{"id":134,"date":"2026-08-29T15:55:24","date_gmt":"2026-08-29T07:55:24","guid":{"rendered":"https:\/\/www.xiaoxindong.cloud\/?p=134"},"modified":"2026-09-06T19:54:17","modified_gmt":"2026-09-06T11:54:17","slug":"gpu-%e5%a4%a7%e6%a8%a1%e5%9e%8b%e6%9c%ac%e5%9c%b0%e9%83%a8%e7%bd%b2%ef%bc%88%e5%ae%b9%e5%99%a8%e5%8c%96%e9%83%a8%e7%bd%b2%ef%bc%89","status":"publish","type":"post","link":"https:\/\/www.xiaoxindong.cloud\/?p=134","title":{"rendered":"GPU \u5927\u6a21\u578b\u672c\u5730\u90e8\u7f72\uff08\u5bb9\u5668\u5316\u90e8\u7f72\uff09"},"content":{"rendered":"\n<h2 class=\"wp-block-heading\">\u4e00\u3001Docker \u57fa\u7840\u4e0e AI \u5bb9\u5668\u5316\u4ef7\u503c<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">1. \u6838\u5fc3\u6982\u5ff5<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Docker \u662f\u5f00\u6e90\u5bb9\u5668\u5316\u5e73\u53f0\uff0c\u5c06\u5e94\u7528\u53ca\u5176\u4f9d\u8d56\u6253\u5305\u6210\u6807\u51c6\u5316\u5bb9\u5668\u5355\u5143\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u865a\u62df\u673a vs \u5bb9\u5668<\/strong>\uff1a\n<ul class=\"wp-block-list\">\n<li>\u865a\u62df\u673a\uff1a\u5b8c\u6574\u64cd\u4f5c\u7cfb\u7edf + \u5e94\u7528 + \u4f9d\u8d56\uff0c\u91cd\u91cf\u7ea7\uff08GB \u7ea7\uff09\uff0c\u542f\u52a8\u6162<\/li>\n\n\n\n<li>\u5bb9\u5668\uff1a\u4ec5\u5e94\u7528 + \u4f9d\u8d56\u5e93\uff0c\u8f7b\u91cf\u7ea7\uff08MB \u7ea7\uff09\uff0c\u5171\u4eab\u5bbf\u4e3b\u673a\u5185\u6838\uff0c\u542f\u52a8\u5feb<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">2. \u5bb9\u5668\u5316\u56db\u5927\u6838\u5fc3\u4ef7\u503c<\/h3>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>\u4ef7\u503c<\/th><th>\u8bf4\u660e<\/th><\/tr><\/thead><tbody><tr><td><strong>\u73af\u5883\u4e00\u81f4\u6027<\/strong><\/td><td>\u4e00\u6b21\u6784\u5efa\uff0c\u5904\u5904\u8fd0\u884c\uff1b\u5f00\u53d1 \/ \u6d4b\u8bd5 \/ \u751f\u4ea7\u73af\u5883\u5b8c\u5168\u4e00\u81f4\uff0c\u89e3\u51b3 \u201c\u6211\u8fd9\u80fd\u8dd1\u4f60\u90a3\u4e0d\u80fd\u8dd1\u201d \u95ee\u9898<\/td><\/tr><tr><td><strong>\u8d44\u6e90\u9694\u79bb\u6027<\/strong><\/td><td>\u4e0d\u540c\u9879\u76ee \/ \u6a21\u578b\u73af\u5883\u9694\u79bb\uff0c\u4e92\u4e0d\u5e72\u6270\uff1b\u53ef\u540c\u65f6\u8fd0\u884c\u4e0d\u540c CUDA\u3001Python \u7248\u672c\u7684\u670d\u52a1<\/td><\/tr><tr><td><strong>\u53ef\u79fb\u690d\u6027<\/strong><\/td><td>\u670d\u52a1\u5668\u3001\u4e91\u4e3b\u673a\u3001Kubernetes \u96c6\u7fa4\u5747\u53ef\u65e0\u7f1d\u8fc1\u79fb<\/td><\/tr><tr><td><strong>\u4ea4\u4ed8\u6548\u7387\u9ad8<\/strong><\/td><td>\u62c9\u53d6\u955c\u50cf\u5373\u53ef\u8fd0\u884c\uff0c\u90e8\u7f72\u3001\u542f\u52a8\u3001\u56de\u6eda\u901f\u5ea6\u5feb\uff0c\u652f\u6301 CI\/CD \u6d41\u6c34\u7ebf<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">3. AI \u670d\u52a1\u5bb9\u5668\u5316\u7684\u5fc5\u8981\u6027<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">AI \u6a21\u578b\u90e8\u7f72\u4f9d\u8d56\u94fe\u590d\u6742\uff1a\u64cd\u4f5c\u7cfb\u7edf \u2192 NVIDIA \u9a71\u52a8 \u2192 CUDA\/cuDNN \u2192 Python \u2192 PyTorch\/Transformers \u2192 \u6570\u5341\u4e2a\u4f9d\u8d56\u5e93\uff0c\u5404\u7ec4\u4ef6\u7248\u672c\u4e25\u683c\u8026\u5408\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u624b\u52a8\u914d\u7f6e\u6548\u7387\u4f4e\u3001\u4e00\u81f4\u6027\u5dee\uff0c\u591a\u9879\u76ee\u5171\u5b58\u6781\u6613\u4f9d\u8d56\u51b2\u7a81<\/li>\n\n\n\n<li>\u5bb9\u5668\u5316\u5c06\u6574\u5957\u73af\u5883\u6253\u5305\u6210\u955c\u50cf\uff0c\u5f7b\u5e95\u89e3\u51b3\u73af\u5883\u4f9d\u8d56\u4e0e\u590d\u73b0\u6027\u95ee\u9898\uff0c\u662f\u751f\u4ea7\u7ea7\u6a21\u578b\u670d\u52a1\u3001K8s \u90e8\u7f72\u7684\u6807\u51c6\u57fa\u7840\u8bbe\u65bd<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u4e8c\u3001Ubuntu 22.04 Docker \u5b89\u88c5\uff08\u2b50\uff09<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">\u5b89\u88c5\u6b65\u9aa4<\/h3>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u66f4\u65b0\u7cfb\u7edf\u5305\u7d22\u5f15<\/strong><\/li>\n<\/ol>\n\n\n\n<pre class=\"wp-block-code\"><code>apt update\n<\/code><\/pre>\n\n\n\n<ol start=\"2\" class=\"wp-block-list\">\n<li><strong>\u5b89\u88c5\u57fa\u7840\u4f9d\u8d56<\/strong><\/li>\n<\/ol>\n\n\n\n<pre class=\"wp-block-code\"><code>apt install apt-transport-https ca-certificates curl software-properties-common gnupg lsb-release\n<\/code><\/pre>\n\n\n\n<ol start=\"3\" class=\"wp-block-list\">\n<li><strong>\u6dfb\u52a0 Docker \u5b98\u65b9 GPG \u5bc6\u94a5<\/strong>\uff08\u56fd\u5185\u963f\u91cc\u4e91\u955c\u50cf\u6e90\uff09<\/li>\n<\/ol>\n\n\n\n<pre class=\"wp-block-code\"><code>curl -fsSL https:\/\/mirrors.aliyun.com\/docker-ce\/linux\/ubuntu\/gpg | sudo gpg --dearmor -o \/usr\/share\/keyrings\/docker-archive-keyring.gpg\n<\/code><\/pre>\n\n\n\n<ol start=\"4\" class=\"wp-block-list\">\n<li><strong>\u6dfb\u52a0 Docker APT \u8f6f\u4ef6\u6e90<\/strong><\/li>\n<\/ol>\n\n\n\n<pre class=\"wp-block-code\"><code>echo \"deb &#91;arch=$(dpkg --print-architecture) signed-by=\/usr\/share\/keyrings\/docker-archive-keyring.gpg] https:\/\/mirrors.aliyun.com\/docker-ce\/linux\/ubuntu $(lsb_release -cs) stable\" | sudo tee \/etc\/apt\/sources.list.d\/docker.list &gt; \/dev\/null\n<\/code><\/pre>\n\n\n\n<ol start=\"5\" class=\"wp-block-list\">\n<li><strong>\u66f4\u65b0\u7d22\u5f15\u5e76\u5b89\u88c5 Docker<\/strong><\/li>\n<\/ol>\n\n\n\n<pre class=\"wp-block-code\"><code>apt update\napt install docker-ce docker-ce-cli containerd.io\n<\/code><\/pre>\n\n\n\n<ol start=\"6\" class=\"wp-block-list\">\n<li><strong>\u542f\u52a8\u670d\u52a1\u5e76\u8bbe\u7f6e\u5f00\u673a\u81ea\u542f<\/strong><\/li>\n<\/ol>\n\n\n\n<pre class=\"wp-block-code\"><code>systemctl start docker\nsystemctl enable docker\n<\/code><\/pre>\n\n\n\n<ol