[{"data":1,"prerenderedAt":175},["ShallowReactive",2],{"cms:global:en-US":3,"case-detail-en-deploy-deepseek-r1-model-without-gpu":153,"cms:website-seo:20":171},{"basic":4,"theme":8,"seo":9,"footer":10,"social":15,"headerItems":19,"footerItems":58,"policyItems":146,"logoUrl":7},{"siteName":5,"tagline":6,"logoUrl":7,"faviconUrl":6},"广东倾云科技有限公司","","https:\u002F\u002Fwww.qingyunclouds.com\u002Fapi\u002Fpublic\u002Fassets\u002F1\u002Ffile",{"primaryColor":6,"accentColor":6},{"title":6,"description":6,"keywords":6,"canonical":6,"robots":6,"ogImageUrl":6},{"description":11,"heroDescription":11,"copyright":12,"contactHeading":13,"showLanguageSwitcher":14,"showBackToTop":14},"A technology R&D and service-oriented enterprise with the core of \"AI+Internet\", focusing on AI scenario transformation, software system construction and network infrastructure.","© CopyRight 2026 广东倾云科技有限公司 版权所有","Cooperation",true,{"website":16,"email":6,"phone":6,"address":17,"postalCode":18},"https:\u002F\u002Fwww.qingyunclouds.com\u002F","Room 717, 7th Floor, Torch Building, Zhongshan Port Street, Zhongshan City, Guangdong Province, China","508400",[20,28,34,40,46,52],{"id":21,"label":22,"kind":23,"openInNewWindow":24,"target":25,"children":27},"d6ba614a-7d89-4c43-a8ce-1d36866f59bf","Home","external_url",false,{"url":26},"\u002F",[],{"id":29,"label":30,"kind":23,"openInNewWindow":24,"target":31,"children":33},"nav-about","Abouts Us",{"url":32},"\u002Fabout",[],{"id":35,"label":36,"kind":23,"openInNewWindow":24,"target":37,"children":39},"nav-news","News",{"url":38},"\u002Fnews",[],{"id":41,"label":42,"kind":23,"openInNewWindow":24,"target":43,"children":45},"nav-cases","Case Center",{"url":44},"\u002Fcases",[],{"id":47,"label":48,"kind":23,"openInNewWindow":24,"target":49,"children":51},"nav-join","Careers",{"url":50},"\u002Fjoin",[],{"id":53,"label":54,"kind":23,"openInNewWindow":24,"target":55,"children":57},"nav-contact","Contact Us",{"url":56},"\u002Fcontact",[],[59,91,122],{"id":60,"label":61,"kind":62,"openInNewWindow":24,"target":63,"children":64},"footer-solutions","Solution","group",{},[65,71,76,81,86],{"id":66,"label":67,"kind":23,"openInNewWindow":24,"target":68,"children":70},"footer-retail","Smart Retail",{"url":69},"\u002F#solutions",[],{"id":72,"label":73,"kind":23,"openInNewWindow":24,"target":74,"children":75},"footer-office","Smart Office",{"url":69},[],{"id":77,"label":78,"kind":23,"openInNewWindow":24,"target":79,"children":80},"footer-city","Smart City",{"url":69},[],{"id":82,"label":83,"kind":23,"openInNewWindow":24,"target":84,"children":85},"footer-consumer","Intelligent C-end",{"url":69},[],{"id":87,"label":88,"kind":23,"openInNewWindow":24,"target":89,"children":90},"footer-tobacco","Digitalization of the tobacco industry",{"url":69},[],{"id":92,"label":93,"kind":62,"openInNewWindow":24,"target":94,"children":95},"footer-products","Products and Services",{},[96,101,106,111,117],{"id":97,"label":98,"kind":23,"openInNewWindow":24,"target":99,"children":100},"footer-llm","Development of Large Model Applications",{"url":69},[],{"id":102,"label":103,"kind":23,"openInNewWindow":24,"target":104,"children":105},"footer-ai","AI solution customization",{"url":69},[],{"id":107,"label":108,"kind":23,"openInNewWindow":24,"target":109,"children":110},"footer-software","Industry