{"id":46412,"date":"2025-07-30T03:07:36","date_gmt":"2025-07-29T19:07:36","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/46412.html"},"modified":"2025-07-30T03:07:36","modified_gmt":"2025-07-29T19:07:36","slug":"8%e5%8d%a1h20%e6%9c%8d%e5%8a%a1%e5%99%a8-vllm%e4%bc%81%e4%b8%9a%e7%ba%a7%e9%83%a8%e7%bd%b2%e6%bb%a1%e8%a1%80%e7%89%88deepseek%e5%ae%9e%e5%bd%95","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/46412.html","title":{"rendered":"8\u5361H20\u670d\u52a1\u5668 + vLLM\u4f01\u4e1a\u7ea7\u90e8\u7f72\u6ee1\u8840\u7248DeepSeek\u5b9e\u5f55"},"content":{"rendered":"<h2>\u524d\u8a00<\/h2>\n<p>\u6700\u8fd1\u7ed9\u4e00\u5bb6\u96c6\u56e2\u516c\u53f8\u90e8\u7f72\u672c\u5730\u5927\u6a21\u578b&#xff0c;\u4fbf\u5199\u4e0b\u672c\u6587\u8bb0\u5f55\u4e00\u4e0b\u4f01\u4e1a\u7ea7\u751f\u4ea7\u73af\u5883\u90e8\u7f72\u5927\u6a21\u578b\u7684\u5168\u8fc7\u7a0b\u3002<\/p>\n<p>\u6574\u4f53\u670d\u52a1\u5668\u914d\u7f6e\u5982\u4e0b&#xff1a;<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729190725-68891bed21982.png\" alt=\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \/><\/p>\n<h3>\u90e8\u7f72\u6846\u67b6\u9009\u578b<\/h3>\n<p>\u8003\u8651\u5230\u5bf9\u65b9\u4f01\u4e1a\u4eba\u6570\u8d85\u8fc72\u4e07&#xff0c;\u6240\u4ee5\u6211\u4eec\u5bf9\u4e8e\u5e76\u53d1\u6027\u3001\u7a33\u5b9a\u6027\u3001\u5b89\u5168\u6027\u90fd\u6709\u4e00\u5b9a\u7684\u8981\u6c42&#xff0c;\u5728\u7efc\u5408\u5bf9\u6bd4\u4e86\u4e00\u4e9b\u4e3b\u6d41\u7684\u63a8\u7406\u6846\u67b6\u540e&#xff0c;\u6700\u7ec8\u786e\u5b9a\u4e86vLLM\u3002\u5177\u4f53\u5404\u6846\u67b6\u7684\u7814\u7a76\u7ec6\u8282\u8fd9\u91cc\u5c31\u4e0d\u518d\u8d58\u8ff0\u4e86&#xff0c;\u611f\u5174\u8da3\u7684\u53ef\u4ee5\u641c\u7d22\u76f8\u5173\u8d44\u6599\u3002<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729190725-68891bed81429.png\" alt=\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \/><\/p>\n<p>\u518d\u9644\u4e0a\u51e0\u4e2a\u5176\u4ed6\u6846\u67b6\u7684\u5bf9\u6bd4\u56fe<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729190725-68891bedea33f.png\" alt=\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \/><\/p>\n<p>\u5728\u6b63\u5f0f\u4e0a\u673a\u4e4b\u524d&#xff0c;\u6211\u4eec\u505a\u4e86\u5927\u91cf\u7684\u8fdc\u7a0b\u90e8\u7f72\u6f14\u7ec3&#xff0c;\u6240\u4ee5\u8fc7\u7a0b\u8fd8\u7b97\u4e00\u5e06\u98ce\u987a\u3002<\/p>\n<p>\u672c\u6587\u5c31\u8bb0\u5f55\u4e0b\u672c\u6b21\u90e8\u7f72\u7684\u5168\u8fc7\u7a0b\u4ee5\u53ca\u9047\u5230\u7684\u5751\u3002\u5148\u4e0a\u4e24\u5f20\u90e8\u7f72\u6210\u529f\u7684\u56fe&#xff08;DS 8\u5361\u8dd1\u6ee1&#xff0c;Qwen\u7528\u4e864\u5361&#xff09;<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729190726-68891bee59032.png\" alt=\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \/><br \/>\n<img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729190726-68891beeec97a.png\" alt=\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \/><\/p>\n<h2>\u6b63\u6587<\/h2>\n<p>\u90e8\u7f72\u7684\u8fc7\u7a0b\u5206\u4e3a\u4e09\u6b65<\/p>\n<p>1. \u73af\u5883\u51c6\u5907 -&gt; 2. \u6b63\u5f0f\u5b89\u88c5 -&gt; 3. \u63a8\u7406\u901f\u5ea6\u6d4b\u8bd5<\/p>\n<h3>\u4e00\u3001\u73af\u5883\u51c6\u5907<\/h3>\n<p>&#8212;&#8212;&#8212;&#8212;&#8212;-<br \/>\nubuntu <span class=\"token number\">22.04<\/span><br \/>\npython <span class=\"token number\">3.12<\/span><br \/>\ncuda <span class=\"token number\">12.4<\/span><br \/>\npytorch <span class=\"token number\">2.5<\/span>.1<br \/>\nnividia driver <span class=\"token number\">550<\/span><br \/>\n&#8212;&#8212;&#8212;&#8212;&#8212;-<\/p>\n<p>\u64cd\u4f5c\u7cfb\u7edf\u9009\u62e9Ubuntu 22.04 LTS\u7248\u672c&#xff0c;\u5176\u4ed6\u5e93\u4ee5\u53ca\u9a71\u52a8\u7248\u672c\u5982\u4e0a&#xff08;\u7248\u672c\u6700\u597d\u5168\u90e8\u5bf9\u5e94&#xff0c;\u5426\u5219\u5bb9\u6613\u51fa\u95ee\u9898&#xff09;<\/p>\n<h4>1. \u5b89\u88c5GPU\u9a71\u52a8<\/h4>\n<p>\u5ba2\u6237\u670d\u52a1\u5668\u5e76\u672a\u9884\u88c5\u9a71\u52a8&#xff0c;\u6240\u4ee5\u6211\u4eec\u9700\u8981\u5c06NVIDIA\u9a71\u52a8\u5148\u88c5\u597d&#xff0c;\u6709\u4e24\u79cd\u5b89\u88c5\u65b9\u5f0f&#xff0c;\u4e00\u79cd\u662f\u4e0b\u8f7d\u597d\u5b98\u65b9\u9a71\u52a8\u540e\u8fd0\u884c\u5b89\u88c5\u7a0b\u5e8f&#xff0c;\u4e00\u79cd\u662f\u66f4\u65b0PPA\u6e90\u540eapt\u5b89\u88c5\u3002\u6211\u8fd9\u91cc\u6f14\u793a\u4e00\u4e0bapt\u7684\u65b9\u5f0f\u5b89\u88c5\u3002<\/p>\n<p>\u5b89\u88c5 