4 style='box-sizing: border-box;margin: 30px 0px 15px;color: rgba(0, 0, 0, 0.85);font-weight: 500;;padding: 0px;display: block;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan style=\"box-sizing: border-box;;font-size: 18px;color: rgb(0, 0, 0);line-height: 1.5em;letter-spacing: 0em;text-align: left;font-weight: bold;display: block;\">\u003Cspan leaf=\"\">网友评论：\u003C/span>\u003C/span>\u003C/h4>\u003Cblockquote style='box-sizing: border-box;margin: 20px 0px;;padding: 10px 10px 10px 20px;border-style: none none none solid;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0.05);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;overflow: auto;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;font-weight: normal;\">\u003Cspan leaf=\"\">This is going to be very popular with companies with small budgets. Congratulations on this achievement; I look forward to V4 and R2. Hopefully, we will also see a video model outperform Veo2 very soon 🙏🙏🙏\u003C/span>\u003C/p>\u003C/blockquote>\u003Cblockquote style='box-sizing: border-box;margin: 20px 0px;;padding: 10px 10px 10px 20px;border-style: none none none solid;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0.05);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;overflow: auto;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;font-weight: normal;\">\u003Cspan leaf=\"\">Love the DualPipe diagram in the paper as well! I added the other 2 in the table for comparison - 1F1B and ZB1P\u003C/span>\u003C/p>\u003C/blockquote>\u003Csection style=\"text-align: center;\" nodeleaf=\"\">\u003Cimg class=\"rich_pages wxw-img js_insertlocalimg\" data-ratio=\"0.4852941176470588\" data-s=\"300,640\" data-type=\"jpeg\" data-w=\"680\" type=\"block\" data-imgfileid=\"100010420\" src=\"https://res.cooltool.vip/article_res/assets/17423769968580.5969464547477064.jpeg\">\u003C/section>\u003Csection>\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/section>\u003Csection data-mpa-template=\"t\" mpa-from-tpl=\"t\" data-mpa-action-id=\"m7nbe5c219uh\">\u003Csection style=\"display: flex;flex-direction: column;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"display: flex;flex-direction: column;align-self: center;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"width: 143px;display: grid;grid-template-columns: 100% 100%;align-items: flex-start;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"display: flex;justify-content: center;align-items: center;width: 143px;height: 73px;\" data-mid=\"\" mpa-from-tpl=\"t\" nodeleaf=\"\">\u003Cimg class=\"rich_pages wxw-img\" data-ratio=\"0.5524475524475524\" data-w=\"286\" src=\"https://res.cooltool.vip/article_res/assets/17423769959830.6007592177485148.png\">\u003C/section>\u003Csection style=\"transform: translateX(-100.1%);padding: 19px 8px 0 8px;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Csection style=\"text-align: left;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Cp style=\"font-weight: bold;font-size: 14px;color: #2D2D2D;line-height: 15px;word-break: break-word;\" data-mid=\"\">\u003Cspan leaf=\"\">DAY 5\u003C/span>\u003C/p>\u003C/section>\u003Csection style=\"text-align: left;\" data-mid=\"\" mpa-from-tpl=\"t\">\u003Cp style=\"font-weight: bold;font-size: 17px;color: #2D2D2D;line-height: 21px;word-break: break-word;\" data-mid=\"\">\u003Cspan leaf=\"\">预热\u003C/span>\u003C/p>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003C/section>\u003Ch4 style='box-sizing: border-box;margin: 30px 0px 15px;color: rgba(0, 0, 0, 0.85);font-weight: 500;;padding: 0px;display: block;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan style=\"box-sizing: border-box;;font-size: 18px;color: rgb(0, 0, 0);line-height: 1.5em;letter-spacing: 0em;text-align: left;font-weight: bold;display: block;\">\u003Cspan leaf=\"\">核心内容：\u003C/span>\u003C/span>\u003C/h4>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">3FS（Fire-Flyer File System）\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，这是一款 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">专为 AI 训练和推理优化的并行文件系统\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，具备超高吞吐量、强一致性架构，适用于 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">大规模数据访问和处理\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">。\u003C/span>\u003C/p>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">1. 3FS = AI 计算的“超高速数据通道”\u003C/span>\u003C/strong>\u003C/p>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan leaf=\"\">类似 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">AI 计算的涡轮增压引擎（Thruster）\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，加速 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">数据访问\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> 和 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">存储读写\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">。 采用 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">并行文件系统架构\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，最大化利用 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">SSD 带宽 & RDMA 网络\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，提供超高数据吞吐量。\u003C/span>\u003C/p>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">2. 超强吞吐性能\u003C/span>\u003C/strong>\u003C/p>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan leaf=\"\">🔥 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">6.6 TiB/s 读取吞吐量\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">（在 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">180 节点集群\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> 测试）\u003C/span>\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003Cspan leaf=\"\">🔥 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">3.66 TiB/min 处理速度\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">（在 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">25 节点 GraySort 测试\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">）\u003C/span>\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003Cspan leaf=\"\">🔥 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">40+ GiB/s 单节点峰值吞吐量\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">（用于 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">KVCache 查询\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">）\u003C/span>\u003C/p>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan leaf=\"\">📌 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">解读\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">：\u003C/span>\u003C/p>\u003Cul style='box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;' class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">6.6 TiB/s\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> 读取吞吐量远超传统分布式存储，确保 AI 训练时 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">数据流畅无阻\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">GraySort 基准测试\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> 证明其数据处理能力强，适合大规模数据预处理。