start=\"7\" class=\"wp-block-list\">\n<li><strong>\u914d\u7f6e\u955c\u50cf\u52a0\u901f<\/strong> \u7f16\u8f91<code>\/etc\/docker\/daemon.json<\/code>\uff0c\u6dfb\u52a0\u56fd\u5185\u955c\u50cf\u6e90\uff1a<\/li>\n<\/ol>\n\n\n\n<pre class=\"wp-block-code\"><code>{\n  \"registry-mirrors\": &#91;\n    \"https:\/\/docker.1ms.run\/\",\n    \"https:\/\/docker.m.daocloud.io\",\n    \"https:\/\/docker.imgdb.de\",\n    \"https:\/\/docker-0.unsee.tech\",\n    \"https:\/\/docker.hlmirror.com\",\n    \"https:\/\/docker.xuanyuan.me\",\n    \"https:\/\/cjie.eu.org\",\n    \"https:\/\/rvjdf6u3.mirror.aliyuncs.com\"\n  ]\n}\n<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">\u91cd\u542f\u670d\u52a1\u751f\u6548\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>systemctl daemon-reload\nsystemctl restart docker\n<\/code><\/pre>\n\n\n\n<ol start=\"8\" class=\"wp-block-list\">\n<li><strong>\u9a8c\u8bc1\u5b89\u88c5<\/strong><\/li>\n<\/ol>\n\n\n\n<pre class=\"wp-block-code\"><code>docker info | grep -A 5 Registry  # \u67e5\u770b\u955c\u50cf\u6e90\u662f\u5426\u751f\u6548\ndocker --version\n<\/code><\/pre>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u4e09\u3001NVIDIA Container Toolkit\uff08\u2b50\uff09<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">1. \u6838\u5fc3\u4f5c\u7528<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u9ed8\u8ba4 Docker \u5bb9\u5668\u65e0\u6cd5\u8bbf\u95ee\u5bbf\u4e3b\u673a GPU \u786c\u4ef6\u3002NVIDIA Container Toolkit \u662f<strong>\u5bb9\u5668\u5f15\u64ce\u4e0e\u5bbf\u4e3b\u673a NVIDIA \u9a71\u52a8\u4e4b\u95f4\u7684\u6865\u6881<\/strong>\uff0c\u8fd0\u884c\u65f6\u52a8\u6001\u5c06 GPU \u8bbe\u5907\u8282\u70b9\u3001\u9a71\u52a8\u5e93\u6587\u4ef6\u6ce8\u5165\u5bb9\u5668\uff0c\u8ba9\u5bb9\u5668\u5185\u5e94\u7528\u53ef\u4ee5\u8c03\u7528 GPU\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">2. \u5de5\u4f5c\u539f\u7406<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u9a71\u52a8\u7559\u5728\u5bbf\u4e3b\u673a<\/strong>\uff1a\u4e0d\u628a\u9a71\u52a8\u6253\u5305\u8fdb\u955c\u50cf\uff0c\u9a71\u52a8\u5fc5\u987b\u5b89\u88c5\u5728\u5bbf\u4e3b\u673a OS \u4e0a<\/li>\n\n\n\n<li><strong>\u8fd0\u884c\u65f6\u52a8\u6001\u6ce8\u5165<\/strong>\uff1a\u5bb9\u5668\u542f\u52a8\u65f6\u901a\u8fc7 OCI Hook \u89e6\u53d1\uff0c\u81ea\u52a8\u8bc6\u522b\u5bbf\u4e3b\u673a\u7684<code>\/dev\/nvidia*<\/code>\u8bbe\u5907\u8282\u70b9\u3001\u9a71\u52a8\u5e93\u6587\u4ef6\uff0c\u6302\u8f7d\u5230\u5bb9\u5668\u9694\u79bb\u73af\u5883\u4e2d<\/li>\n\n\n\n<li>\u7c7b\u6bd4\uff1a\u5bb9\u5668\u662f\u5bc6\u5c01\u592a\u7a7a\u8231\uff0cToolkit \u662f\u667a\u80fd\u5bf9\u63a5\u7ba1\u9053\uff0c\u6309\u9700\u5bf9\u63a5 GPU \u201c\u8d44\u6e90\u63a5\u53e3\u201d\uff0c\u65e2\u4fdd\u8bc1\u9694\u79bb\u53c8\u80fd\u4f7f\u7528\u786c\u4ef6<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">3. \u5b89\u88c5\u914d\u7f6e\u6b65\u9aa4<\/h3>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u6dfb\u52a0 NVIDIA \u8f6f\u4ef6\u6e90\u4e0e GPG \u5bc6\u94a5<\/strong><\/li>\n<\/ol>\n\n\n\n<pre