software customization",{"url":69},[],{"id":112,"label":113,"kind":23,"openInNewWindow":14,"target":114,"children":116},"footer-video","Video Creation Platform",{"url":115},"https:\u002F\u002Fwww.xuyingai.com\u002F",[],{"id":118,"label":119,"kind":23,"openInNewWindow":24,"target":120,"children":121},"footer-compute","Computing power server",{"url":69},[],{"id":123,"label":124,"kind":62,"openInNewWindow":24,"target":125,"children":126},"footer-company","Join us",{},[127,133,138,142],{"id":128,"label":129,"kind":23,"openInNewWindow":24,"target":130,"children":132},"footer-about","Company Profile",{"url":131},"\u002F#about",[],{"id":134,"label":135,"kind":23,"openInNewWindow":24,"target":136,"children":137},"footer-cases","Customer Case",{"url":44},[],{"id":139,"label":36,"kind":23,"openInNewWindow":24,"target":140,"children":141},"footer-news",{"url":38},[],{"id":143,"label":48,"kind":23,"openInNewWindow":24,"target":144,"children":145},"footer-join",{"url":50},[],[147],{"id":148,"label":149,"kind":23,"openInNewWindow":14,"target":150,"children":152},"5401443a-f196-4ff4-a970-5f6b79b6ae79","粤ICP备2021127448号",{"url":151},"https:\u002F\u002Fbeian.miit.gov.cn\u002F#\u002FIntegrated\u002Findex",[],{"id":154,"slug":155,"path":156,"title":157,"description":158,"date":159,"category":160,"cover":161,"client":6,"period":6,"tags":162,"metrics":163,"body":164,"seo":165,"relatedCases":170},8,"deploy-deepseek-r1-model-without-gpu","\u002Fcases\u002Fdeploy-deepseek-r1-model-without-gpu","零门槛部署满血版deepseek-r1模型，无显卡也可部署","提供了一份零门槛部署 DeepSeek-R1 大模型的实操教程。文章详细讲解了在无需高性能显卡的硬件条件下，如何选择合适的动态量化模型版本，并给出了从环境准备到命令行运行的具体步骤，让普通配置设备也能顺畅运行私有化 AI 模型。","2025-02-16T10:09:01.848+08:00","企业大模型","https:\u002F\u002Fwww.qingyunclouds.com\u002Fapi\u002Fpublic\u002Fassets\u002F119\u002Ffile",[],[],"\u003Cp style=\"text-align: justify\">废话不多说直接开始：\u003C\u002Fp>\u003Cp style=\"text-align: justify\">1. 首先下载安装ollama，他可以让cpu与gpu共同配合使用，不用堆那么多显卡即可运行，单推理速度肯定是gpu来得快，后面我会讲解如何分配cpu与gpu内存，无显卡也可使用ollama，但推理效率极低\u003C\u002Fp>\u003Cp style=\"text-align: justify\">https:\u002F\u002Follama.com\u002Fdownload\u003C\u002Fp>\u003Cp style=\"text-align: center\">\u003Cimg src=\"https:\u002F\u002Fwww.qingyunclouds.com\u002Fapi\u002Fpublic\u002Fassets\u002F118\u002Ffile\" alt=\"01.png\">\u003C\u002Fp>\u003Cp style=\"text-align: left\">\u003Cstrong>普通蒸馏过的量化版本模型可以通过ollama run deepseek-r1:xxb 即可一键运行，这里主要介绍满血如何构建ollama服务然后运行起来。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp style=\"text-align: left\">2. 下再需要部署的满血量化模型，推荐unsloth提供的动态量化模型版本，这里我们下周的是Q2_K_XL\u003C\u002Fp>\u003Cp style=\"text-align: center\">\u003Cimg src=\"https:\u002F\u002Fwww.qingyunclouds.com\u002Fapi\u002Fpublic\u002Fassets\u002F117\u002Ffile\" alt=\"02.png\">\u003C\u002Fp>\u003Cp style=\"text-align: left\">3. 把模型上传至服务器\u003C\u002Fp>\u003Cp style=\"text-align: center\">\u003Cimg src=\"https:\u002F\u002Fwww.qingyunclouds.com\u002Fapi\u002Fpublic\u002Fassets\u002F116\u002Ffile\" alt=\"03.png\">\u003C\u002Fp>\u003Cp style=\"text-align: left\">4. 