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">KVCache 查询高吞吐量\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，使 AI 推理任务 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">更快\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">、\u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">更高效\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">3. 解耦架构 & 强一致性\u003C/span>\u003C/strong>\u003C/p>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan leaf=\"\">存储与计算分离，提升扩展性，避免单点瓶颈。保证数据一致性，适用于 AI 训练和推理任务。\u003C/span>\u003C/p>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">4. AI 训练 & 推理的全能存储支持\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> 3FS 适用于 AI 计算的多个关键场景，包括：训练数据预处理、数据集加载（Dataset Loading）、Checkpoint 存储/恢复（训练中断 & 继续）、Embedding 向量搜索（AI 语义检索）、KVCache 查询**（推理加速）\u003C/span>\u003C/p>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">5. 生态扩展：Smallpond\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> 🌊 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">Smallpond\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> 是基于 3FS 的 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">数据处理框架\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，专注于：\u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">高效数据处理\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">、\u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">优化存储管理\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">、\u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">适配 AI 训练 & 推理\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">。链接：https://github.com/deepseek-ai/smallpond\u003C/span>\u003C/p>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan leaf=\"\">🚀 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">DeepSeek 通过 3FS，让 AI 计算的数据流动更快、更智能，推动大模型训练和推理效率的全面提升！\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> 🔥\u003C/span>\u003C/p>\u003Cp style='box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: normal;text-align: left;text-indent: 0px;padding: 8px 0px;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;orphans: 2;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan leaf=\"\">📥 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">GitHub\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">: 3FS https://github.com/deepseek-ai/3FS\u003C/span>\u003C/p>\u003Ch4 style='box-sizing: border-box;margin: 30px 0px 15px;color: rgba(0, 0, 0, 0.85);font-weight: 500;;padding: 0px;display: block;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan style=\"box-sizing: border-box;;font-size: 18px;color: rgb(0, 0, 0);line-height: 1.5em;letter-spacing: 0em;text-align: left;font-weight: bold;display: block;\">\u003Cspan leaf=\"\">解析：\u003C/span>\u003C/span>\u003C/h4>\u003Csection style='box-sizing: border-box;;margin: 20px 0px;padding: 10px 20px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgb(250, 250, 250);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Csection style=\"box-sizing: border-box;;\">\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">3FS\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">（Fire-Flyer File System）是一款 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">高性能并行文件系统\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，专为 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">AI 训练 & 推理\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> 设计，提供\u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">超快数据访问\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，避免存储瓶颈。\u003C/span>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">💡 3FS 如何工作？\u003C/span>\u003C/strong>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">想象一个 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">超级图书馆\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> 📖，书籍（数据）存放在多个分馆（存储节点）。研究者（计算节点）可以\u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">随时借阅任何书籍\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，而且\u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">不受位置影响\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，因为 3FS 采用 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">高速传输通道（RDMA）\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，让数据访问\u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">更快、更顺畅\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">。\u003C/span>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">🚀 关键优势\u003C/span>\u003C/strong>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">✅ \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">超快数据访问\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">：\u003C/span>\u003C/p>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">6.6 TiB/s 读取吞吐量\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，适配超大规模 AI 数据训练。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">避免数据加载瓶颈，让 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">AI 训练和推理更流畅\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">✅ \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">可扩展性强\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">：\u003C/span>\u003C/p>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">存储 & 计算解耦\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，可以轻松扩展 AI 训练规模。