class=\"wp-block-code\"><code>distribution=$(. \/etc\/os-release;echo $ID$VERSION_ID)\ncurl -fsSL https:\/\/nvidia.github.io\/libnvidia-container\/gpgkey | sudo gpg --dearmor -o \/usr\/share\/keyrings\/nvidia-container-toolkit-keyring.gpg\ncurl -s -L \"https:\/\/nvidia.github.io\/libnvidia-container\/$distribution\/libnvidia-container.list\" | \\\nsed 's#deb https:\/\/#deb &#91;signed-by=\/usr\/share\/keyrings\/nvidia-container-toolkit-keyring.gpg] https:\/\/#' | \\\nsudo tee \/etc\/apt\/sources.list.d\/nvidia-container-toolkit.list\n<\/code><\/pre>\n\n\n\n<ol start=\"2\" class=\"wp-block-list\">\n<li><strong>\u5b89\u88c5\u5de5\u5177\u5305<\/strong><\/li>\n<\/ol>\n\n\n\n<pre class=\"wp-block-code\"><code>apt update\napt install -y nvidia-container-toolkit\n<\/code><\/pre>\n\n\n\n<ol start=\"3\" class=\"wp-block-list\">\n<li><strong>\u914d\u7f6e Docker \u8fd0\u884c\u65f6<\/strong><\/li>\n<\/ol>\n\n\n\n<pre class=\"wp-block-code\"><code>nvidia-ctk runtime configure --runtime=docker\nsystemctl restart docker  # \u5173\u952e\u6b65\u9aa4\uff0c\u914d\u7f6e\u540e\u5fc5\u987b\u91cd\u542fDocker\n<\/code><\/pre>\n\n\n\n<h3 class=\"wp-block-heading\">4. \u9a8c\u8bc1\u4e0e GPU \u53c2\u6570<\/h3>\n\n\n\n<h4 class=\"wp-block-heading\">\u57fa\u7840\u9a8c\u8bc1\uff1a\u5bb9\u5668\u5185\u6267\u884c nvidia-smi<\/h4>\n\n\n\n<pre class=\"wp-block-code\"><code>docker run --rm --name test --gpus all nvidia\/cuda:12.1.1-base-ubuntu22.04 nvidia-smi\n<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">\u8f93\u51fa\u4e0e\u5bbf\u4e3b\u673a\u4e00\u81f4\u5219\u8bf4\u660e GPU \u900f\u4f20\u6210\u529f\u3002<\/p>\n\n\n\n<h4 class=\"wp-block-heading\"><code>--gpus<\/code>\u53c2\u6570\u7528\u6cd5<\/h4>\n\n\n\n<ul class=\"wp-block-list\">\n<li><code>--gpus all<\/code>\uff1a\u6620\u5c04\u6240\u6709 GPU<\/li>\n\n\n\n<li><code>--gpus 2<\/code>\uff1a\u6620\u5c04\u4efb\u610f 2 \u5757 GPU<\/li>\n\n\n\n<li><code>--gpus '\"device=0,1\"'<\/code>\uff1a\u6307\u5b9a\u4f7f\u7528 0 \u53f7\u548c 1 \u53f7 GPU<\/li>\n\n\n\n<li><code>--gpus '\"device=GPU-UUID\"'<\/code>\uff1a\u901a\u8fc7 UUID \u7cbe\u786e\u6307\u5b9a<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">5. \u81ea\u5b9a\u4e49 GPU \u955c\u50cf\u6784\u5efa\u793a\u4f8b<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u6784\u5efa PyTorch GPU \u5e94\u7528\u955c\u50cf\uff0c\u63a8\u8350\u76f4\u63a5\u4f7f\u7528\u5b98\u65b9 CUDA\/PyTorch \u57fa\u7840\u955c\u50cf\uff0c\u907f\u514d\u91cd\u590d\u914d\u7f6e\u73af\u5883\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code># \u57fa\u7840\u955c\u50cf\uff1a\u5b98\u65b9\u9884\u88c5CUDA+cuDNN+PyTorch\nFROM pytorch\/pytorch:2.1.0-cuda12.1-cudnn8-runtime\n\nWORKDIR \/app\nCOPY app.py .\n\nCMD &#91;\"python\", \"app.py\"]\n<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">\u6784\u5efa\u8fd0\u884c\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>docker build -t my-pytorch-app .