下载llama.cpp工具，不同服务器环境下载不同版本，因为这里需要将这多个gguf模型文件合并成单个gguf，合并后才能给予ollama生成服务\u003C\u002Fp>\u003Cp style=\"text-align: left\">https:\u002F\u002Fgithub.com\u002Fggml-org\u002Fllama.cpp\u003C\u002Fp>\u003Cp style=\"text-align: center\">\u003Cimg src=\"https:\u002F\u002Fwww.qingyunclouds.com\u002Fapi\u002Fpublic\u002Fassets\u002F115\u002Ffile\" alt=\"04.png\">\u003C\u002Fp>\u003Cp style=\"text-align: left\">下载成功后上传服务器通过unzip解压，解压完成得到如下工具集，我们只需要使用llama-gguf-split工具即可\u003C\u002Fp>\u003Cp style=\"text-align: center\">\u003Cimg src=\"https:\u002F\u002Fwww.qingyunclouds.com\u002Fapi\u002Fpublic\u002Fassets\u002F114\u002Ffile\" alt=\"05.png\">\u003C\u002Fp>\u003Cp style=\"text-align: left\">在llama.cpp文件夹下执行如下命令，注意下面文件路径需要修改，00001-of-00005代表有多少个文件，最后得到DeepSeek-R1-UD-Q2_K_XL.gguf文件\u003C\u002Fp>\u003Cpre>\u003Ccode>.\u002Fllama-gguf-split  --merge \n\u002Fopt\u002Fds\u002Fmodel\u002FdeepseekR1\u002FDeepSeek-R1-UD-Q2_K_XL-00001-of-00005.gguf \n\u002Fopt\u002Fds\u002Fmodel\u002FdeepseekR1\u002FDeepSeek-R1-UD-Q2_K_XL.gguf\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>执行如下显示正在进行中\u003C\u002Fp>\u003Cp style=\"text-align: center\">\u003Cimg src=\"https:\u002F\u002Fwww.qingyunclouds.com\u002Fapi\u002Fpublic\u002Fassets\u002F113\u002Ffile\" alt=\"06.png\">\u003C\u002Fp>\u003Cp style=\"text-align: left\">5. 任意目录创建Modelfile文件,打开文件\u003C\u002Fp>\u003Cpre>\u003Ccode>touch Modelfile\nvim Modelfile\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>创建Modelfile文件内容\u003C\u002Fp>\u003Cpre>\u003Ccode>FROM \u002Fopt\u002Fds\u002Fmodel\u002FdeepseekR1\u002FDeepSeek-R1-UD-Q2_K_XL.gguf \u002F\u002F前面合并生成的gguf\nPARAMETER num_predict -1 \u002F\u002F不限制回答数量\nTEMPLATE "<｜User｜>{{ .System }} {{ .Prompt }}<｜Assistant｜>" \u002F\u002F提问模板\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>执行命令：\u003C\u002Fp>\u003Cpre>\u003Ccode>ollama create deepseek-r1 -f Modelfile\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>6. 修改ollama配置文件，运行外网方便，配置gpu占用100%\u003C\u002Fp>\u003Cpre>\u003Ccode>vim \u002Fetc\u002Fsystemd\u002Fsystem\u002Follama.service\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>注意这里CUDA_VISIBLE_DEVICES根据自己显卡来输入，我这里显卡8张则0-7编号可通过nvidia-smi查看\u003C\u002Fp>\u003Cp style=\"text-align: center\">\u003Cimg src=\"https:\u002F\u002Fwww.qingyunclouds.com\u002Fapi\u002Fpublic\u002Fassets\u002F112\u002Ffile\" alt=\"07.png\">\u003C\u002Fp>\u003Cp style=\"text-align: left\">再执行如下命令刷新服务\u003C\u002Fp>\u003Cpre>\u003Ccode>systemctl daemon-reload\nsystemctl restart ollama\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>7. ollama查看与运行服务\u003C\u002Fp>\u003Cp style=\"text-align: center\">\u003Cimg src=\"https:\u002F\u002Fwww.qingyunclouds.com\u002Fapi\u002Fpublic\u002Fassets\u002F111\u002Ffile\" alt=\"08.png\">\u003C\u002Fp>\u003Cp style=\"text-align: left\">运行任务：ollama run deepseek-r1:latest\u003C\u002Fp>\u003Cp style=\"text-align: center\">\u003Cimg src=\"https:\u002F\u002Fwww.qingyunclouds.com\u002Fapi\u002Fpublic\u002Fassets\u002F110\u002Ffile\" alt=\"09.png\">\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>",{"metaTitle":166,"metaDescription":167,"canonical":168,"robots":169,"ogImage":-1},"零门槛部署 DeepSeek-R1 模型：无需显卡即可轻松部署指南","如何在无需独立显卡配置的环境下，零门槛本地部署 DeepSeek-R1 大模型。包含模型版本选择、动态量化版本对比及详细步骤命令，助你轻松实现私有 AI 模型搭建。","https:\u002F\u002Fwww.qingyunclouds.com\u002Fcases\u002Fdeploy-deepseek-r1-model-without-gpu","index, follow",[],{"siteOrigin":172,"automaticCanonical":14,"managedOrigins":173},"https:\u002F\u002Fwww.qingyunclouds.com",[174],"https:\u002F\u002Fwww.gdqycloud.com",1794301314935]