NVIDIA \u9a71\u52a8<\/p>\n<p><span class=\"token function\">sudo<\/span> add-apt-repository ppa:graphics-drivers\/ppa  <span class=\"token comment\"># \u6dfb\u52a0\u7b2c\u4e09\u65b9PPA\u6e90<\/span><br \/>\n<span class=\"token function\">sudo<\/span> <span class=\"token function\">apt<\/span> update<br \/>\n<span class=\"token function\">sudo<\/span> <span class=\"token function\">apt<\/span> <span class=\"token function\">install<\/span> nvidia-driver-550  <span class=\"token comment\"># \u5b89\u88c5550\u9a71\u52a8<\/span><\/p>\n<p>\u5b89\u88c5CUDA Toolkit<\/p>\n<p><span class=\"token function\">wget<\/span> https:\/\/developer.download.nvidia.com\/compute\/cuda\/repos\/ubuntu2204\/x86_64\/cuda-keyring_1.1-1_all.deb<br \/>\n<span class=\"token function\">sudo<\/span> dpkg -i cuda-keyring_1.1-1_all.deb<br \/>\n<span class=\"token function\">sudo<\/span> <span class=\"token function\">apt-get<\/span> update<br \/>\n<span class=\"token function\">sudo<\/span> <span class=\"token function\">apt-get<\/span> -y <span class=\"token function\">install<\/span> cuda-toolkit-12-4<\/p>\n<p>\u67e5\u770b\u9a71\u52a8\u662f\u5426\u5b89\u88c5\u6210\u529f<\/p>\n<p>nvidia-smi<\/p>\n<p>\u6b64\u65f6\u5b89\u88c5\u6210\u529f&#xff0c;\u6211\u4eec\u770b\u5230\u67098\u5f20H20\u7684GPU&#xff0c;\u5355\u5f20143G\u663e\u5b58&#xff0c;\u5982\u4e0b\u56fe&#xff1a;<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729190727-68891bef90c4e.png\" alt=\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \/><\/p>\n<p>\u6b64\u65f6\u6211\u4eec\u8fd8\u8981\u518d\u770b\u4e00\u4e0bNVLink&#xff08;\u9ad8\u901f GPU \u4e92\u8054&#xff09;\u662f\u5426\u5df2\u7ecf\u8054\u901a<\/p>\n<p><span class=\"token comment\"># \u67e5\u770bGPU\u62d3\u6251\u56fe<\/span><br \/>\nnvidia-smi topo -m<br \/>\n<span class=\"token comment\"># \u67e5\u770bnvlink\u8fde\u63a5\u72b6\u6001<\/span><br \/>\nnvidia-smi nvlink &#8211;status<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729190728-68891bf09c32e.png\" alt=\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \/><\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729190729-68891bf1d48f5.png\" alt=\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \/><\/p>\n<p>\u6211\u4eec\u770b\u5230GPU\u4e4b\u95f4\u901a\u8fc718\u6761NVLink\u4e92\u8054&#xff0c;\u6574\u4f53\u5e26\u5bbd\u53ef\u8fbe 26.562 GB\/s * 18 &#061; 478.116 GB\/s<\/p>\n<p>\u81f3\u6b64\u9a71\u52a8\u548cGPU\u73af\u5883\u51c6\u5907\u5b8c\u6210<\/p>\n<h4>2. \u7528conda\u5b89\u88c5\u865a\u62df\u73af\u5883&#xff08;\u63a8\u8350&#xff09;<\/h4>\n<p>\u9a71\u52a8\u5b89\u88c5\u597d\u4e4b\u540e&#xff0c;\u6211\u4eec\u7528conda\u521b\u5efa\u4e00\u4e2apython\u865a\u62df\u73af\u5883&#xff0c;\u907f\u514d\u5404\u79cd\u7248\u672c\u7684\u517c\u5bb9\u6027\u95ee\u9898<\/p>\n<p><span class=\"token comment\"># \u4e0b\u8f7d Miniconda \u5b89\u88c5\u5305<\/span><br \/>\n<span class=\"token builtin class-name\">cd<\/span> ~ <span class=\"token operator\">&amp;&amp;<\/span> <span class=\"token function\">wget<\/span> https:\/\/repo.anaconda.com\/miniconda\/Miniconda3-latest-Linux-x86_64.sh<br \/>\n<span class=\"token comment\"># \u5b89\u88c5\u811a\u672c<\/span><br \/>\n<span class=\"token function\">bash<\/span> Miniconda3-latest-Linux-x86_64.sh<br \/>\n<span class=\"token comment\"># \u6fc0\u6d3b<\/span><br \/>\n<span class=\"token builtin class-name\">source<\/span> ~\/.bashrc<br \/>\nconda &#8211;version<\/p>\n<p>\u521b\u5efa\u65b0\u73af\u5883&#xff08;python&#061;3.12&#xff09;<\/p>\n<p>conda create -n py312 <span class=\"token assign-left variable\">python<\/span><span class=\"token operator\">&#061;<\/span><span class=\"token number\">3.12<\/span><br \/>\n<span class=\"token comment\"># \u9700\u8981conda init&#xff0c;\u5982\u679c\u6ca1\u6709\u5219\u8f93\u5165&#xff1a;source \/root\/miniconda3\/etc\/profile.d\/conda.sh<\/span><br \/>\nconda activate py312<\/p>\n<h3>\u4e8c\u3001\u5b89\u88c5vllm<\/h3>\n<p>\u4e0a\u9762\u7684\u73af\u5883\u90fd\u51c6\u5907\u597d\u540e&#xff0c;\u8fdb\u5165\u865a\u62df\u73af\u5883&#xff0c;\u5b89\u88c5vllm<\/p>\n<p>pip <span class=\"token function\">install<\/span> vllm<\/p>\n<p>\u7ecf\u8fc7\u4e00\u6bb5\u65f6\u95f4\u540evllm\u5b8c\u6210\u5b89\u88c5&#xff08;\u5927\u7ea6\u67092-3\u4e2aG&#xff09;<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729190730-68891bf2c7365.png\" alt=\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \/><\/p>\n<h3>\u4e09\u3001\u4e0b\u8f7d\u6a21\u578b<\/h3>\n<p>\u56e0\u4e3a\u90e8\u7f72\u7684\u662f\u6ee1\u8840\u7684DeepSeek&#xff0c;\u6240\u4ee5\u4e24\u4e2a\u6a21\u578b\u6587\u4ef6\u5c31\u67091T\u591a&#xff0c;\u6240\u4ee5\u6211\u4eec\u63d0\u524d\u51c6\u5907\u4e864T\u7684\u79fb\u52a8\u786c\u76d8&#xff0c;\u5c06\u6a21\u578b\u4e0b\u8f7d\u597d&#xff0c;\u4ea4\u7ed9\u5bf9\u65b9\u7684\u8fd0\u7ef4\u8fdb\u884c\u62f7\u8d1d\u3002<\/p>\n<p>\u6211\u4eec\u5148\u4f7f\u7528\u5c06\u6a21\u578b\u6587\u4ef6\u4e0b\u8f7d\u5230\u786c\u76d8\u91cc&#xff0c;\u8fd9\u91cc\u4f7f\u7528\u56fd\u5185\u7684\u9b54\u642d\u8fdb\u884c\u4e0b\u8f7d&#xff08;Model Scope&#xff09;<\/p>\n<p>\u5148\u770b\u4e00\u4e0b\u78c1\u76d8\u7a7a\u95f4&#xff0c;\u53ef\u4ee5\u770b\u5230\u67096T\u7684\u6570\u636e\u7a7a\u95f4&#xff0c;\u591f\u7528\u4e86<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729190731-68891bf37a1a3.png\" alt=\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \/><\/p>\n<h5>\u5f00\u59cb\u4e0b\u8f7d<\/h5>\n<p>pip <span class=\"token function\">install<\/span> modelscope<br \/>\n<span class=\"token comment\"># \u4f7f\u7528 &#8211;local_dir\u53c2\u6570\u4e0b\u8f7d\u5230\u6307\u5b9a\u76ee\u5f55<\/span><br \/>\nmodelscope download &#8211;model deepseek-ai\/DeepSeek-R1-0528 &#8211;local_dir \/data\/models\/DeepSeek-R1-0528<\/p>\n<p>\u5168\u90e8\u4e0b\u8f7d\u597d\u540e&#xff0c;\u5927\u6982\u67091T\u7684\u5360\u7528&#xff0c;\u6211\u8fd9\u91cc\u8fd8\u4e0b\u8f7d\u4e86\u522b\u7684\u6a21\u578b&#xff0c;\u6240\u4ee5\u663e\u793a1.8T&#xff0c;\u5982\u56fe&#xff1a;<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729190732-68891bf4572a8.png\" alt=\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \/><\/p>\n<h3>\u56db\u3001vllm\u542f\u52a8\u5927\u6a21\u578b<\/h3>\n<p>\u73b0\u5728\u6211\u4eec\u7684\u9a71\u52a8\u5b89\u88c5\u597d\u4e86&#xff0c;\u6a21\u578b\u4e0b\u8f7d\u597d\u4e86&#xff0c;\u4e14\u5df2\u7ecf\u5728py312\u7684\u865a\u62df\u73af\u5883\u4e2d&#xff0c;\u4e07\u4e8b\u5177\u5907\u53ea\u6b20\u4e1c\u98ce&#xff0c;\u6b64\u65f6\u8fd0\u884c\u4e00\u4e2a\u6700\u5c0f\u5316\u6267\u884c\u547d\u4ee4&#xff0c;\u770b\u770bvllm\u662f\u5426\u80fd\u6210\u529f\u542f\u52a8<\/p>\n<p><span class=\"token comment\"># \u8fd0\u884c<\/span><br \/>\nvllm serve \/data\/packages\/models\/DeepSeek\/DeepSeek-V3-0324\/ <span class=\"token punctuation\">\\\\<\/span><br \/>\n    &#8211;tensor-parallel-size<span class=\"token operator\">&#061;<\/span><span class=\"token number\">8<\/span> <span class=\"token punctuation\">\\\\<\/span> <span class=\"token comment\"># \u5f20\u91cf\u5e76\u884c&#xff0c;GPU\u6570<\/span><br \/>\n    &#8211;served-model-name DeepSeek-V3-0324 <span class=\"token punctuation\">\\\\<\/span> <span class=\"token comment\"># \u6a21\u578b\u540d\u79f0<\/span><\/p>\n<p>\u7b49\u5f853-5\u5206\u949f\u540e&#xff0c;\u542f\u52a8\u6210\u529f&#xff01;\u6211\u4eec\u770b\u5230\u542f\u52a8\u6ee1\u8840\u7248DeepSeek-V3-0324&#xff0c;\u5355\u5361\u5360\u7528\u4e86132g\u7684\u663e\u5b58\u3002<\/p>\n<p>\u540c\u65f6vLLM\u4f1a\u63d0\u4f9b\u4e00\u4e2a\u517c\u5bb9OpenAI API\u683c\u5f0f\u7684\u63a5\u53e3&#xff0c;\u9ed8\u8ba4\u7aef\u53e3\u4e3a8000<\/p>\n<p>\u81f3\u6b64&#xff0c;\u6211\u4eec\u7684\u5927\u6a21\u578b\u670d\u52a1\u5c31\u901a\u8fc7vLLM\u6210\u529f\u542f\u52a8\u8d77\u6765\u4e86&#xff0c;\u5f53\u7136\u4e3a\u4e86\u5728\u751f\u4ea7\u73af\u5883\u6301\u4e45\u5316\u90e8\u7f72&#xff0c;\u8fd8\u9700\u8981\u505a\u4e00\u4e9b\u53c2\u6570\u4f18\u5316<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729190732-68891bf4d954e.png\" alt=\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \/><\/p>\n<h3>\u4e94\u3001\u6301\u4e45\u5316\u90e8\u7f72<\/h3>\n<h4>\u4e00\u3001\u7528nohup\u8fd0\u884c<\/h4>\n<p><span