\ndocker run --rm --gpus all my-pytorch-app\n<\/code><\/pre>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u56db\u3001Docker \u90e8\u7f72 Ollama\uff08\u2b50\uff09<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">1. \u79c1\u6709\u4ed3\u5e93\u914d\u7f6e\uff08\u5185\u7f51\u73af\u5883\uff09<\/h3>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u5728<code>daemon.json<\/code>\u4e2d\u6dfb\u52a0\u79c1\u6709\u4ed3\u5e93\u5730\u5740\uff0c\u5141\u8bb8 HTTP \u8bbf\u95ee\uff1a<\/li>\n<\/ol>\n\n\n\n<pre class=\"wp-block-code\"><code>{\n  \"insecure-registries\": &#91;\"192.168.100.149\"]\n}\n<\/code><\/pre>\n\n\n\n<ol start=\"2\" class=\"wp-block-list\">\n<li>\u91cd\u542f Docker \u5e76\u767b\u5f55\u4ed3\u5e93\uff1a<\/li>\n<\/ol>\n\n\n\n<pre class=\"wp-block-code\"><code>systemctl daemon-reload &amp;&amp; systemctl restart docker\ndocker login 192.168.100.149 -u 0331admin -p Admin@123\n<\/code><\/pre>\n\n\n\n<h3 class=\"wp-block-heading\">2. \u62c9\u53d6\u955c\u50cf\u5e76\u542f\u52a8\u5bb9\u5668<\/h3>\n\n\n\n<pre class=\"wp-block-code\"><code># \u62c9\u53d6\u955c\u50cf\ndocker pull 192.168.100.149\/gpuimages\/ollama:latest\n\n# \u542f\u52a8\u5bb9\u5668\ndocker run -d \\\n  --name=ollama \\\n  --gpus=all \\\n  -v ollama:\/root\/.ollama \\\n  -p 11434:11434 \\\n  192.168.100.149\/gpuimages\/ollama:latest\n<\/code><\/pre>\n\n\n\n<h3 class=\"wp-block-heading\">3. \u5173\u952e\u53c2\u6570\u8bf4\u660e<\/h3>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>\u53c2\u6570<\/th><th>\u4f5c\u7528<\/th><\/tr><\/thead><tbody><tr><td><code>--gpus=all<\/code><\/td><td>\u6388\u4e88\u5bb9\u5668\u8bbf\u95ee\u6240\u6709 GPU \u6743\u9650<\/td><\/tr><tr><td><code>-v ollama:\/root\/.ollama<\/code><\/td><td>\u6302\u8f7d\u6570\u636e\u5377\uff0c\u6301\u4e45\u5316\u6a21\u578b\u6587\u4ef6\uff0c\u5bb9\u5668\u91cd\u5efa\u4e0d\u4e22\u5931<\/td><\/tr><tr><td><code>-p 11434:11434<\/code><\/td><td>\u7aef\u53e3\u6620\u5c04\uff0c\u5bbf\u4e3b\u673a 11434 \u7aef\u53e3\u6620\u5c04\u5bb9\u5668\u5185\u670d\u52a1\u7aef\u53e3<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">4. \u5bb9\u5668\u5185\u6a21\u578b\u7ba1\u7406<\/h3>\n\n\n\n<pre class=\"wp-block-code\"><code># \u8fdb\u5165\u5bb9\u5668\ndocker exec -it ollama bash\n\n# \u62c9\u53d6\u6a21\u578b\nollama pull qwen:7b\n\n# \u67e5\u770b\u6a21\u578b\u5217\u8868\nollama list\n<\/code><\/pre>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u670d\u52a1\u5730\u5740\uff1a<code>http:\/\/\u5bbf\u4e3b\u673aIP:11434<\/code>\uff0cAPI \u4e0e\u539f\u751f\u90e8\u7f72\u5b8c\u5168\u4e00\u81f4<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u4e94\u3001Docker \u90e8\u7f72 vLLM<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">1. \u955c\u50cf\u62c9\u53d6<\/h3>\n\n\n\n<pre class=\"wp-block-code\"><code>docker pull 192.168.100.149\/gpuimages\/vllm:latest\n<\/code><\/pre>\n\n\n\n<h3 