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">适用于 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">企业 & 研究级 AI 模型\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，支持大规模计算集群。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">✅ \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">专为 AI 优化\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">：\u003C/span>\u003C/p>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">支持数据加载、Checkpoint 存储、推理缓存\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，提升 AI 训练效率。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">KVCache 高速查询\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，加速 AI 推理任务。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cspan leaf=\"\">✅ \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">数据一致性 & 可靠性\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">：\u003C/span>\u003C/p>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">所有计算节点访问相同数据\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，保证 AI 训练结果稳定。\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cspan leaf=\"\">采用 \u003C/span>\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">强一致性架构（CRAQ）\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">，避免数据版本不匹配问题。\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003Csection style=\"box-sizing:border-box;cursor:pointer;margin-top:5px;margin-bottom:5px;color:rgb(1, 1, 1);font-size:16px;line-height:1.8em;letter-spacing:0em;text-align:left;font-weight:normal;\">\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003C/section>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">🎯 总结\u003C/span>\u003C/strong>\u003C/p>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">3FS = AI 计算的“数据高速公路”\u003C/span>\u003C/strong>\u003Cspan leaf=\"\"> 🚗💨\u003C/span>\u003C/p>\u003Cul style=\"box-sizing: border-box;margin: 8px 0px;;list-style-type: disc;padding: 0px 0px 0px 25px;color: rgb(0, 0, 0);\" class=\"list-paddingleft-1\">\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">更快\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">（6.6 TiB/s 读取速度，避免数据瓶颈）\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">更稳\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">（强一致性，保证 AI 训练结果准确）\u003C/span>\u003C/section>\u003C/li>\u003Cli style=\"box-sizing: border-box;;\">\u003Csection style=\"box-sizing: border-box;;margin-top: 5px;margin-bottom: 5px;color: rgb(1, 1, 1);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;font-weight: normal;\">\u003Cstrong style=\"box-sizing: border-box;font-weight: bold;;color: rgb(0, 0, 0);background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0);width: auto;height: auto;margin: 0px;padding: 0px;border-style: none;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;\">\u003Cspan leaf=\"\">更灵活\u003C/span>\u003C/strong>\u003Cspan leaf=\"\">（适配 AI 训练、推理、存储需求）\u003C/span>\u003C/section>\u003C/li>\u003C/ul>\u003C/section>\u003C/section>\u003Ch4 style='box-sizing: border-box;margin: 30px 0px 15px;color: rgba(0, 0, 0, 0.85);font-weight: 500;;padding: 0px;display: block;font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;background-color: rgb(255, 255, 255);text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cspan style=\"box-sizing: border-box;;font-size: 18px;color: rgb(0, 0, 0);line-height: 1.5em;letter-spacing: 0em;text-align: left;font-weight: bold;display: block;\">\u003Cspan leaf=\"\">网友评论：\u003C/span>\u003C/span>\u003C/h4>\u003Cblockquote style='box-sizing: border-box;margin: 20px 0px;;padding: 10px 10px 10px 20px;border-style: none none none solid;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0.05);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;overflow: auto;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;font-weight: normal;\">\u003Cspan leaf=\"\">Last day of DeepSeek cooking my Nvidia Stock. 😢\u003C/span>\u003C/p>\u003C/blockquote>\u003Cblockquote style='box-sizing: border-box;margin: 20px 0px;;padding: 10px 10px 10px 20px;border-style: none none none solid;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0.05);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;overflow: auto;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;font-weight: normal;\">\u003Cspan leaf=\"\">3FS and Smallpond are setting new benchmarks in AI data processing—blazing speeds and seamless integration.\u003C/span>\u003C/p>\u003C/blockquote>\u003Cblockquote style='box-sizing: border-box;margin: 20px 0px;;padding: 10px 10px 10px 20px;border-style: none none none solid;border-width: 3px;border-color: rgba(0, 0, 0, 0.4);border-radius: 0px;background: none 0% 0% / auto no-repeat scroll padding-box border-box rgba(0, 0, 0, 0.05);width: auto;height: auto;box-shadow: rgba(0, 0, 0, 0) 0px 0px 0px 0px;display: block;overflow: auto;color: rgb(0, 0, 0);font-family: Optima, \"Microsoft YaHei\", PingFangSC-regular, serif;font-size: 16px;font-style: normal;font-variant-ligatures: normal;font-variant-caps: normal;font-weight: 400;letter-spacing: normal;orphans: 2;text-align: left;text-indent: 0px;text-transform: none;widows: 2;word-spacing: 0px;-webkit-text-stroke-width: 0px;white-space: normal;text-decoration-thickness: initial;text-decoration-style: initial;text-decoration-color: initial;'>\u003Cp style=\"box-sizing: border-box;margin: 0px;;color: rgb(0, 0, 0);font-size: 16px;line-height: 1.8em;letter-spacing: 0em;text-align: left;text-indent: 0em;padding: 8px 0px;font-weight: normal;\">\u003Cspan leaf=\"\">Incredible, I was expecting that.👏\u003C/span>\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003Cspan leaf=\"\">Congratulations guys🎉. These days, you guys deliver real value that no one else delivers.\u003C/span>\u003Cspan leaf=\"\">\u003Cbr>\u003C/span>\u003Cspan leaf=\"\">Nex, please, we need the Video model and V4+R2 🙏🙏🙏🙏, the whole world is waiting for you guys.\u003C/span>\u003C/p>\u003C/blockquote>\u003Cp style=\"display: none;\">\u003Cmp-style-type data-value=\"3\">\u003C/mp-style-type>\u003C/p>\u003C/div>",1752585425138]