class=\"token function\">nohup<\/span> vllm serve \/data\/packages\/models\/DeepSeek\/DeepSeek-V3-0324\/ <span class=\"token punctuation\">\\\\<\/span><br \/>\n    &#8211;tensor-parallel-size<span class=\"token operator\">&#061;<\/span><span class=\"token number\">8<\/span> <span class=\"token punctuation\">\\\\<\/span> <span class=\"token comment\"># \u5f20\u91cf\u5e76\u884c&#xff0c;GPU\u6570<\/span><br \/>\n    &#8211;served-model-name DeepSeek-V3-0324 <span class=\"token punctuation\">\\\\<\/span> <span class=\"token comment\"># \u6a21\u578b\u540d\u79f0<\/span><br \/>\n    &#8211;max-model-len <span class=\"token number\">8192<\/span> <span class=\"token punctuation\">\\\\<\/span> <span class=\"token comment\"># \u9650\u5236token\u957f\u5ea6&#xff08;\u53ef\u9009&#xff09;<\/span><br \/>\n    <span class=\"token operator\">&gt;<\/span> \/logs\/vllm.log <span class=\"token operator\"><span class=\"token file-descriptor important\">2<\/span>&gt;<\/span><span class=\"token file-descriptor important\">&amp;1<\/span> <span class=\"token operator\">&amp;<\/span> <span class=\"token comment\"># \u8f93\u51fa\u65e5\u5fd7<\/span><\/p>\n<h4>\u4e8c\u3001\u7528docker\u8fd0\u884c<\/h4>\n<h5>1. \u5b89\u88c5docker<\/h5>\n<p><span class=\"token comment\"># \u8fd0\u884cdocker\u5b89\u88c5\u811a\u672c<\/span><br \/>\n<span class=\"token function\">sh<\/span> install_docker.sh<\/p>\n<h5>2.2 \u5b89\u88c5NVIDIA Container Toolkit<\/h5>\n<p><span class=\"token comment\"># \u6dfb\u52a0\u6e90\u5e76\u5b89\u88c5NVIDIA Container Toolkit GPU\u7a7f\u900f\u8fc7\u53bb<\/span><br \/>\n<span class=\"token assign-left variable\">distribution<\/span><span class=\"token operator\">&#061;<\/span><span class=\"token variable\"><span class=\"token variable\">$(<\/span><span class=\"token builtin class-name\">.<\/span> \/etc\/os-release<span class=\"token punctuation\">;<\/span><span class=\"token builtin class-name\">echo<\/span> $ID$VERSION_ID<span class=\"token variable\">)<\/span><\/span><br \/>\n<span class=\"token function\">curl<\/span> -s -L https:\/\/nvidia.github.io\/nvidia-docker\/gpgkey <span class=\"token operator\">|<\/span> <span class=\"token function\">sudo<\/span> apt-key <span class=\"token function\">add<\/span> &#8211;<br \/>\n<span class=\"token function\">curl<\/span> -s -L https:\/\/nvidia.github.io\/nvidia-docker\/<span class=\"token variable\">$distribution<\/span>\/nvidia-docker.list <span class=\"token operator\">|<\/span> <span class=\"token function\">sudo<\/span> <span class=\"token function\">tee<\/span> \/etc\/apt\/sources.list.d\/nvidia-docker.list<br \/>\n<span class=\"token function\">sudo<\/span> <span class=\"token function\">apt-get<\/span> update <span class=\"token operator\">&amp;&amp;<\/span> <span class=\"token function\">sudo<\/span> <span class=\"token function\">apt-get<\/span> <span class=\"token function\">install<\/span> -y nvidia-container-toolkit<br \/>\n<span class=\"token function\">sudo<\/span> systemctl restart docker<\/p>\n<h5>2.3 \u914d\u7f6e Docker \u8fd0\u884c\u65f6<\/h5>\n<p><span class=\"token comment\"># \u751f\u6210 NVIDIA \u8fd0\u884c\u65f6\u914d\u7f6e<\/span><br \/>\n<span class=\"token function\">sudo<\/span> nvidia-ctk runtime configure &#8211;runtime<span class=\"token operator\">&#061;<\/span>docker<br \/>\n<span class=\"token comment\"># \u91cd\u542f Docker \u670d\u52a1<\/span><br \/>\n<span class=\"token function\">sudo<\/span> systemctl restart docker<\/p>\n<h4>3. \u62c9\u53d6\u955c\u50cf<\/h4>\n<p>\u9700\u8981\u6307\u5b9a\u53f7\u76f8\u5e94\u7248\u672c&#xff0c;\u907f\u514dCUDA\u7248\u672c\u548c\u9a71\u52a8\u5bf9\u5e94\u4e0d\u4e0a&#xff0c;\u6211\u8fd9\u91cc\u7528v0.8.5.post1<\/p>\n<p><span class=\"token comment\"># \u6307\u5b9av0.8.5.post1<\/span><br \/>\ndocker pull vllm\/vllm-openai:v0.8.5.post1<\/p>\n<h4>4. \u8fd0\u884cdocker<\/h4>\n<p>docker