{"id":837,"date":"2026-07-12T01:15:00","date_gmt":"2026-07-11T17:15:00","guid":{"rendered":"https:\/\/blog.itsmygo.uk\/2026\/07\/12\/%e3%80%90vllm-%e5%b7%a5%e7%a8%8b%e5%b8%ab%e5%af%a6%e6%88%b0%e3%80%9101-%e5%be%9e-ollama-%e5%88%b0-vllm%ef%bc%9a%e7%82%ba%e4%bb%80%e9%ba%bc%e8%a6%81%e6%8f%9b%e3%80%81%e6%80%8e%e9%ba%bc%e8%b7%91\/"},"modified":"2026-07-12T01:15:00","modified_gmt":"2026-07-11T17:15:00","slug":"%e3%80%90vllm-%e5%b7%a5%e7%a8%8b%e5%b8%ab%e5%af%a6%e6%88%b0%e3%80%9101-%e5%be%9e-ollama-%e5%88%b0-vllm%ef%bc%9a%e7%82%ba%e4%bb%80%e9%ba%bc%e8%a6%81%e6%8f%9b%e3%80%81%e6%80%8e%e9%ba%bc%e8%b7%91","status":"publish","type":"post","link":"https:\/\/blog.itsmygo.uk\/wordpress\/2026\/07\/12\/%e3%80%90vllm-%e5%b7%a5%e7%a8%8b%e5%b8%ab%e5%af%a6%e6%88%b0%e3%80%9101-%e5%be%9e-ollama-%e5%88%b0-vllm%ef%bc%9a%e7%82%ba%e4%bb%80%e9%ba%bc%e8%a6%81%e6%8f%9b%e3%80%81%e6%80%8e%e9%ba%bc%e8%b7%91\/","title":{"rendered":"\u3010vLLM \u5de5\u7a0b\u5e2b\u5be6\u6230\u3011#01 \u5f9e Ollama \u5230 vLLM\uff1a\u70ba\u4ec0\u9ebc\u8981\u63db\u3001\u600e\u9ebc\u8dd1\u8d77\u7b2c\u4e00\u500b\u670d\u52d9"},"content":{"rendered":"\n<div id=\"quiz-3db2b512\">\n<style>\n#quiz-3db2b512 .question { background: white; border-radius: 8px; padding: 20px; margin: 20px 0; box-shadow: 0 2px 4px rgba(0,0,0,0.1); }#quiz-3db2b512 .question h3 { margin-top: 0; color: #333; }#quiz-3db2b512 .options { list-style: none; padding: 0; }#quiz-3db2b512 .options li { padding: 12px 15px; margin: 8px 0; border: 2px solid #ddd; border-radius: 6px; cursor: pointer; transition: all 0.2s; }#quiz-3db2b512 .options li:hover:not(.disabled) { border-color: #4caf50; background: #f0fff0; }#quiz-3db2b512 .options li.correct { border-color: #4caf50; background: #e8f5e9; }#quiz-3db2b512 .options li.wrong { border-color: #f44336; background: #ffebee; }#quiz-3db2b512 .options li.disabled { cursor: default; opacity: 0.7; }#quiz-3db2b512 .options li.show-answer { border-color: #4caf50; background: #e8f5e9; }#quiz-3db2b512 .feedback { margin-top: 15px; padding: 12px; border-radius: 6px; display: none; }#quiz-3db2b512 .feedback.correct { background: #e8f5e9; color: #2e7d32; display: block; }#quiz-3db2b512 .feedback.wrong { background: #ffebee; color: #c62828; display: block; }#quiz-3db2b512 .score { text-align: center; font-size: 1.5em; padding: 20px; background: white; border-radius: 8px; margin-top: 20px; }#quiz-3db2b512 .code-block { background: #1e1e1e; color: #d4d4d4; padding: 14px; border-radius: 6px; font-family: \"Monaco\", \"Consolas\", \"Courier New\", monospace; font-size: 0.9em; overflow-x: auto; margin: 12px 0; white-space: pre-wrap; }\n<\/style>\n<h1>\u6e2c\u9a57\uff1a\u5f9e Ollama \u5230 vLLM<\/h1>\n    <p>\u5171 5 \u984c\uff0c\u9ede\u9078\u7b54\u6848\u5f8c\u6703\u7acb\u5373\u986f\u793a\u7d50\u679c<\/p>\n\n    <div class=\"question\" id=\"q1\">\n      <h3>1. \u6839\u64da\u6587\u7ae0\uff0cvLLM \u548c Ollama \u6700\u6838\u5fc3\u7684\u5b9a\u4f4d\u5dee\u7570\u662f\u4ec0\u9ebc\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(1, 'A', 'B')\">A. vLLM \u53ea\u80fd\u8dd1\u91cf\u5316\u6a21\u578b\uff0cOllama \u53ea\u80fd\u8dd1\u5168\u7cbe\u5ea6\u6a21\u578b<\/li>\n        <li onclick=\"checkAnswer(1, 'B', 'B')\">B. Ollama \u662f\u7d66\u300c\u4eba\u300d\u55ae\u6a5f\u8a66\u73a9\u7528\u7684\uff0cvLLM \u662f\u7d66\u300c\u670d\u52d9\u300d\u625b\u9ad8\u4f75\u767c\u7528\u7684<\/li>\n        <li onclick=\"checkAnswer(1, 'C', 'B')\">C. vLLM \u53ea\u80fd\u8dd1\u5728\u96f2\u7aef\uff0cOllama \u53ea\u80fd\u8dd1\u5728\u672c\u6a5f<\/li>\n        <li onclick=\"checkAnswer(1, 'D', 'B')\">D. \u5169\u8005\u529f\u80fd\u5b8c\u5168\u76f8\u540c\uff0c\u53ea\u662f\u4ecb\u9762\u4e0d\u540c<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb1\"><\/div>\n    <\/div>\n\n    <div class=\"question\" id=\"q2\">\n      <h3>2. \u6587\u7ae0\u7528\u300c\u6703\u81ea\u5df1\u505a\u6279\u6b21\u7684\u9ad8\u541e\u5410 LLM \u4f3a\u670d\u5668\u300d\u5f62\u5bb9 vLLM\u3002\u9019\u500b\u300c\u81ea\u5df1\u505a\u6279\u6b21\u300d\u5e36\u4f86\u7684\u4e3b\u8981\u597d\u8655\u662f\u4ec0\u9ebc\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(2, 'A', 'C')\">A. \u8b93\u6a21\u578b\u6a94\u6848\u8b8a\u5c0f\uff0c\u7701\u4e0b\u786c\u789f\u7a7a\u9593<\/li>\n        <li onclick=\"checkAnswer(2, 'B', 'C')\">B. \u8b93\u5b83\u4e0d\u9700\u8981 GPU \u4e5f\u80fd\u8dd1<\/li>\n        <li onclick=\"checkAnswer(2, 'C', 'C')\">C. \u628a\u591a\u500b\u8acb\u6c42\u4f75\u6210\u4e00\u6279\u4e00\u8d77\u7b97\uff0c\u662f\u591a\u4eba\u540c\u6642\u6253\u6642\u7dad\u6301\u9ad8\u541e\u5410\u7684\u95dc\u9375<\/li>\n        <li onclick=\"checkAnswer(2, 'D', 'C')\">D. \u81ea\u52d5\u5e6b\u4f60\u628a\u56de\u7b54\u7ffb\u8b6f\u6210\u591a\u570b\u8a9e\u8a00<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb2\"><\/div>\n    <\/div>\n\n    <div class=\"question\" id=\"q3\">\n      <h3>3. \u4f60\u7528 Python \u7684 openai \u5957\u4ef6\u8981\u6539\u6253\u672c\u6a5f\u7684 vLLM\uff0c\u6700\u95dc\u9375\u8981\u6539\u7684\u662f\u54ea\u4e00\u9805\uff1f<\/h3>\n      <div class=\"code-block\">client = OpenAI(\n    base_url=&#8221;http:\/\/localhost:8000\/v1&#8243;,\n    api_key=&#8221;EMPTY&#8221;,\n)<\/div>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(3, 'A', 'A')\">A. \u628a base_url \u6307\u5230\u672c\u6a5f\u7684 vLLM \u4f4d\u5740\uff0c\u5176\u4ed6\u5beb\u6cd5\u5e7e\u4e4e\u4e0d\u7528\u6539<\/li>\n        <li onclick=\"checkAnswer(3, 'B', 'A')\">B. \u5fc5\u9808\u63db\u6389\u6574\u500b openai \u5957\u4ef6\uff0c\u6539\u7528 vLLM \u5c08\u5c6c SDK<\/li>\n        <li onclick=\"checkAnswer(3, 'C', 'A')\">C. api_key \u4e00\u5b9a\u8981\u586b\u771f\u6b63\u7684 OpenAI \u91d1\u9470\u624d\u80fd\u7528<\/li>\n        <li onclick=\"checkAnswer(3, 'D', 'A')\">D. messages \u683c\u5f0f\u8981\u6539\u6210 vLLM \u5c08\u7528\u683c\u5f0f<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb3\"><\/div>\n    <\/div>\n\n    <div class=\"question\" id=\"q4\">\n      <h3>4. \u95dc\u65bc <code>--gpu-memory-utilization 0.9<\/code>\uff0c\u4e0b\u5217\u6558\u8ff0\u4f55\u8005\u6b63\u78ba\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(4, 'A', 'C')\">A. \u5b83\u6703\u5f37\u5236 vLLM \u525b\u597d\u7528\u6389\u4e5d\u6210\u986f\u5b58\uff0c\u4e00\u9ede\u90fd\u4e0d\u80fd\u591a\u4e0d\u80fd\u5c11<\/li>\n        <li onclick=\"checkAnswer(4, 'B', 'C')\">B. \u5b83\u662f\u8a2d\u5b9a\u56de\u7b54\u9577\u5ea6\uff0c\u8ddf\u986f\u5b58\u7121\u95dc<\/li>\n        <li onclick=\"checkAnswer(4, 'C', 'C')\">C. \u5b83\u8a2d\u5b9a vLLM \u6700\u591a\u53ef\u7528\u7684\u986f\u5b58\u6bd4\u4f8b\uff08\u4e0a\u9650\uff09\uff0c\u986f\u5b58\u7206\u6389\u6642\u53ef\u4ee5\u8abf\u4f4e<\/li>\n        <li onclick=\"checkAnswer(4, 'D', 'C')\">D. \u5b83\u6c7a\u5b9a\u8981\u7528\u5e7e\u5f35 GPU \u4e00\u8d77\u7b97<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb4\"><\/div>\n    <\/div>\n\n    <div class=\"question\" id=\"q5\">\n      <h3>5. \u5718\u968a\u8981\u628a\u4e00\u500b\u958b\u6e90\u6a21\u578b\u67b6\u6210 API \u7d66\u6574\u500b\u5f8c\u7aef\u670d\u52d9\u5171\u7528\uff0c\u4e14\u624b\u4e0a\u6709 GPU\u3002\u4f9d\u6587\u7ae0\u5efa\u8b70\u8a72\u9078\u54ea\u500b\u3001\u70ba\u4ec0\u9ebc\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(5, 'A', 'B')\">A. \u9078 Ollama\uff0c\u56e0\u70ba\u5b83\u5b89\u88dd\u6700\u7c21\u55ae<\/li>\n        <li onclick=\"checkAnswer(5, 'B', 'B')\">B. \u9078 vLLM\uff0c\u56e0\u70ba\u8981\u5c0d\u591a\u4eba\/\u5f8c\u7aef\u63d0\u4f9b\u670d\u52d9\u3001\u4e14\u5728\u610f\u9ad8\u4f75\u767c\u4e0b\u7684\u541e\u5410\u91cf<\/li>\n        <li onclick=\"checkAnswer(5, 'C', 'B')\">C. \u5169\u500b\u90fd\u4e0d\u884c\uff0c\u9019\u7a2e\u9700\u6c42\u53ea\u80fd\u7528 OpenAI \u5b98\u65b9 API<\/li>\n        <li onclick=\"checkAnswer(5, 'D', 'B')\">D. \u9078 Ollama\uff0c\u56e0\u70ba vLLM \u4e0d\u652f\u63f4 OpenAI \u76f8\u5bb9\u683c\u5f0f<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb5\"><\/div>\n    <\/div>\n\n    <div class=\"score\" id=\"score\" style=\"display:none;\"><\/div>\n<script>\n\n      let answered = 0;\n      let correct = 0;\n      const total = 5;\n      const explanations = {\n        1: \"\u6587\u7ae0\u7684\u4e00\u53e5\u8a71\u7e3d\u7d50\uff1aOllama \u662f\u7d66\u300e\u4eba\u300f\u7528\u7684\uff08\u672c\u6a5f\u3001\u55ae\u4eba\u8a66\u73a9\uff09\uff0cvLLM \u662f\u7d66\u300e\u670d\u52d9\u300f\u7528\u7684\uff08\u751f\u7522\u7d1a\u3001\u625b\u9ad8\u4f75\u767c\uff09\u3002\u5206\u91ce\u5728\u65bc\u8981\u4e0d\u8981\u625b\u4f75\u767c\u3002\",\n        2: \"vLLM \u6703\u628a\u591a\u500b\u8acb\u6c42\u81ea\u52d5\u4f75\u6210\u4e00\u6279\u4e00\u8d77\u8a08\u7b97\uff0c\u9019\u6b63\u662f\u5b83\u5728\u591a\u4eba\u540c\u6642\u6253\u6642\u4ecd\u80fd\u7dad\u6301\u9ad8\u541e\u5410\u91cf\u7684\u95dc\u9375\uff0c\u4e5f\u662f\u8207 Ollama\u300e\u4e00\u500b\u4e00\u500b\u8655\u7406\u300f\u6700\u5927\u7684\u5dee\u5225\u3002\",\n        3: \"OpenAI \u76f8\u5bb9\u6700\u5927\u7684\u597d\u8655\u5c31\u662f\uff1a\u53ea\u8981\u628a base_url \u63db\u6210\u672c\u6a5f vLLM\uff0c\u5176\u4ed6\u7a0b\u5f0f\u78bc\u5e7e\u4e4e\u4e0d\u52d5\u3002api_key \u56e0\u70ba vLLM \u9810\u8a2d\u4e0d\u9a57\u8b49\uff0c\u586b\u500b\u975e\u7a7a\u5b57\u4e32\u5373\u53ef\u3002\",\n        4: \"gpu-memory-utilization \u8a2d\u5b9a\u7684\u662f\u300e\u5141\u8a31\u4f7f\u7528\u7684\u986f\u5b58\u6bd4\u4f8b\u4e0a\u9650\u300f\uff080\uff5e1\uff0c\u9810\u8a2d 0.9\uff09\uff0c\u4e0d\u662f\u5f37\u5236\u7528\u6eff\u3002OOM \u6642\u53ef\u8abf\u4f4e\uff08\u4f8b\u5982 0.85\uff09\u7559\u66f4\u591a\u7a7a\u9593\u7d66\u7cfb\u7d71\u3002\",\n        5: \"\u8981\u5c0d\u591a\u4eba\u6216\u5f8c\u7aef\u670d\u52d9\u63d0\u4f9b API\u3001\u53c8\u6709 GPU\u3001\u5728\u610f\u9ad8\u4f75\u767c\u541e\u5410\u91cf\uff0c\u6b63\u662f\u6587\u7ae0\u5b9a\u7fa9\u7684\u300e\u8a72\u63db vLLM\u300f\u60c5\u5883\u3002Ollama \u9069\u5408\u81ea\u5df1\u55ae\u6a5f\u8a66\u73a9\u3002\"\n      };\n      function checkAnswer(qNum, selected, correctAns) {\n        const question = document.getElementById(\"q\" + qNum);\n        if (question.classList.contains(\"answered\")) return;\n        question.classList.add(\"answered\");\n        answered++;\n        const options = question.querySelectorAll(\".options li\");\n        options.forEach((opt) => {\n          opt.classList.add(\"disabled\");\n          if (opt.textContent.startsWith(correctAns + \".\")) { opt.classList.add(\"show-answer\"); }\n        });\n        const feedback = document.getElementById(\"fb\" + qNum);\n        if (selected === correctAns) {\n          correct++;\n          feedback.className = \"feedback correct\";\n          feedback.innerHTML = \"\u2713 \u6b63\u78ba\uff01\" + explanations[qNum];\n        } else {\n          feedback.className = \"feedback wrong\";\n          feedback.innerHTML = \"\u2717 \u932f\u8aa4\u3002\u6b63\u78ba\u7b54\u6848\u662f \" + correctAns + \"\u3002\" + explanations[qNum];\n          options.forEach((opt) => { if (opt.textContent.startsWith(selected + \".