class=\"wp-block-heading\">2. \u542f\u52a8\u5bb9\u5668<\/h3>\n\n\n\n<pre class=\"wp-block-code\"><code>docker run -d \\\n  --name vllm-qwen \\\n  --gpus all \\\n  --network host \\\n  --ipc=host \\\n  --shm-size 16g \\\n  -e VLLM_USE_MODELSCOPE=True \\\n  -e MODELSCOPE_CACHE=\/root\/.cache\/modelscope \\\n  -v \/data\/modelscope:\/root\/.cache\/modelscope \\\n  192.168.100.149\/gpuimages\/vllm:latest \\\n  Qwen\/Qwen2-1.5B-Instruct \\\n  --tensor-parallel-size 1 \\\n  --max-model-len 4096 \\\n  --gpu-memory-utilization 0.9 \\\n  --max-num-seqs 32\n<\/code><\/pre>\n\n\n\n<h3 class=\"wp-block-heading\">3. \u5173\u952e\u914d\u7f6e\u8be6\u89e3<\/h3>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>\u53c2\u6570<\/th><th>\u4f5c\u7528<\/th><th>\u91cd\u8981\u6027<\/th><\/tr><\/thead><tbody><tr><td><code>--network host<\/code><\/td><td>\u4f7f\u7528\u5bbf\u4e3b\u673a\u7f51\u7edc\uff0c\u51cf\u5c11\u7f51\u7edc\u8f6c\u53d1\u635f\u8017\uff0c\u63d0\u5347\u6027\u80fd<\/td><td>\u9ad8<\/td><\/tr><tr><td><code>--ipc=host<\/code><\/td><td>\u4f7f\u7528\u5bbf\u4e3b\u673a IPC \u547d\u540d\u7a7a\u95f4\uff0c\u652f\u6301\u8fdb\u7a0b\u95f4\u9ad8\u901f\u901a\u4fe1<\/td><td>\u9ad8<\/td><\/tr><tr><td><code>--shm-size 16g<\/code><\/td><td>\u8bbe\u7f6e\u5171\u4eab\u5185\u5b58 16GB\uff0cvLLM \u9ad8\u5e76\u53d1\u573a\u666f\u9700\u8981\u5927\u5171\u4eab\u5185\u5b58<\/td><td>\u9ad8<\/td><\/tr><tr><td><code>-e VLLM_USE_MODELSCOPE=True<\/code><\/td><td>\u5f00\u542f ModelScope \u6e90\uff0c\u56fd\u5185\u52a0\u901f\u4e0b\u8f7d\u6a21\u578b<\/td><td>\u4e2d<\/td><\/tr><tr><td><code>-v \/data\/modelscope:...<\/code><\/td><td>\u6302\u8f7d\u6a21\u578b\u7f13\u5b58\u76ee\u5f55\uff0c\u907f\u514d\u91cd\u590d\u4e0b\u8f7d<\/td><td>\u4e2d<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u670d\u52a1\u5730\u5740\uff1a<code>http:\/\/\u5bbf\u4e3b\u673aIP:8000\/v1<\/code>\uff0c\u517c\u5bb9 OpenAI API \u6807\u51c6<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u516d\u3001Docker \u90e8\u7f72 Open WebUI\uff08\u2b50\uff09<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">1. \u542f\u52a8\u5bb9\u5668<\/h3>\n\n\n\n<pre class=\"wp-block-code\"><code>docker run -d \\\n  --name open-webui \\\n  -p 3000:8080 \\\n  --restart always \\\n  -v open-webui:\/app\/backend\/data \\\n  -e ENABLE_RAG_WEB_SEARCH=false \\\n  -e ENABLE_RAG_HYBRID_SEARCH=false \\\n  -e ENABLE_OLLAMA_API=true \\\n  192.168.100.149\/gpuimages\/open-webui:latest\n<\/code><\/pre>\n\n\n\n<h3 class=\"wp-block-heading\">2. \u73af\u5883\u53d8\u91cf\u8bf4\u660e<\/h3>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>\u53c2\u6570<\/th><th>\u4f5c\u7528<\/th><\/tr><\/thead><tbody><tr><td><code>-p 3000:8080<\/code><\/td><td>\u5bbf\u4e3b\u673a 3000 \u7aef\u53e3\u6620\u5c04 WebUI \u7684 8080 \u7aef\u53e3<\/td><\/tr><tr><td><code>-v open-webui:\/app\/backend\/data<\/code><\/td><td>\u6301\u4e45\u5316\u7528\u6237\u6570\u636e\u3001\u5bf9\u8bdd\u8bb0\u5f55\u3001\u914d\u7f6e<\/td><\/tr><tr><td><code>--restart