run &#8211;runtime nvidia &#8211;gpus all <span class=\"token punctuation\">\\\\<\/span><br \/>\n    -v \/root\/models:\/models <span class=\"token punctuation\">\\\\<\/span><br \/>\n    -p <span class=\"token number\">8000<\/span>:8000 -d <span class=\"token punctuation\">\\\\<\/span><br \/>\n    &#8211;ipc<span class=\"token operator\">&#061;<\/span>host <span class=\"token punctuation\">\\\\<\/span><br \/>\n    vllm\/vllm-openai:v0.8.5.post1 <span class=\"token punctuation\">\\\\<\/span><br \/>\n    &#8211;model \/models\/Qwen\/Qwen3-4B-FP8<\/p>\n<p>\u4ee5\u4e0b\u662f\u6da6\u8272\u540e\u7684\u7248\u672c&#xff0c;\u4f18\u5316\u4e86\u8bed\u8a00\u8868\u8fbe\u548c\u6280\u672f\u7ec6\u8282\u7684\u5448\u73b0&#xff0c;\u540c\u65f6\u4fdd\u6301\u4e86\u4e13\u4e1a\u6027\u548c\u53ef\u8bfb\u6027&#xff1a;<\/p>\n<hr \/>\n<h3>\u516d\u3001\u63a8\u7406\u901f\u5ea6\u6d4b\u8bd5<\/h3>\n<p>\u4e3a\u6d4b\u8bd5\u5927\u6a21\u578b\u7684\u63a8\u7406\u6027\u80fd&#xff0c;\u6211\u4eec\u4f7f\u7528EvalScope\u6d4b\u8bd5\u5de5\u5177&#xff0c;\u5b83\u662f\u7531\u9b54\u642d\u793e\u533a\u63a8\u51fa\u7684\u8bc4\u6d4b\u6846\u67b6&#xff0c;\u652f\u6301\u63a8\u7406\u901f\u5ea6\u538b\u6d4b&#xff08;\u5982TTFT\u3001TPS\u3001\u541e\u5410\u91cf&#xff09;&#xff0c;\u5185\u7f6e\u591a\u6570\u636e\u96c6&#xff08;MMLU\u3001CMMLU\u7b49&#xff09;&#xff0c;\u53ef\u672c\u5730\u6216API\u6d4b\u8bd5&#xff0c;\u751f\u6210\u53ef\u89c6\u5316\u62a5\u544a\u3002<\/p>\n<p>\u4e0b\u9762\u662f\u5177\u4f53\u7684\u6d4b\u8bd5\u6d41\u7a0b&#xff1a;<\/p>\n<h5>1. \u73af\u5883\u914d\u7f6e<\/h5>\n<p><span class=\"token comment\"># \u521b\u5efa\u4e13\u7528conda\u73af\u5883<\/span><br \/>\nconda create -n evalscope <span class=\"token assign-left variable\">python<\/span><span class=\"token operator\">&#061;<\/span><span class=\"token number\">3.10<\/span> -y<br \/>\nconda activate evalscope<\/p>\n<p><span class=\"token comment\"># \u5b89\u88c5\u8bc4\u6d4b\u5de5\u5177&#xff08;\u542b\u6027\u80fd\u6d4b\u8bd5\u7ec4\u4ef6&#xff09;<\/span><br \/>\npip <span class=\"token function\">install<\/span> <span class=\"token string\">&#039;evalscope[all]&#039;<\/span> evalscope<span class=\"token punctuation\">[<\/span>perf<span class=\"token punctuation\">]<\/span> -U<\/p>\n<h5>4.2 \u6d4b\u8bd5\u53c2\u6570\u8bbe\u8ba1<\/h5>\n<p>\u91c7\u7528EvalScope\u6027\u80fd\u6d4b\u8bd5\u6a21\u5757&#xff0c;\u901a\u8fc7\u4ee5\u4e0b\u914d\u7f6e\u6a21\u62df\u771f\u5b9e\u4e1a\u52a1\u573a\u666f&#xff1a;<\/p>\n<ul>\n<li>\n<p>API\u7aef\u70b9\u200b&#xff1a;\u672c\u5730\u90e8\u7f72\u7684OpenAI\u517c\u5bb9\u63a5\u53e3&#xff08;http:\/\/127.0.0.1:8000\/v1\/chat\/completions&#xff09;<\/p>\n<\/li>\n<li>\n<p>\u5e76\u53d1\u91cf\u200b&#xff1a;32\u7ebf\u7a0b\u5e76\u884c\u8bf7\u6c42<\/p>\n<\/li>\n<li>\n<p>\u6d4b\u8bd5\u8f6e\u6b21\u200b&#xff1a;\u6bcf\u7ebf\u7a0b20\u6b21\u8fed\u4ee3<\/p>\n<\/li>\n<li>\n<p>\u6d4b\u8bd5\u6570\u636e\u96c6\u200b&#xff1a;openqa&#xff08;\u5f00\u653e\u57df\u95ee\u7b54\u57fa\u51c6&#xff09;<\/p>\n<\/li>\n<li>\n<p>\u6d41\u5f0f\u8f93\u51fa\u200b&#xff1a;\u542f\u7528stream\u6a21\u5f0f<\/p>\n<\/li>\n<\/ul>\n<p>evalscope perf \\\\<br \/>\n    &#8211;url &#034;http:\/\/127.0.0.1:8000\/v1\/chat\/completions&#034; \\\\<br \/>\n    &#8211;parallel 32 \\\\<br \/>\n    &#8211;model \/data\/packages\/models\/DeepSeek\/DeepSeek-V3-0324 \\\\<br \/>\n    &#8211;number 20 \\\\<br \/>\n    &#8211;api openai \\\\<br \/>\n    &#8211;dataset openqa \\\\<br \/>\n    &#8211;stream<\/p>\n<h5>4.3 \u6d4b\u8bd5\u7ed3\u679c\u5206\u6790<\/h5>\n<p>\u57fa\u51c6\u6d4b\u8bd5\u603b\u7ed3&#xff08;32\u5e76\u53d1\/20\u8bf7\u6c42&#xff09;\u200b\u200b&#xff1a;<\/p>\n<ul>\n<li>\n<p>\u603b\u6d4b\u8bd5\u65f6\u95f4&#xff1a;39.53\u79d2<\/p>\n<\/li>\n<li>\n<p>\u8bf7\u6c42\u6210\u529f\u7387&#xff1a;100%&#xff08;20\/20&#xff09;<\/p>\n<\/li>\n<li>\n<p>\u8f93\u51faToken\u541e\u5410\u91cf&#xff1a;592.26 tok\/s<\/p>\n<\/li>\n<li>\n<p>\u603bToken\u541e\u5410\u91cf&#xff1a;607.06 