\")) { opt.classList.add(\"wrong\"); } });\n        }\n        if (answered === total) {\n          const scoreDiv = document.getElementById(\"score\");\n          scoreDiv.style.display = \"block\";\n          scoreDiv.innerHTML = \"\u6e2c\u9a57\u5b8c\u6210\uff01\u5f97\u5206\uff1a\" + correct + \" \/ \" + total + (correct === total ? \" \ud83c\udf89 \u5168\u5c0d\uff01\" : correct >= Math.ceil(total * 0.6) ? \" \ud83d\udc4d \u4e0d\u932f\uff01\" : \" \ud83d\udcda \u518d\u8907\u7fd2\u4e00\u4e0b\u5427\");\n        }\n      }\n    \n<\/script>\n<\/div>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\">\u7cfb\u5217\u7b2c 1 \u7bc7\uff08\u5171 4 \u7bc7\uff09\uff5c\u96e3\u5ea6\uff1aL2-\u9032\u968e<br><br>\u524d\u7f6e\u77e5\u8b58\uff1a\u57fa\u672c Linux\/\u7d42\u7aef\u6a5f\u64cd\u4f5c\u3001Python \u74b0\u5883\u3001\u77e5\u9053\u4ec0\u9ebc\u662f LLM \u63a8\u8ad6\u8207 OpenAI API \u683c\u5f0f\u3002<\/p>\n<\/blockquote>\n\n\n\n<h2 class=\"wp-block-heading\">\u4e00\u53e5\u8a71\u8aaa\u660e<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>vLLM \u662f\u4e00\u53f0\u300c\u6703\u81ea\u5df1\u505a\u6279\u6b21\u300d\u7684\u9ad8\u541e\u5410 LLM \u4f3a\u670d\u5668<\/strong>\u2014\u2014\u4f60\u7528\u5b83\u628a\u4e00\u500b\u958b\u6e90\u6a21\u578b\u67b6\u6210\u4e00\u500b OpenAI \u76f8\u5bb9\u7684 API\uff0c\u8b93\u5f88\u591a\u4eba\u540c\u6642\u6253\u4e5f\u4e0d\u6703\u5361\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5982\u679c Ollama \u662f\u300c\u5728\u4f60\u7b46\u96fb\u4e0a\u8dd1\u500b\u6a21\u578b\u73a9\u73a9\u300d\uff0c\u90a3 vLLM \u5c31\u662f\u300c\u628a\u6a21\u578b\u67b6\u6210\u6b63\u5f0f\u670d\u52d9\uff0c\u625b\u5f97\u4f4f\u4e00\u7968\u4eba\u540c\u6642\u7528\u300d\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u4e00\u7bc7\u4e0d\u8ac7\u5e95\u5c64\u7406\u8ad6\uff08PagedAttention\u3001continuous batching \u90a3\u4e9b\u7559\u5230\u5f8c\u9762\uff09\uff0c\u53ea\u5e36\u4f60\uff1a\u641e\u6e05\u695a\u4ec0\u9ebc\u6642\u5019\u8a72\u5f9e Ollama \u63db\u5230 vLLM\u3001\u628a\u7b2c\u4e00\u500b\u670d\u52d9\u8dd1\u8d77\u4f86\u3001\u770b\u61c2\u555f\u52d5\u6307\u4ee4\u6bcf\u500b\u53c3\u6578\u5728\u5e79\u561b\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u5148\u641e\u6e05\u695a\uff1avLLM \u548c Ollama \u5230\u5e95\u5dee\u5728\u54ea<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u5f88\u591a\u4eba\u7b2c\u4e00\u500b\u554f\u984c\u662f\u300c\u6211\u5df2\u7d93\u6709 Ollama \u4e86\uff0c\u70ba\u4ec0\u9ebc\u8981\u5b78 vLLM\uff1f\u300d<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e00\u53e5\u8a71\uff1a<strong>Ollama \u662f\u7d66\u300c\u4eba\u300d\u7528\u7684\uff0cvLLM \u662f\u7d66\u300c\u670d\u52d9\u300d\u7528\u7684\u3002<\/strong><\/p>\n\n\n\n<figure class=\"wp-block-table\"><table>\n<thead>\n<tr>\n<th>\u9762\u5411<\/th>\n<th>Ollama<\/th>\n<th>vLLM<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>\u5b9a\u4f4d<\/td>\n<td>\u672c\u6a5f\u958b\u767c\u3001\u55ae\u4eba\u8a66\u73a9<\/td>\n<td>\u751f\u7522\u7d1a\u63a8\u8ad6\u670d\u52d9<\/td>\n<\/tr>\n<tr>\n<td>\u4f75\u767c\u80fd\u529b<\/td>\n<td>\u4e00\u6b21\u670d\u52d9\u4e00\u5169\u500b\u8acb\u6c42\u5c31\u5403\u529b<\/td>\n<td>\u5e7e\u5341\u4e0a\u767e\u4f75\u767c\u9084\u80fd\u7dad\u6301\u9ad8\u541e\u5410<\/td>\n<\/tr>\n<tr>\n<td>\u786c\u9ad4\u53cb\u5584\u5ea6<\/td>\n<td>CPU \u4e5f\u80fd\u8dd1\u3001\u5403 GGUF \u91cf\u5316\u6a94<\/td>\n<td>\u5e7e\u4e4e\u4e00\u5b9a\u8981 GPU\uff08\u5403\u986f\u5b58\uff09<\/td>\n<\/tr>\n<tr>\n<td>\u5b89\u88dd<\/td>\n<td>\u4e00\u500b installer\uff0c\u958b\u7bb1\u5373\u7528<\/td>\n<td><code>pip install<\/code>\uff0c\u8981\u5c0d\u597d CUDA \u74b0\u5883<\/td>\n<\/tr>\n<tr>\n<td>\u6279\u6b21\u8655\u7406<\/td>\n<td>\u57fa\u672c\u4e0a\u4e00\u500b\u4e00\u500b\u8655\u7406<\/td>\n<td>\u81ea\u52d5\u628a\u591a\u500b\u8acb\u6c42\u4f75\u6210\u4e00\u6279\u4e00\u8d77\u7b97\uff08\u541e\u5410\u91cf\u7ffb\u500d\u7684\u95dc\u9375\uff09<\/td>\n<\/tr>\n<tr>\n<td>\u9069\u5408\u5834\u666f<\/td>\n<td>\u6211\u4e00\u500b\u4eba\u5728 side project \u554f\u554f\u984c<\/td>\n<td>\u4e00\u500b API \u8981\u7d66\u6574\u500b\u5718\u968a \/ \u7522\u54c1\u5f8c\u7aef\u7528<\/td>\n<\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">\u4ec0\u9ebc\u6642\u5019\u8a72\u63db\uff1f<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n\n<li>\ud83d\udfe2 <strong>\u7e7c\u7e8c\u7528 Ollama<\/strong>\uff1a\u53ea\u6709\u4f60\u81ea\u5df1\u7528\u3001\u8dd1\u5728\u7b46\u96fb\/\u6c92\u6709\u50cf\u6a23 GPU\u3001\u53ea\u662f\u60f3\u8a66\u6a21\u578b\u6548\u679c\u3002<\/li>\n\n\n<li>\ud83d\udd34 <strong>\u8a72\u63db vLLM<\/strong>\uff1a\u8981\u67b6\u6210 API \u7d66\u591a\u4eba\u6216\u5f8c\u7aef\u670d\u52d9\u7528\u3001\u6709\u4e00\u5f35\uff08\u6216\u591a\u5f35\uff09GPU\u3001\u5728\u610f\u300c\u540c\u6642\u5f88\u591a\u4eba\u6253\u7684\u6642\u5019\u9084\u5feb\u4e0d\u5feb\u300d\u3002<\/li>\n\n<\/ul>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\">\u767d\u8a71\u6bd4\u55bb\uff1aOllama \u50cf\u5bb6\u88e1\u7684\u74e6\u65af\u7210\uff0c\u4e00\u6b21\u716e\u4e00\u5169\u4eba\u4efd\u5f88\u65b9\u4fbf\uff1bvLLM \u50cf\u9910\u5ef3\u7684\u7210\u7076\uff0c\u4e00\u6b21\u51fa\u5e7e\u5341\u4efd\u4e5f\u7a69\u3002\u4f60\u81ea\u5df1\u5403\u4e0d\u9700\u8981\u9910\u5ef3\u7210\u7076\uff0c\u4f46\u8981\u958b\u5e97\u5c31\u5f97\u63db\u3002<\/p>\n<\/blockquote>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u74b0\u5883\u9700\u6c42\uff1a\u5148\u78ba\u8a8d\u4f60\u6709\u6c92\u6709\u672c\u9322\u8dd1<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">vLLM \u5e7e\u4e4e\u662f\u70ba GPU \u800c\u751f\u7684\u3002\u52d5\u624b\u524d\u5148\u78ba\u8a8d\u4e09\u4ef6\u4e8b\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n\n<li><strong>\u6709 NVIDIA GPU \u4e14\u88dd\u597d\u9a45\u52d5<\/strong>\u2014\u2014\u8dd1 <code>nvidia-smi<\/code>\uff0c\u770b\u5f97\u5230\u4f60\u7684\u986f\u5361\u548c CUDA \u7248\u672c\u624d\u7b97\u904e\u95dc\u3002<\/li>\n\n\n<li><strong>\u986f\u5b58\u5920\u4e0d\u5920<\/strong>\u2014\u2014\u9019\u662f\u6700\u5bb9\u6613\u7ffb\u8eca\u7684\u5730\u65b9\u3002\u6a21\u578b\u53c3\u6578\u91cf \u00d7 \u6bcf\u500b\u53c3\u6578\u7684\u4f4d\u5143\u7d44\u6578\uff0c\u5927\u6982\u5c31\u662f\u6700\u4f4e\u986f\u5b58\u9700\u6c42\u3002\u4e00\u500b\u7c97\u7565\u7684\u6293\u6cd5\uff1a<\/li>\n\n<\/ol>\n\n\n\n<p class=\"wp-block-paragraph\">> 7B \u6a21\u578b\u7528 