always<\/code><\/td><td>\u5f00\u673a\u81ea\u542f\uff0c\u5f02\u5e38\u81ea\u52a8\u91cd\u542f<\/td><\/tr><tr><td><code>ENABLE_RAG_WEB_SEARCH=false<\/code><\/td><td>\u5173\u95ed\u8054\u7f51\u641c\u7d22\uff0c\u7eaf\u79bb\u7ebf\u8fd0\u884c<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">3. \u540e\u7aef\u5bf9\u63a5<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u5bf9\u63a5 Ollama<\/strong>\uff1a\u754c\u9762\u8bbe\u7f6e \u2192 Ollama \u8fde\u63a5 \u2192 \u586b\u5199<code>http:\/\/\u5bbf\u4e3b\u673aIP:11434<\/code><\/li>\n\n\n\n<li><strong>\u5bf9\u63a5 vLLM<\/strong>\uff1a\u754c\u9762\u8bbe\u7f6e \u2192 \u5916\u90e8\u8fde\u63a5 \u2192 \u6dfb\u52a0 OpenAI \u63a5\u53e3<code>http:\/\/\u5bbf\u4e3b\u673aIP:8000\/v1<\/code><\/li>\n\n\n\n<li>\u8bbf\u95ee\u5730\u5740\uff1a<code>http:\/\/\u5bbf\u4e3b\u673aIP:3000<\/code>\uff0c\u9996\u6b21\u8bbf\u95ee\u6ce8\u518c\u7ba1\u7406\u5458\u8d26\u53f7<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u4e03\u3001\u9ad8\u9891\u6c47\u603b<\/h2>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u4e3a\u4ec0\u4e48 AI \u5927\u6a21\u578b\u670d\u52a1\u8981\u505a\u5bb9\u5668\u5316\uff1f<\/strong> \u7b54\uff1aAI \u6a21\u578b\u4f9d\u8d56\u94fe\u957f\u4e14\u7248\u672c\u8026\u5408\u4e25\u683c\uff0c\u624b\u52a8\u90e8\u7f72\u73af\u5883\u4e00\u81f4\u6027\u5dee\u3001\u6613\u51b2\u7a81\u3001\u8fc1\u79fb\u96be\u3002\u5bb9\u5668\u5316\u5c06\u6574\u5957\u8fd0\u884c\u73af\u5883\u6253\u5305\uff0c\u4fdd\u8bc1\u5f00\u53d1 \/ \u6d4b\u8bd5 \/ \u751f\u4ea7\u73af\u5883\u4e00\u81f4\uff0c\u652f\u6301\u5feb\u901f\u90e8\u7f72\u3001\u5f39\u6027\u6269\u7f29\u3001\u7248\u672c\u56de\u6eda\uff0c\u662f\u751f\u4ea7\u7ea7\u670d\u52a1\u548c K8s \u7f16\u6392\u7684\u57fa\u7840\u3002<\/li>\n\n\n\n<li><strong>NVIDIA Container Toolkit \u7684\u5de5\u4f5c\u539f\u7406\u662f\u4ec0\u4e48\uff1f<\/strong> \u7b54\uff1a\u9a71\u52a8\u5b89\u88c5\u5728\u5bbf\u4e3b\u673a\uff0c\u4e0d\u6253\u5305\u8fdb\u955c\u50cf\uff1b\u5bb9\u5668\u542f\u52a8\u65f6\u901a\u8fc7 OCI Hook \u673a\u5236\uff0c\u52a8\u6001\u5c06\u5bbf\u4e3b\u673a\u7684 GPU \u8bbe\u5907\u8282\u70b9\uff08<code>\/dev\/nvidia*<\/code>\uff09\u3001\u9a71\u52a8\u5e93\u6587\u4ef6\u3001\u73af\u5883\u53d8\u91cf\u6ce8\u5165\u5bb9\u5668\u9694\u79bb\u7a7a\u95f4\uff0c\u8ba9\u5bb9\u5668\u5185\u5e94\u7528\u900f\u660e\u8c03\u7528 GPU\uff0c\u540c\u65f6\u4fdd\u6301\u5bb9\u5668\u7684\u9694\u79bb\u6027\u3002<\/li>\n\n\n\n<li><strong>\u5bb9\u5668\u8bbf\u95ee GPU \u7684\u6838\u5fc3\u53c2\u6570\u662f\u4ec0\u4e48\uff1f\u6709\u54ea\u4e9b\u7528\u6cd5\uff1f<\/strong> \u7b54\uff1a\u6838\u5fc3\u53c2\u6570\u662f<code>--gpus<\/code>\u3002\u652f\u6301<code>all<\/code>\uff08\u5168\u90e8 GPU\uff09\u3001\u6307\u5b9a\u6570\u91cf\u3001\u6307\u5b9a\u8bbe\u5907 ID\u3001\u6307\u5b9a UUID \u7b49\u591a\u79cd\u7c92\u5ea6\u7684 GPU \u5206\u914d\u3002<\/li>\n\n\n\n<li><strong>vLLM \u5bb9\u5668\u90e8\u7f72\u4e3a\u4ec0\u4e48\u8981\u7528 host \u7f51\u7edc\u548c\u5927\u5171\u4eab\u5185\u5b58\uff1f<\/strong> \u7b54\uff1a\n<ul class=\"wp-block-list\">\n<li>host \u7f51\u7edc\u51cf\u5c11 Docker NAT \u8f6c\u53d1\u7684\u6027\u80fd\u635f\u8017\uff0c\u964d\u4f4e\u63a8\u7406\u5ef6\u8fdf\uff1b<\/li>\n\n\n\n<li>vLLM \u7684\u8fde\u7eed\u6279\u5904\u7406\u3001PagedAttention \u673a\u5236\u4f9d\u8d56\u5927\u91cf\u8fdb\u7a0b\u95f4\u901a\u4fe1\uff0c\u5927\u5171\u4eab\u5185\u5b58\uff08<code>--shm-size<\/code>\uff09\u548c<code>--ipc=host<\/code>\u80fd\u4fdd\u969c\u901a\u4fe1\u6548\u7387\uff0c\u907f\u514d\u9ad8\u5e76\u53d1\u4e0b\u51fa\u73b0\u74f6\u9888\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><strong>\u5bb9\u5668\u5316\u90e8\u7f72\u4e2d\uff0c\u6a21\u578b\u6587\u4ef6\u4e3a\u4ec0\u4e48\u63a8\u8350\u7528\u6570\u636e\u5377\u6302\u8f7d\uff1f<\/strong> \u7b54\uff1a\u5bb9\u5668\u672c\u8eab\u662f\u65e0\u72b6\u6001\u7684\uff0c\u5220\u9664\u91cd\u5efa\u540e\u6570\u636e\u4f1a\u4e22\u5931\u3002\u5c06\u6a21\u578b\u3001\u7f13\u5b58\u3001\u7528\u6237\u6570\u636e\u6302\u8f7d\u5230\u5bbf\u4e3b\u673a\u6570\u636e\u5377\uff0c\u53ef\u6301\u4e45\u5316\u5b58\u50a8\uff0c\u907f\u514d\u91cd\u590d\u4e0b\u8f7d\u6a21\u578b\uff0c\u540c\u65f6\u65b9\u4fbf\u5bb9\u5668\u5347\u7ea7\u3001\u8fc1\u79fb\u3002<\/li>\n<\/ol>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u4e00\u3001Docker \u57fa\u7840\u4e0e AI \u5bb9\u5668\u5316\u4ef7\u503c 1. \u6838\u5fc3\u6982\u5ff5 Docker \u662f\u5f00\u6e90\u5bb9\u5668\u5316\u5e73\u53f0\uff0c\u5c06\u5e94\u7528\u53ca\u5176\u4f9d\u8d56\u6253\u5305 [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[6],"tags":[],"class_list":["post-134","post","type-post","status-publish","format-standard","hentry","category-6"],"_links":{"self":[{"href":"https:\/\/www.xiaoxindong.cloud\/index.php?rest_route=\/wp\/v2\/posts\/134","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.xiaoxindong.cloud\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.xiaoxindong.cloud\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.xiaoxindong.cloud\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.xiaoxindong.cloud\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=134"}],"version-history":[{"count":2,"href":"https:\/\/www.xiaoxindong.cloud\/index.php?rest_route=\/wp\/v2\/posts\/134\/revisions"}],"predecessor-version":[{"id":186,"href":"https:\/\/www.xiaoxindong.cloud\/index.php?rest_route=\/wp\/v2\/posts\/134\/revisions\/186"}],"wp:attachment":[{"href":"https:\/\/www.xiaoxindong.cloud\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=134"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.xiaoxindong.cloud\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=134"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.xiaoxindong.cloud\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=134"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}