tok\/s<\/p>\n<\/li>\n<li>\n<p>\u5e73\u5747\u9996Token\u5ef6\u8fdf&#xff08;TTFT&#xff09;&#xff1a;0.99\u79d2<\/p>\n<\/li>\n<li>\n<p>\u5e73\u5747\u8f93\u51faToken\u95f4\u9694&#xff08;ITL&#xff09;&#xff1a;0.0213\u79d2<\/p>\n<\/li>\n<li>\n<p>\u5e73\u5747\u8f93\u5165Token\u6570\/\u8bf7\u6c42&#xff1a;29.25<\/p>\n<\/li>\n<li>\n<p>\u5e73\u5747\u8f93\u51faToken\u6570\/\u8bf7\u6c42&#xff1a;1170.5<\/p>\n<\/li>\n<li>\n<p>\u541e\u5410\u80fd\u529b\u200b&#xff1a;\u7cfb\u7edf\u572832\u5e76\u53d1\u4e0b\u4fdd\u6301607.06 tok\/s\u7684\u603b\u541e\u5410\u91cf&#xff0c;\u5f88\u4e0d\u9519<\/p>\n<\/li>\n<li>\n<p>\u5ef6\u8fdf\u8868\u73b0\u200b&#xff1a;<\/p>\n<ul>\n<li>\n<p>\u9996Token\u5ef6\u8fdf\u7a33\u5b9a\u57281.02\u79d2&#xff08;P95&#xff09;<\/p>\n<\/li>\n<li>\n<p>Token\u95f4\u5ef6\u8fdf&#xff08;ITL&#xff09;\u63a7\u5236\u572822ms\u5de6\u53f3&#xff0c;\u6ee1\u8db3\u5b9e\u65f6\u4ea4\u4e92\u9700\u6c42<\/p>\n<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u7a33\u5b9a\u6027\u200b&#xff1a;\u4e0d\u540c\u767e\u5206\u4f4d\u6307\u6807\u6ce2\u52a8\u5c0f\u4e8e5%&#xff0c;\u8868\u73b0\u7a33\u5b9a<\/p>\n<\/li>\n<li>\n<p>\u8d44\u6e90\u5229\u7528\u7387\u200b&#xff1a;\u957f\u6587\u672c\u751f\u6210\u573a\u666f&#xff08;\u5e73\u57471170\u8f93\u51faToken\/\u8bf7\u6c42&#xff09;\u4e0b\u4ecd\u4fdd\u6301\u9ad8\u541e\u5410<\/p>\n<\/li>\n<\/ul>\n<p>\u6d4b\u8bd5\u622a\u56fe\u5982\u4e0b&#xff1a;<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729190733-68891bf5c4e30.png\" alt=\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \/><\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729190734-68891bf6bad75.png\" alt=\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \/><\/p>\n<h3>\u4e03\u3001\u95ee\u9898\u4e0e\u89e3\u51b3\u65b9\u6848<\/h3>\n<p>\u56e0\u4e3a\u6d89\u53ca\u5f88\u591a\u8f6f\u786c\u4ef6\u7684\u56e0\u7d20&#xff0c;\u90e8\u7f72\u8fc7\u7a0b\u4e2d\u96be\u514d\u4f1a\u9047\u5230\u4e00\u4e9b\u95ee\u9898&#xff0c;\u8fd9\u91cc\u6211\u4eec\u5217\u4e3e\u4e00\u4e9b\u95ee\u9898\u548c\u89e3\u51b3\u529e\u6cd5<\/p>\n<li>NVLink\u672a\u6b63\u786e\u8fde\u63a5<\/li>\n<p>\u6709\u7684\u65f6\u5019NVLink\u7684\u670d\u52a1\u53ef\u80fd\u672a\u542f\u52a8&#xff0c;\u6216\u8005nvlink\u7684\u7248\u672c\u548c\u9a71\u52a8\u7248\u672c\u4e0d\u4e00\u81f4<\/p>\n<p><span class=\"token comment\"># \u67e5\u770bnvlink\u72b6\u6001<\/span><br \/>\nsystemctl status nvidia-fabricmanager.service<\/p>\n<p>\u4f1a\u51fa\u73b0\u7c7b\u4f3cfabric manager NVIDIA GPU driver interface version 535.216.01 dont match with driver version 550.183.01\u8fd9\u79cd\u63d0\u793a&#xff0c;\u8fd9\u65f6\u6211\u4eec\u9700\u8981\u5c06\u4e24\u8005\u7684\u7248\u672c\u53f7\u7edf\u4e00&#xff0c;\u91cd\u65b0\u5b89\u88c5nvidia-fabricmanager\u6216\u8005\u9a71\u52a8\u7248\u672c<\/p>\n<ul>\n<li>\u663e\u5b58\u4e0d\u8db3&#xff08;OOM&#xff09;\u95ee\u9898<\/li>\n<\/ul>\n<p>\u63a8\u7406\u8fc7\u7a0b\u4e2d\u5982\u679c\u51fa\u73b0CUDA out of memory\u9519\u8bef&#xff0c;\u53ef\u4ee5\u5c1d\u8bd5\u5982\u4e0b\u89e3\u51b3\u65b9\u6cd5&#xff1a;<\/p>\n<ul>\n<li>\n<p>\u964d\u4f4e\u6279\u6b21\u5927\u5c0f&#xff08;Batch Size&#xff09;<\/p>\n<\/li>\n<li>\n<p>\u4f7f\u7528\u534a\u7cbe\u5ea6&#xff08;FP16&#xff09;\u6216\u6df7\u5408\u7cbe\u5ea6\u63a8\u7406<\/p>\n<\/li>\n<li>\n<p>\u5c1d\u8bd5\u91cf\u5316\u6a21\u578b&#xff08;\u59828-bit\u62164-bit\u91cf\u5316&#xff09;<\/p>\n<\/li>\n<li>\n<p>\u8003\u8651\u4f7f\u7528\u66f4\u5c0f\u7684\u6a21\u578b\u53d8\u4f53<\/p>\n<\/li>\n<li>\n<p>\u53ca\u65f6\u6e05\u9664\u4e0d\u518d\u9700\u8981\u7684\u53d8\u91cf\u548c\u7f13\u5b58<\/p>\n<\/li>\n<li>\n<p>\u4f7f\u7528torch.cuda.empty_cache()\u91ca\u653e\u672a\u4f7f\u7528\u7684\u663e\u5b58<\/p>\n<\/li>\n<li>\n<p>Docker\u90e8\u7f72\u65f6&#xff0c;GPU\u672a\u6620\u5c04<\/p>\n<\/li>\n<\/ul>\n<ul>\n<li>\n<p>\u786e\u4fdd\u5b89\u88c5\u4e86NVIDIA Container Toolkit<\/p>\n<\/li>\n<li>\n<p>\u786e\u4fdd\u914d\u7f6e\u4e86docker\u8fd0\u884c\u65f6nvidia-ctk runtime configure &#8211;runtime&#061;docker<\/p>\n<\/li>\n<\/ul>\n<ul>\n<li>\u591aGPU\u8d1f\u8f7d\u4e0d\u5747\u8861<\/li>\n<\/ul>\n<p>\u95ee\u9898\u73b0\u8c61\u200b&#xff1a;<\/p>\n<ul>\n<li>\n<p>nvidia-smi\u663e\u793a\u5404GPU\u5229\u7528\u7387\u5dee\u5f02\u5927<\/p>\n<\/li>\n<li>\n<p>\u90e8\u5206GPU\u663e\u5b58\u5360\u7528\u660e\u663e\u9ad8\u4e8e\u5176\u4ed6\u5361<\/p>\n<\/li>\n<\/ul>\n<p>\u89e3\u51b3\u65b9\u6848\u200b&#xff1a;<\/p>\n<li>\u8c03\u6574\u5f20\u91cf\u5e76\u884c\u7b56\u7565&#xff1a;<\/li>\n<p>&#8211;tensor-parallel-size 