FP16\uff08\u534a\u7cbe\u5ea6\uff09\u8dd1\uff0c\u5149\u6a21\u578b\u6b0a\u91cd\u5c31\u7d04 14GB\uff0c\u518d\u52a0\u4e0a KV cache \u548c\u5176\u4ed6\u958b\u92b7\uff0c\u5be6\u52d9\u4e0a\u8981\u6293 16GB \u4ee5\u4e0a\u986f\u5b58\u624d\u6bd4\u8f03\u5b89\u5168\u3002<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n\n<li><strong>Python \u74b0\u5883<\/strong>\u2014\u2014\u5efa\u8b70\u7528\u865b\u64ec\u74b0\u5883\uff08venv \u6216 uv\uff09\uff0c\u5225\u76f4\u63a5\u88dd\u9032\u7cfb\u7d71 Python\u3002<\/li>\n\n<\/ol>\n\n\n\n<h3 class=\"wp-block-heading\">\u5b89\u88dd\u8207\u5192\u7159\u6e2c\u8a66<\/h3>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-1\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\"><span class=\"hljs-comment\"># \u5efa\u500b\u4e7e\u6de8\u7684\u865b\u64ec\u74b0\u5883\uff08\u9019\u88e1\u7528 uv\uff0c\u63db\u6210 venv \u4e5f\u884c\uff09<\/span>\nuv venv &amp;&amp; source .venv\/bin\/activate\n\n<span class=\"hljs-comment\"># \u88dd vLLM\uff08\u5b83\u6703\u4e00\u8d77\u5e36\u9032\u5c0d\u61c9\u7684 PyTorch \/ CUDA \u76f8\u4f9d\uff09<\/span>\npip install vllm\n\n<span class=\"hljs-comment\"># \u5192\u7159\u6e2c\u8a66\uff1a\u78ba\u8a8d\u88dd\u8d77\u4f86\u3001import \u5f97\u5230\u3001\u7248\u672c\u5370\u5f97\u51fa\u4f86<\/span>\npython -c <span class=\"hljs-string\">\"import vllm; print(vllm.__version__)\"<\/span>\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-1\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">PHP<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">php<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\">\u2705 **\u9019\u5728\u5e79\u561b**\uff1a\u6700\u5f8c\u90a3\u884c <code>python -c \"...\"<\/code> \u662f\u300c\u4e00\u53e5\u8a71 Python\u300d\u2014\u2014\u4e0d\u9032\u4e92\u52d5\u6a21\u5f0f\uff0c\u76f4\u63a5\u8dd1\u5f15\u865f\u88e1\u7684\u7a0b\u5f0f\u3002\u80fd\u5370\u51fa\u7248\u672c\u865f\uff0c\u4ee3\u8868 vLLM \u81f3\u5c11\u88dd\u5c0d\u4e86\u3001\u8f09\u5f97\u9032\u4f86\u3002\u5370\u4e0d\u51fa\u4f86\uff08\u5831\u932f\uff09\u5c31\u5225\u5f80\u4e0b\u8d70\uff0c\u5148\u628a\u74b0\u5883\u4fee\u597d\u3002<\/p>\n<\/blockquote>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u8dd1\u8d77\u7b2c\u4e00\u500b\u670d\u52d9\uff1a<code>vllm serve<\/code><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u88dd\u597d\u4e4b\u5f8c\uff0c\u4e00\u884c\u6307\u4ee4\u5c31\u80fd\u628a\u6a21\u578b\u67b6\u6210\u4e00\u500b OpenAI \u76f8\u5bb9\u7684 API\uff1a<\/p>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\"><span><code class=\"hljs\">vllm serve Qwen\/Qwen2.5-7B-Instruct\n<\/code><\/span><\/pre><\/div>\n\n\n<h3 class=\"wp-block-heading\">\u9010\u6bb5\u7ffb\u8b6f<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n\n<li><code>vllm serve<\/code>\uff1a\u555f\u52d5 vLLM \u7684 API \u4f3a\u670d\u5668\u6a21\u5f0f\u3002<\/li>\n\n\n<li><code>Qwen\/Qwen2.5-7B-Instruct<\/code>\uff1a\u8981\u8f09\u5165\u7684\u6a21\u578b\u3002\u9019\u662f Hugging Face \u4e0a\u7684\u6a21\u578b ID\uff08<code>\u7d44\u7e54\u540d\/\u6a21\u578b\u540d<\/code>\uff09\uff0cvLLM \u6703\u81ea\u52d5\u5e6b\u4f60\u4e0b\u8f09\u3002<\/li>\n\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u8dd1\u8d77\u4f86\u5f8c\uff0c\u9810\u8a2d\u6703\u5728 <code>http:\/\/localhost:8000<\/code> \u958b\u4e00\u500b\u670d\u52d9\uff0c\u800c\u4e14<strong>\u9577\u5f97\u8ddf OpenAI \u7684 API \u4e00\u6a21\u4e00\u6a23<\/strong>\u3002\u9019\u9ede\u5f88\u91cd\u8981\u2014\u2014\u610f\u601d\u662f\u4f60\u539f\u672c\u6253 OpenAI \u7684\u7a0b\u5f0f\u78bc\uff0c\u5e7e\u4e4e\u53ea\u8981\u6539\u500b\u7db2\u5740\u5c31\u80fd\u6539\u6253\u4f60\u81ea\u5df1\u7684 vLLM\u3002<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\">\u23f3 \u7b2c\u4e00\u6b21\u8dd1\u6703\u5361\u5f88\u4e45\u5f88\u6b63\u5e38\u2014\u2014\u5b83\u5728\u4e0b\u8f09\u5e7e GB \u5230\u5e7e\u5341 GB \u7684\u6a21\u578b\u6b0a\u91cd\u3002\u770b\u5230\u985e\u4f3c <code>Uvicorn running on http:\/\/0.0.0.0:8000<\/code> \u9019\u884c\uff0c\u624d\u4ee3\u8868\u670d\u52d9\u771f\u7684\u8d77\u4f86\u4e86\u3002<\/p>\n<\/blockquote>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u6253\u901a\u5b83\uff1acurl \u8207 Python<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">\u7528 curl \u6253\uff08\u6700\u5feb\u78ba\u8a8d\u670d\u52d9\u6d3b\u8457\uff09<\/h3>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-2\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">curl http:<span class=\"hljs-comment\">\/\/localhost:8000\/v1\/chat\/completions \\<\/span>\n  -H <span class=\"hljs-string\">\"Content-Type: application\/json\"<\/span> \\\n  -d <span class=\"hljs-string\">'{\n    \"model\": \"Qwen\/Qwen2.5-7B-Instruct\",\n    \"messages\": &#91;{\"role\": \"user\", \"content\": \"\u4f60\u597d\uff0c\u4e00\u53e5\u8a71\u81ea\u6211\u4ecb\u7d39\"}]\n  }'<\/span>\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-2\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">PHP<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">php<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u9010\u884c\u7ffb\u8b6f<\/strong>\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n\n<li><code>\/v1\/chat\/completions<\/code>\uff1a\u9019\u5c31\u662f OpenAI \u7684\u804a\u5929 API \u8def\u5f91\uff0cvLLM \u539f\u5c01\u4e0d\u52d5\u7167\u6284\u3002<\/li>\n\n\n<li><code>-H \"Content-Type: application\/json\"<\/code>\uff1a\u544a\u8a34\u4f3a\u670d\u5668\u300c\u6211\u9001\u7684\u662f JSON\u300d\u3002<\/li>\n\n\n<li><code>-d '{...