4  # \u9700\u4e0eGPU\u6570\u91cf\u5339\u914d<\/p>\n<li>\u68c0\u67e5CUDA\u53ef\u89c1\u8bbe\u5907\u8bbe\u7f6e&#xff1a;<\/li>\n<p>export CUDA_VISIBLE_DEVICES&#061;0,1,2,3,4,5,6,7<\/p>\n<li>\u4f7f\u7528NCCL\u8c03\u8bd5\u5de5\u5177&#xff1a;<\/li>\n<p>nccl-tests\/build\/all_reduce_perf -b 8 -e 256M -f 2<\/p>\n<p>\u5efa\u8bae\u5728\u90e8\u7f72\u524d\u4f7f\u7528vllm.entrypoints.benchmark\u8fdb\u884c\u57fa\u51c6\u6d4b\u8bd5&#xff0c;\u63d0\u524d\u53d1\u73b0\u6027\u80fd\u74f6\u9888\u3002\u5178\u578b\u6d4b\u8bd5\u547d\u4ee4&#xff1a;<\/p>\n<p>python -m vllm.entrypoints.benchmark \\\\<br \/>\n    &#8211;model deepseek-ai\/deepseek-v3 \\\\<br \/>\n    &#8211;dataset huggingface:openai\/summarize_from_feedback \\\\<br \/>\n    &#8211;quantization awq \\\\<br \/>\n    &#8211;tensor-parallel-size 2<\/p>\n<h3>\u603b\u7ed3<\/h3>\n<p>\u672c\u6b21\u751f\u4ea7\u73af\u5883\u90e8\u7f72\u603b\u5171\u5386\u65f65\u5929&#xff0c;\u524d\u671f\u5927\u6a21\u578b\u6a21\u578b\u6587\u4ef6\u3001\u811a\u672c\u3001\u955c\u50cf\u7b49\u6587\u4ef6\u7684\u4e0b\u8f7d\u548c\u4f20\u8f93\u7528\u4e862\u5929&#xff0c;\u5b9e\u9645\u90e8\u7f72\u7528\u4e863\u5929&#xff0c;\u8fd9\u91cc\u7ed9\u4e00\u4e9b\u5b9e\u9645\u5efa\u8bae&#xff1a;<\/p>\n<li>\n<p>\u524d\u671f\u51c6\u5907&#xff1a;\u63d0\u524d\u51c6\u5907\u79fb\u52a8\u786c\u76d8&#xff08;\u81f3\u5c112T&#xff09;\u5c06\u5927\u6a21\u578b\u3001docker\u955c\u50cf\u4ee5\u53ca\u4e00\u4e9b\u81ea\u52a8\u5316\u811a\u672c\u62f7\u8d1d\u597d&#xff0c;\u53e6\u5916\u5fc5\u987b\u505a\u597d\u6d4b\u8bd5\u670d\u52a1\u5668\u7684\u5b89\u88c5\u6f14\u7ec3&#xff0c;\u5426\u5219\u5230\u4e86\u751f\u4ea7\u73af\u5883\u4f60\u4f1a\u624b\u5fd9\u811a\u4e71<\/p>\n<\/li>\n<li>\n<p>\u7f51\u7edc\u73af\u5883&#xff1a;\u672c\u6b21\u975e\u7eaf\u79bb\u7ebf\u5b89\u88c5&#xff0c;\u6240\u4ee5\u9700\u8981\u4fdd\u8bc1\u673a\u623f\u7684\u7f51\u7edc\u73af\u5883\u7a33\u5b9a<\/p>\n<\/li>\n<li>\n<p>\u786c\u4ef6\u4e0a&#xff1a;8 * H20&#xff08;141G&#xff09;\u53ef\u4ee5\u4fdd\u8bc1\u7a33\u5b9a\u8fd0\u884c\u6ee1\u8840DeepSeek&#xff0c;4 * H20\u53ef\u4ee5\u7a33\u5b9a\u8fd0\u884c\u6ee1\u8840Qwen3<\/p>\n<\/li>\n<li>\n<p>\u90e8\u7f72&#xff1a;\u4f7f\u7528vLLM\u90e8\u7f72&#xff0c;\u4e00\u5b9a\u8981\u521b\u5efa\u865a\u62df\u73af\u5883&#xff0c;\u6574\u4f53\u90e8\u7f72\u8fc7\u7a0b\u90fd\u6bd4\u8f83\u4e1d\u6ed1<\/p>\n<\/li>\n<li>\n<p>\u5e76\u53d1&#xff1a;32\u5e76\u53d1\u8fd8\u6709\u5f88\u5927\u7a7a\u95f4&#xff0c;100\u4ee5\u5185\u538b\u529b\u4e0d\u5927&#xff0c;\u9996token\u5ef6\u8fdf1\u79d2\u5de6\u53f3&#xff0c;\u5355\u8bf7\u6c42600token\/s<\/p>\n<\/li>\n<h3>\u6700\u540e<\/h3>\n<p>\u611f\u8c22\u5927\u5bb6\u770b\u5b8c\u5168\u6587&#xff0c;\u672c\u6587\u662f\u57fa\u4e8e\u6211\u5728\u751f\u4ea7\u73af\u5883\u90e8\u7f72DS\u7684\u4e00\u4e9b\u5b9e\u64cd\u8bb0\u5f55&#xff0c;\u5f53\u7136\u56e0\u4e3a\u90e8\u7f72\u65f6\u5728\u7532\u65b9\u516c\u53f8&#xff0c;\u6240\u4ee5\u5f88\u591a\u622a\u56fe\u548cBUG\u8bb0\u5f55\u6709\u6240\u9057\u6f0f&#xff0c;\u6574\u4f53\u4e0a\u8fd8\u662f\u5177\u6709\u53c2\u8003\u610f\u4e49\u7684&#xff0c;\u5e0c\u671b\u5bf9\u5927\u5bb6\u6709\u6240\u5e2e\u52a9\u548c\u542f\u53d1\u3002<\/p>\n<p>\u53e6\u5916&#xff0c;\u5927\u5bb6\u6709\u5927\u6a21\u578b\u3001AI\u667a\u80fd\u4f53\u76f8\u5173\u7684\u5546\u4e1a\u5408\u4f5c\u548c\u54a8\u8be2\u53ef\u4ee5\u548c\u6211\u804a\u804a\u54e6&#xff1a;lwbg66<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb825\u6b21\uff0c\u70b9\u8d5e18\u6b21\uff0c\u6536\u85cf12\u6b21\u3002\u6700\u8fd1\u7ed9\u4e00\u5bb6\u96c6\u56e2\u516c\u53f8\u90e8\u7f72\u672c\u5730\u5927\u6a21\u578b\uff0c\u4fbf\u5199\u4e0b\u672c\u6587\u8bb0\u5f55\u4e00\u4e0b**\u4f01\u4e1a\u7ea7\u751f\u4ea7\u73af\u5883\u90e8\u7f72\u5927\u6a21\u578b\u7684\u5168\u8fc7\u7a0b**\u3002<\/p>\n","protected":false},"author":2,"featured_media":46398,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[66,68,2681,132,603],"topic":[],"class_list":["post-46412","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-ai","tag-deepseek","tag-vllm","tag-132","tag-603"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v20.3 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>8\u5361H20\u670d\u52a1\u5668 + vLLM\u4f01\u4e1a\u7ea7\u90e8\u7f72\u6ee1\u8840\u7248DeepSeek\u5b9e\u5f55 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/46412.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"8\u5361H20\u670d\u52a1\u5668 + vLLM\u4f01\u4e1a\u7ea7\u90e8\u7f72\u6ee1\u8840\u7248DeepSeek\u5b9e\u5f55 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb825\u6b21\uff0c\u70b9\u8d5e18\u6b21\uff0c\u6536\u85cf12\u6b21\u3002\u6700\u8fd1\u7ed9\u4e00\u5bb6\u96c6\u56e2\u516c\u53f8\u90e8\u7f72\u672c\u5730\u5927\u6a21\u578b\uff0c\u4fbf\u5199\u4e0b\u672c\u6587\u8bb0\u5f55\u4e00\u4e0b**\u4f01\u4e1a\u7ea7\u751f\u4ea7\u73af\u5883\u90e8\u7f72\u5927\u6a21\u578b\u7684\u5168\u8fc7\u7a0b**\u3002\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/46412.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2025-07-29T19:07:36+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729190725-68891bed21982.png\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"4 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/46412.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/46412.html\",\"name\":\"8\u5361H20\u670d\u52a1\u5668 + vLLM\u4f01\u4e1a\u7ea7\u90e8\u7f72\u6ee1\u8840\u7248DeepSeek\u5b9e\u5f55 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2025-07-29T19:07:36+00:00\",\"dateModified\":\"2025-07-29T19:07:36+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/46412.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/46412.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/46412.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"8\u5361H20\u670d\u52a1\u5668 + vLLM\u4f01\u4e1a\u7ea7\u90e8\u7f72\u6ee1\u8840\u7248DeepSeek\u5b9e\u5f55\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"8\u5361H20\u670d\u52a1\u5668 + vLLM\u4f01\u4e1a\u7ea7\u90e8\u7f72\u6ee1\u8840\u7248DeepSeek\u5b9e\u5f55 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/46412.html","og_locale":"zh_CN","og_type":"article","og_title":"8\u5361H20\u670d\u52a1\u5668 + vLLM\u4f01\u4e1a\u7ea7\u90e8\u7f72\u6ee1\u8840\u7248DeepSeek\u5b9e\u5f55 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb825\u6b21\uff0c\u70b9\u8d5e18\u6b21\uff0c\u6536\u85cf12\u6b21\u3002\u6700\u8fd1\u7ed9\u4e00\u5bb6\u96c6\u56e2\u516c\u53f8\u90e8\u7f72\u672c\u5730\u5927\u6a21\u578b\uff0c\u4fbf\u5199\u4e0b\u672c\u6587\u8bb0\u5f55\u4e00\u4e0b**\u4f01\u4e1a\u7ea7\u751f\u4ea7\u73af\u5883\u90e8\u7f72\u5927\u6a21\u578b\u7684\u5168\u8fc7\u7a0b**\u3002","og_url":"https:\/\/www.wsisp.com\/helps\/46412.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2025-07-29T19:07:36+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250729190725-68891bed21982.png"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"4 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/46412.html","url":"https:\/\/www.wsisp.com\/helps\/46412.html","name":"8\u5361H20\u670d\u52a1\u5668 + vLLM\u4f01\u4e1a\u7ea7\u90e8\u7f72\u6ee1\u8840\u7248DeepSeek\u5b9e\u5f55 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2025-07-29T19:07:36+00:00","dateModified":"2025-07-29T19:07:36+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/46412.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/46412.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/46412.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"8\u5361H20\u670d\u52a1\u5668 + vLLM\u4f01\u4e1a\u7ea7\u90e8\u7f72\u6ee1\u8840\u7248DeepSeek\u5b9e\u5f55"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/46412","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=46412"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/46412\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/46398"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=46412"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=46412"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=46412"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=46412"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}