}'<\/code>\uff1a\u8981\u9001\u51fa\u53bb\u7684\u5167\u5bb9\u3002<code>model<\/code> \u8981\u8ddf\u4f60 serve \u6642\u7684\u6a21\u578b ID \u4e00\u81f4\uff0c<code>messages<\/code> \u662f\u5c0d\u8a71\u5167\u5bb9\u3002<\/li>\n\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">\u7528 Python \u7684 OpenAI SDK \u6253\uff08\u5be6\u52d9\u4e0a\u90fd\u9019\u6a23\u63a5\uff09<\/h3>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-3\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">from openai import OpenAI\n\n<span class=\"hljs-comment\"># \u91cd\u9ede\uff1abase_url \u6307\u5230\u4f60\u81ea\u5df1\u7684 vLLM\uff0c\u4e0d\u662f OpenAI<\/span>\nclient = OpenAI(\n    base_url=<span class=\"hljs-string\">\"http:\/\/localhost:8000\/v1\"<\/span>,\n    api_key=<span class=\"hljs-string\">\"EMPTY\"<\/span>,   <span class=\"hljs-comment\"># vLLM \u9810\u8a2d\u4e0d\u9a57\u8b49\u91d1\u9470\uff0c\u96a8\u4fbf\u586b\u4e00\u500b\u975e\u7a7a\u5b57\u4e32\u5373\u53ef<\/span>\n)\n\nresp = client.chat.completions.create(\n    model=<span class=\"hljs-string\">\"Qwen\/Qwen2.5-7B-Instruct\"<\/span>,\n    messages=&#91;{<span class=\"hljs-string\">\"role\"<\/span>: <span class=\"hljs-string\">\"user\"<\/span>, <span class=\"hljs-string\">\"content\"<\/span>: <span class=\"hljs-string\">\"\u4f60\u597d\uff0c\u4e00\u53e5\u8a71\u81ea\u6211\u4ecb\u7d39\"<\/span>}],\n)\n\n<span class=\"hljs-keyword\">print<\/span>(resp.choices&#91;<span class=\"hljs-number\">0<\/span>].message.content)\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-3\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">PHP<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">php<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u9019\u6bb5\u5728\u5e79\u561b<\/strong>\uff1a\u7528\u5b98\u65b9 <code>openai<\/code> \u5957\u4ef6\uff0c\u4f46\u628a <code>base_url<\/code> \u63db\u6210\u672c\u6a5f\u7684 vLLM\u3002\u9664\u4e86\u9019\u4e00\u884c\uff0c\u5176\u4ed6\u5beb\u6cd5\u8ddf\u4f60\u6253\u771f\u6b63 OpenAI \u5b8c\u5168\u4e00\u6a23\u2014\u2014\u9019\u5c31\u662f\u300cOpenAI \u76f8\u5bb9\u300d\u6700\u5be6\u969b\u7684\u597d\u8655\uff1a<strong>\u4f60\u7684\u61c9\u7528\u7a0b\u5f0f\u78bc\u5e7e\u4e4e\u4e0d\u7528\u6539\u3002<\/strong><\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\">\ud83d\udccc **\u77e5\u9053\u5c31\u597d**\uff1a<code>api_key=\"EMPTY\"<\/code> \u662f\u56e0\u70ba vLLM \u9810\u8a2d\u6c92\u958b\u9a57\u8b49\u3002\u9019\u4ee3\u8868**\u4efb\u4f55\u80fd\u9023\u5230\u9019\u500b\u4f4d\u5740\u7684\u4eba\u90fd\u80fd\u7528\u4f60\u7684\u6a21\u578b**\u2014\u2014\u4e0a\u7dda\u524d\u52d9\u5fc5\u52a0\u4e0a <code>--api-key<\/code> \u6216\u7528\u53cd\u5411\u4ee3\u7406\u64cb\u4e00\u5c64\uff08\u5f8c\u9762\u6703\u63d0\uff09\u3002<\/p>\n<\/blockquote>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u770b\u61c2\u555f\u52d5\u53c3\u6578\uff1aAI \u5e6b\u4f60\u751f\u6307\u4ee4\u6642\uff0c\u4f60\u8981\u770b\u5f97\u61c2<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u5be6\u52d9\u4e0a <code>vllm serve<\/code> \u5f8c\u9762\u5e38\u5e38\u639b\u4e00\u4e32\u53c3\u6578\u3002AI \u5e6b\u4f60\u751f\u51fa\u4f86\u7684\u6307\u4ee4\uff0c\u4f60\u81f3\u5c11\u8981\u770b\u61c2\u9019\u56db\u500b\u6700\u5e38\u898b\u7684\uff1a<\/p>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\"><span><code class=\"hljs\">vllm serve Qwen\/Qwen2.5-7B-Instruct \\\n  --gpu-memory-utilization 0.9 \\\n  --max-model-len 8192 \\\n  --tensor-parallel-size 2\n<\/code><\/span><\/pre><\/div>\n\n\n<figure class=\"wp-block-table\"><table>\n<thead>\n<tr>\n<th>\u53c3\u6578<\/th>\n<th>\u767d\u8a71\u89e3\u91cb<\/th>\n<th>\u4ec0\u9ebc\u6642\u5019\u8981\u52d5\u5b83<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><code>--model<\/code>\uff08\u6216\u76f4\u63a5\u5beb\u5728\u5f8c\u9762\uff09<\/td>\n<td>\u8981\u8f09\u5165\u54ea\u500b\u6a21\u578b\uff08HF ID \u6216\u672c\u6a5f\u8def\u5f91\uff09<\/td>\n<td>\u4e00\u5b9a\u8981<\/td>\n<\/tr>\n<tr>\n<td><code>--gpu-memory-utilization<\/code><\/td>\n<td>\u5141\u8a31 vLLM \u5403\u6389\u591a\u5c11\u6bd4\u4f8b\u7684\u986f\u5b58\uff080\uff5e1\uff09\uff0c\u9810\u8a2d 0.9<\/td>\n<td>\u986f\u5b58\u7206\u4e86\u5c31\u8abf\u4f4e\uff08\u5982 0.8\uff09\uff1b\u540c\u6a5f\u9084\u8981\u8dd1\u5225\u7684\u6771\u897f\u4e5f\u8abf\u4f4e<\/td>\n<\/tr>\n<tr>\n<td><code>--max-model-len<\/code><\/td>\n<td>\u55ae\u6b21\u5c0d\u8a71\u6700\u9577\u80fd\u5403\u591a\u5c11 token\uff08prompt + \u56de\u7b54\u52a0\u8d77\u4f86\uff09<\/td>\n<td>\u986f\u5b58\u4e0d\u5920\u6642\u8abf\u5c0f\uff1b\u9700\u8981\u9577\u5c0d\u8a71\u6642\u8abf\u5927\uff08\u4f46\u66f4\u5403\u986f\u5b58\uff09<\/td>\n<\/tr>\n<tr>\n<td><code>--tensor-parallel-size<\/code><\/td>\n<td>\u628a\u6a21\u578b\u62c6\u5230\u5e7e\u5f35 GPU \u4e0a\u4e00\u8d77\u7b97<\/td>\n<td>\u4e00\u5f35\u5361\u88dd\u4e0d\u4e0b\u5927\u6a21\u578b\u6642\uff0c\u8a2d\u6210\u4f60\u6709\u7684 GPU \u5f35\u6578<\/td>\n<\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">\u9010\u500b\u767d\u8a71<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n\n<li><strong><code>--gpu-memory-utilization 0.9<\/code><\/strong>\uff1a\u8ddf vLLM \u8aaa\u300c\u9019\u5f35\u5361\u7684\u986f\u5b58\uff0c\u4f60\u6700\u591a\u7528\u4e5d\u6210\u300d\u3002\u7559\u4e00\u6210\u7d66\u7cfb\u7d71\u548c\u5176\u4ed6\u7a0b\u5f0f\u5598\u606f\u3002<strong>\u4e0d\u662f<\/strong>\u8a2d 0.9 \u5c31\u4e00\u5b9a\u7528\u6eff\uff0c\u800c\u662f\u8a2d\u4e00\u500b\u4e0a\u9650\u3002<\/li>\n\n\n<li><strong><code>--max-model-len 8192<\/code><\/strong>\uff1a\u4e00\u6b21\u5c0d\u8a71\uff08\u4f60\u554f\u7684 + \u5b83\u7b54\u7684\uff09\u52a0\u8d77\u4f86\u6700\u591a 8192 \u500b token\u3002\u8a2d\u592a\u5927\u5f88\u5403\u986f\u5b58\uff0c\u8a2d\u592a\u5c0f\u9577\u6587\u6703\u88ab\u622a\u3002<\/li>\n\n\n<li><strong><code>--tensor-parallel-size 2<\/code><\/strong>\uff1a\u4e00\u5f35 GPU \u88dd\u4e0d\u4e0b 70B \u9019\u7a2e\u5927\u6a21\u578b\u6642\uff0c\u7528\u5169\u5f35\u5361\u628a\u6a21\u578b\u300c\u5207\u958b\u300d\u4e00\u8d77\u625b\u3002<strong>\u9019\u500b\u6578\u5b57\u4e0d\u80fd\u8d85\u904e\u4f60\u5be6\u969b\u6709\u7684 GPU \u6578\u91cf<\/strong>\uff0c\u4e5f\u901a\u5e38\u8981\u80fd\u6574\u9664\u6a21\u578b\u7684\u6ce8\u610f\u529b\u982d\u6578\uff08\u8a2d\u932f\u6703\u76f4\u63a5\u555f\u52d5\u5931\u6557\uff09\u3002<\/li>\n\n<\/ul>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\">\u2705 **\u5fc5\u770b\u61c2**\uff1a<code>--gpu-memory-utilization<\/code> \u548c <code>--max-model-len<\/code> \u9019\u5169\u500b\u662f\u4f60\u6700\u5e38\u88ab\u8feb\u8abf\u6574\u7684\uff0c\u56e0\u70ba\u5b83\u5011\u76f4\u63a5\u95dc\u4fc2\u5230\u300c\u8dd1\u4e0d\u8dd1\u5f97\u8d77\u4f86\u300d\u3002<br><br>\ud83d\udccc **\u77e5\u9053\u5c31\u597d**\uff1a<code>--tensor-parallel-size<\/code> \u53ea\u6709\u4f60\u73a9\u591a\u5361 \/ \u5927\u6a21\u578b\u6642\u624d\u7528\u5f97\u5230\uff0c\u55ae\u5361\u8dd1 7B \u7528\u4e0d\u5230\u3002<\/p>\n<\/blockquote>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\ud83d\udea9 \u7b2c\u4e00\u6b21\u8dd1\u4e0d\u8d77\u4f86\u7684\u7d05\u65d7<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">AI \u751f\u7684\u555f\u52d5\u6307\u4ee4\u300c\u770b\u8d77\u4f86\u6c38\u9060\u5f88\u5c0d\u300d\uff0c\u4f46 vLLM \u5c0d\u786c\u9ad4\u5f88\u8aa0\u5be6\u2014\u2014\u4e0d\u5c0d\u5c31\u662f\u8d77\u4e0d\u4f86\u3002\u4ee5\u4e0b\u662f\u6700\u5e38\u649e\u7684\u5e7e\u500b\uff0c\u6559\u4f60\u8a8d\u5f97\u75c7\u72c0\u3001\u77e5\u9053\u4e0b\u4e00\u6b65\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\ud83d\udd34 \u7d05\u65d7 1\uff1aOOM\uff08\u986f\u5b58\u7206\u4e86\uff09<\/h3>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-4\" data-shcb-language-name=\"CSS\" data-shcb-language-slug=\"css\"><span><code class=\"hljs language-css\"><span class=\"hljs-selector-tag\">torch<\/span><span class=\"hljs-selector-class\">.OutOfMemoryError<\/span>: <span class=\"hljs-selector-tag\">CUDA<\/span> <span class=\"hljs-selector-tag\">out<\/span> <span class=\"hljs-selector-tag\">of<\/span> <span class=\"hljs-selector-tag\">memory<\/span>.\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-4\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">CSS<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">css<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u70ba\u4ec0\u9ebc<\/strong>\uff1a\u6a21\u578b + KV cache \u8981\u7684\u986f\u5b58\u8d85\u904e\u4f60\u7684\u5361\u3002\u9019\u662f vLLM \u65b0\u624b\u7b2c\u4e00\u540d\u6b7b\u56e0\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u600e\u9ebc\u6551<\/strong>\uff08\u7531\u6eab\u548c\u5230\u6fc0\u9032\uff09\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n\n<li>\u8abf\u4f4e <code>--gpu-memory-utilization<\/code>\uff08\u5982 0.85\uff09\u2014\u2014\u6709\u6642\u53ea\u662f\u7559\u7d66\u7cfb\u7d71\u7684\u592a\u5c11\u3002<\/li>\n\n\n<li>\u8abf\u5c0f <code>--max-model-len<\/code>\uff08\u5982\u5f9e 8192 \u964d\u5230 4096\uff09\u2014\u2014KV cache \u76f4\u63a5\u7701\u4e00\u534a\u3002<\/li>\n\n\n<li>\u63db\u66f4\u5c0f\u7684\u6a21\u578b\uff0c\u6216\u7528\u91cf\u5316\u7248\u672c\uff08\u5f8c\u9762\u7bc7\u7ae0\u6703\u8ac7\uff09\u3002<\/li>\n\n<\/ol>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u8ddf AI \u9019\u6a23\u8aaa<\/strong>\uff1a\u300c\u6211\u7684 GPU \u53ea\u6709 16GB \u986f\u5b58\uff0c\u8dd1 7B \u6a21\u578b OOM \u4e86\uff0c\u5e6b\u6211\u8abf\u6574 <code>--max-model-len<\/code> \u548c <code>--gpu-memory-utilization<\/code>\uff0c\u4e26\u89e3\u91cb\u70ba\u4ec0\u9ebc\u9019\u6a23\u6539\u3002\u300d<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\ud83d\udd34 \u7d05\u65d7 2\uff1amax-model-len \u8d85\u904e\u6a21\u578b\u4e0a\u9650<\/h3>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\"><span><code class=\"hljs\">ValueError: User-specified max_model_len (32768) is greater than\nthe derived max_model_len (8192) ...\n<\/code><\/span><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u70ba\u4ec0\u9ebc<\/strong>\uff1a\u4f60\u8a2d\u7684 <code>--max-model-len<\/code> \u8d85\u904e\u9019\u500b\u6a21\u578b\u672c\u8eab\u652f\u63f4\u7684\u4e0a\u4e0b\u6587\u9577\u5ea6\u3002\u4e0d\u662f\u986f\u5b58\u554f\u984c\uff0c\u662f<strong>\u8a2d\u5b9a\u503c\u672c\u8eab\u4e0d\u5408\u6cd5<\/strong>\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u600e\u9ebc\u6551<\/strong>\uff1a\u628a <code>--max-model-len<\/code> \u8a2d\u56de\u6a21\u578b\u652f\u63f4\u7684\u7bc4\u570d\u5167\uff08\u932f\u8aa4\u8a0a\u606f\u88e1\u7684 <code>derived max_model_len<\/code> \u5c31\u662f\u4e0a\u9650\uff09\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u8ddf AI \u9019\u6a23\u8aaa<\/strong>\uff1a\u300c\u9019\u500b\u6a21\u578b\u7684\u6700\u5927\u4e0a\u4e0b\u6587\u9577\u5ea6\u662f\u591a\u5c11\uff1f\u6211\u7684 <code>--max-model-len<\/code> \u8a2d\u591a\u5c11\u624d\u5408\u6cd5\uff1f\u300d<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\ud83d\udfe1 \u7d05\u65d7 3\uff1aPort \u88ab\u5360\u7528<\/h3>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-5\" data-shcb-language-name=\"CSS\" data-shcb-language-slug=\"css\"><span><code class=\"hljs language-css\"><span class=\"hljs-selector-attr\">&#91;Errno 98]<\/span> <span class=\"hljs-selector-tag\">Address<\/span> <span class=\"hljs-selector-tag\">already<\/span> <span class=\"hljs-selector-tag\">in<\/span> <span class=\"hljs-selector-tag\">use<\/span>\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-5\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">CSS<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">css<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u70ba\u4ec0\u9ebc<\/strong>\uff1a8000 \u57e0\u5df2\u7d93\u6709\u6771\u897f\u5728\u8dd1\uff08\u53ef\u80fd\u662f\u4e0a\u4e00\u6b21\u6c92\u95dc\u4e7e\u6de8\u7684 vLLM\uff0c\u6216\u5225\u7684\u670d\u52d9\uff09\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u600e\u9ebc\u6551<\/strong>\uff1a\u63db\u500b\u57e0 <code>--port 8001<\/code>\uff0c\u6216\u5148\u628a\u5360\u7528 8000 \u7684\u7a0b\u5e8f\u95dc\u6389\uff08<code>lsof -i:8000<\/code> \u627e\u51fa\u4f86\uff09\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u8ddf AI \u9019\u6a23\u8aaa<\/strong>\uff1a\u300c8000 \u57e0\u88ab\u5360\u7528\u4e86\uff0c\u5e6b\u6211\u6539\u7528\u5225\u7684\u57e0\u555f\u52d5\uff0c\u4e26\u63d0\u9192\u6211\u6253 API \u6642\u7db2\u5740\u8981\u8ddf\u8457\u6539\u3002\u300d<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\ud83d\udfe1 \u7d05\u65d7 4\uff1a<code>api_key=\"EMPTY\"<\/code> \u76f4\u63a5\u4e0a\u7dda<\/h3>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-6\" data-shcb-language-name=\"JavaScript\" data-shcb-language-slug=\"javascript\"><span><code class=\"hljs language-javascript\">client = OpenAI(base_url=<span class=\"hljs-string\">\"http:\/\/0.0.0.0:8000\/v1\"<\/span>, api_key=<span class=\"hljs-string\">\"EMPTY\"<\/span>)\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-6\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">JavaScript<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">javascript<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\">\u770b\u5230\u670d\u52d9\u7d81\u5728 <code>0.0.0.0<\/code>\uff08\u5c0d\u5916\u5168\u958b\uff09\u53c8\u6c92\u8a2d\u91d1\u9470\uff0c\u4ee3\u8868<strong>\u4efb\u4f55\u9023\u5f97\u5230\u7684\u4eba\u90fd\u80fd\u514d\u8cbb\u7528\u4f60\u7684 GPU<\/strong>\u3002\u672c\u6a5f\u6e2c\u8a66\u7121\u6240\u8b02\uff0c\u4f46\u8981\u5c0d\u5916\u5c31\u662f\u7d05\u65d7\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u8ddf AI \u9019\u6a23\u8aaa<\/strong>\uff1a\u300c\u9019\u500b\u670d\u52d9\u8981\u5c0d\u5916\u958b\u653e\uff0c\u5e6b\u6211\u52a0\u4e0a <code>--api-key<\/code> \u9a57\u8b49\uff0c\u4e26\u8aaa\u660e\u9084\u6709\u54ea\u4e9b\u5b89\u5168\u8a2d\u5b9a\u8a72\u88dc\uff08\u4f8b\u5982\u9650\u5236\u7d81\u5b9a\u4f4d\u5740\u3001\u52a0\u53cd\u5411\u4ee3\u7406\uff09\u3002\u300d<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">Vibe Coder \u9a57\u6536\u6aa2\u67e5\u9ede<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u8ddf\u8457\u505a\uff0c\u505a\u5f97\u5230\u5c31\u4ee3\u8868\u4f60\u9019\u7bc7\u901a\u95dc\u4e86\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n\n<li><strong>\u78ba\u8a8d\u88dd\u597d<\/strong>\uff1a\u8dd1<\/li>\n\n\n<li><strong>\u78ba\u8a8d\u670d\u52d9\u8d77\u4f86<\/strong>\uff1a\u8dd1 <code>vllm serve &lt;\u4f60\u7684\u6a21\u578b&gt;<\/code> \u5f8c\uff0c\u7d42\u7aef\u6a5f\u51fa\u73fe <code>Uvicorn running on http:\/\/0.0.0.0:8000<\/code> \u9019\u985e\u8a0a\u606f\u3002<\/li>\n\n\n<li><strong>\u78ba\u8a8d\u6253\u5f97\u901a<\/strong>\uff1a\u53e6\u958b\u4e00\u500b\u7d42\u7aef\u6a5f\u8dd1\u4e0a\u9762\u7684 <code>curl<\/code> \u6307\u4ee4\uff0c\u61c9\u8a72\u6536\u5230\u4e00\u6bb5 JSON\uff0c\u88e1\u9762 <code>choices[0].message.content<\/code> \u6709\u6a21\u578b\u7684\u56de\u7b54\u3002<\/li>\n\n\n<li><strong>\u78ba\u8a8d\u81ea\u5df1\u770b\u5f97\u61c2\u53c3\u6578<\/strong>\uff1a\u628a\u4f60\u7684\u555f\u52d5\u6307\u4ee4\u8cbc\u7d66 AI\uff0c\u554f\u300c\u9019\u5e7e\u500b\u53c3\u6578\u5404\u81ea\u5728\u63a7\u5236\u4ec0\u9ebc\uff1f\u5982\u679c\u6211 OOM \u4e86\u8a72\u5148\u52d5\u54ea\u4e00\u500b\uff1f\u300d\u2014\u2014\u4f60\u8981\u80fd\u5224\u65b7\u5b83\u7684\u56de\u7b54\u5c0d\u4e0d\u5c0d\uff0c\u800c\u4e0d\u662f\u7167\u55ae\u5168\u6536\u3002<\/li>\n\n<\/ol>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u770b\u4e0d\u61c2\u5c31\u9019\u6a23\u554f AI<\/h2>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\">\u300c\u7528\u767d\u8a71\u89e3\u91cb <code>vllm serve<\/code> \u9019\u884c\u6307\u4ee4\u7684\u6bcf\u500b\u53c3\u6578\u5728\u5e79\u561b\uff0c\u6211\u662f\u7b2c\u4e00\u6b21\u7528 vLLM\uff0c\u53ea\u6709\u4e00\u5f35 16GB \u7684 GPU\u3002\u300d<\/p>\n<\/blockquote>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\">\u300c\u6211\u8dd1 vLLM \u51fa\u73fe\u9019\u500b\u932f\u8aa4\uff08\u8cbc\u4e0a\u932f\u8aa4\u8a0a\u606f\uff09\uff0c\u9019\u662f\u986f\u5b58\u4e0d\u5920\u3001\u53c3\u6578\u8a2d\u932f\u3001\u9084\u662f\u57e0\u88ab\u5360\u7528\uff1f\u4e0b\u4e00\u6b65\u8a72\u600e\u9ebc\u505a\uff1f\u300d<\/p>\n<\/blockquote>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\">\u300c\u6211\u539f\u672c\u9019\u6bb5\u7a0b\u5f0f\u78bc\u662f\u6253 OpenAI \u7684\uff08\u8cbc\u4e0a\u7a0b\u5f0f\u78bc\uff09\uff0c\u8981\u6539\u6210\u6253\u6211\u672c\u6a5f\u7684 vLLM \u670d\u52d9\uff0c\u9700\u8981\u6539\u54ea\u5e7e\u884c\uff1f\u300d<\/p>\n<\/blockquote>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u5c0f\u7d50<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n\n<li>vLLM \u548c Ollama \u4e0d\u662f\u8ab0\u53d6\u4ee3\u8ab0\uff1a<strong>\u81ea\u5df1\u73a9\u7528 Ollama\uff0c\u67b6\u670d\u52d9\u7528 vLLM<\/strong>\u3002\u5206\u91ce\u5728\u300c\u8981\u4e0d\u8981\u625b\u4f75\u767c\u300d\u3002<\/li>\n\n\n<li>vLLM \u5e7e\u4e4e\u7d81 GPU\uff0c\u8dd1\u4e4b\u524d\u5148\u78ba\u8a8d\u986f\u5b58\u5920\u4e0d\u5920\u2014\u2014\u9019\u662f\u6700\u5bb9\u6613\u7ffb\u8eca\u7684\u9ede\u3002<\/li>\n\n\n<li><code>vllm serve &lt;\u6a21\u578b&gt;<\/code> \u4e00\u884c\u5c31\u80fd\u958b\u4e00\u500b <strong>OpenAI \u76f8\u5bb9<\/strong> \u7684 API\uff0c\u4f60\u539f\u672c\u7684\u7a0b\u5f0f\u6539\u500b <code>base_url<\/code> \u5c31\u80fd\u63a5\u3002<\/li>\n\n\n<li>\u56db\u500b\u5fc5\u8a8d\u53c3\u6578\uff1a<code>--model<\/code>\u3001<code>--gpu-memory-utilization<\/code>\u3001<code>--max-model-len<\/code>\u3001<code>--tensor-parallel-size<\/code>\u3002\u524d\u5169\u500b\u95dc\u4fc2\u5230\u300c\u8dd1\u4e0d\u8dd1\u5f97\u8d77\u4f86\u300d\uff0c\u6700\u5e38\u88ab\u8feb\u8abf\u3002<\/li>\n\n\n<li>\u8dd1\u4e0d\u8d77\u4f86\u6642\u7684\u56db\u5927\u7d05\u65d7\uff1a<strong>OOM\u3001max-model-len \u8d85\u4e0a\u9650\u3001\u57e0\u88ab\u5360\u3001\u91d1\u9470\u5168\u958b<\/strong>\u3002\u8a8d\u5f97\u75c7\u72c0\uff0c\u5c31\u77e5\u9053\u8a72\u554f AI \u4ec0\u9ebc\u3002<\/li>\n\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e0b\u4e00\u7bc7\uff0c\u6211\u5011\u6703\u9032\u5230\u300c\u600e\u9ebc\u8b93\u5b83\u8dd1\u5f97\u66f4\u5feb\u3001\u66f4\u7701\u300d\u2014\u2014\u8abf\u53c3\u6578\u3001\u770b\u541e\u5410\u91cf\u6307\u6a19\uff0c\u628a\u9019\u53f0\u4f3a\u670d\u5668\u69a8\u51fa\u50f9\u503c\u3002<\/p>\n\n\n\n<div id=\"quiz-38cfed2b\">\n<style>\n#quiz-38cfed2b .badge { display: inline-block; padding: 4px 10px; border-radius: 12px; font-size: 0.75em; font-weight: bold; margin-left: 8px; }#quiz-38cfed2b .badge.scenario { background: #ebf8ff; color: #2b6cb0; }#quiz-38cfed2b .badge.debug { background: #fef3c7; color: #92400e; }#quiz-38cfed2b .question-adv { background: white; border-radius: 10px; padding: 24px; margin: 20px 0; box-shadow: 0 2px 8px rgba(0,0,0,0.08); border-left: 4px solid #ed8936; }#quiz-38cfed2b .question-adv h3 { margin-top: 0; color: #2d3748; }#quiz-38cfed2b .code-block { background: #1e1e1e; color: #d4d4d4; padding: 14px; border-radius: 6px; font-family: \"Menlo\", \"Monaco\", \"Courier New\", monospace; font-size: 0.9em; overflow-x: auto; margin: 12px 0; white-space: pre-wrap; }#quiz-38cfed2b .options-adv { list-style: none; padding: 0; }#quiz-38cfed2b .options-adv li { padding: 14px 16px; margin: 10px 0; border: 2px solid #e2e8f0; border-radius: 8px; cursor: pointer; transition: all 0.2s; }#quiz-38cfed2b .options-adv li:hover:not(.disabled) { border-color: #ed8936; background: #fffaf0; }#quiz-38cfed2b .options-adv li.disabled { cursor: default; opacity: 0.7; }#quiz-38cfed2b .options-adv li.show-answer { border-color: #48bb78; background: #f0fff4; }#quiz-38cfed2b .options-adv li.wrong { border-color: #fc8181; background: #fff5f5; }#quiz-38cfed2b .feedback-adv { margin-top: 16px; padding: 14px; border-radius: 8px; display: none; line-height: 1.6; }#quiz-38cfed2b .feedback-adv.correct { background: #f0fff4; color: #276749; display: block; }#quiz-38cfed2b .feedback-adv.wrong { background: #fff5f5; color: #c53030; display: block; }#quiz-38cfed2b .feedback-adv \n      .score-adv { text-align: center; font-size: 1.4em; padding: 24px; background: white; border-radius: 10px; margin-top: 24px; box-shadow: 0 2px 8px rgba(0,0,0,0.08); }#quiz-38cfed2b .intro { background: #ebf8ff; padding: 16px; border-radius: 8px; margin-bottom: 20px; color: #2c5282; }\n<\/style>\n<h1>\u9032\u968e\u6e2c\u9a57\uff1a\u5f9e Ollama \u5230 vLLM<\/h1>\n    <div class=\"intro\">\n      <strong>\u6e2c\u9a57\u76ee\u6a19\uff1a<\/strong>\u9a57\u8b49\u4f60\u662f\u5426\u80fd\u5728\u5be6\u969b\u60c5\u5883\u4e2d\u61c9\u7528\u6240\u5b78\u3002<br \/>\n      \u5171 5 \u984c\uff0c\u5305\u542b\u60c5\u5883\u984c\u8207\u932f\u8aa4\u8a3a\u65b7\u984c\u3002\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv1\">\n      <h3>1. \u60c5\u5883\u5224\u65b7 <span class=\"badge scenario\">\u60c5\u5883\u984c<\/span><\/h3>\n      <div class=\"code-block\">\u4f60\u6709\u4e00\u5f35 16GB \u986f\u5b58\u7684 GPU\uff0c\u60f3\u8dd1\u4e00\u500b 7B \u6a21\u578b\uff08FP16\uff09\u3002\n\u555f\u52d5\u5f8c\u7acb\u523b\u51fa\u73fe\uff1a\ntorch.OutOfMemoryError: CUDA out of memory.\n\u4f60\u5e0c\u671b\u7528\u300c\u6700\u6eab\u548c\u3001\u5148\u4e0d\u63db\u6a21\u578b\u300d\u7684\u65b9\u5f0f\u8b93\u5b83\u5148\u8dd1\u8d77\u4f86\u3002<\/div>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(1, 'A', 'B')\">A. \u76f4\u63a5\u628a &#8211;tensor-parallel-size \u8a2d\u6210 4<\/li>\n        <li onclick=\"checkAnswerAdv(1, 'B', 'B')\">B. \u8abf\u4f4e &#8211;gpu-memory-utilization \u4e26\u8abf\u5c0f &#8211;max-model-len\uff0c\u5148\u7701\u4e0b KV cache<\/li>\n        <li onclick=\"checkAnswerAdv(1, 'C', 'B')\">C. \u628a &#8211;max-model-len \u8abf\u5230 32768 \u8b93\u5b83\u6709\u66f4\u591a\u7a7a\u9593<\/li>\n        <li onclick=\"checkAnswerAdv(1, 'D', 'B')\">D. \u79fb\u9664 api_key \u53c3\u6578<\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv1\"><\/div>\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv2\">\n      <h3>2. \u670d\u52d9\u5c0d\u5916\u958b\u653e <span class=\"badge scenario\">\u60c5\u5883\u984c<\/span><\/h3>\n      <div class=\"code-block\">\u4f60\u8981\u628a vLLM \u670d\u52d9\u958b\u653e\u7d66\u516c\u53f8\u5167\u7db2\u5176\u4ed6\u540c\u4e8b\u7684\u5f8c\u7aef\u547c\u53eb\uff0c\n\u76ee\u524d\u555f\u52d5\u6307\u4ee4\u6c92\u6709\u4efb\u4f55\u9a57\u8b49\uff0cclient \u4e5f\u9084\u5beb\u8457 api_key=&#8221;EMPTY&#8221;\u3002\n\u4f60\u6700\u8a72\u512a\u5148\u88dc\u4e0a\u7684\u662f\u4ec0\u9ebc\uff1f<\/div>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(2, 'A', 'C')\">A. \u4ec0\u9ebc\u90fd\u4e0d\u7528\u505a\uff0cvLLM \u672c\u4f86\u5c31\u5b89\u5168<\/li>\n        <li onclick=\"checkAnswerAdv(2, 'B', 'C')\">B. \u628a &#8211;max-model-len \u8abf\u66f4\u5927\u4ee5\u652f\u63f4\u66f4\u591a\u4eba<\/li>\n        <li onclick=\"checkAnswerAdv(2, 'C', 'C')\">C. \u52a0\u4e0a &#8211;api-key \u9a57\u8b49\uff08\u6216\u7528\u53cd\u5411\u4ee3\u7406\u64cb\u4e00\u5c64\uff09\uff0c\u907f\u514d\u4efb\u4f55\u9023\u5f97\u5230\u7684\u4eba\u90fd\u80fd\u7528\u4f60\u7684 GPU<\/li>\n        <li onclick=\"checkAnswerAdv(2, 'D', 'C')\">D. \u628a base_url \u6539\u6210 OpenAI \u5b98\u65b9\u4f4d\u5740<\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv2\"><\/div>\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv3\">\n      <h3>3. \u9078\u578b\u6c7a\u7b56 <span class=\"badge scenario\">\u60c5\u5883\u984c<\/span><\/h3>\n      <div class=\"code-block\">\u4f60\u5728\u81ea\u5df1\u7684\u7b46\u96fb\uff08\u6c92\u6709\u7368\u7acb GPU\uff09\u4e0a\uff0c\u60f3\u5feb\u901f\u8a66\u5e7e\u500b\u958b\u6e90\u6a21\u578b\u7684\u56de\u7b54\u54c1\u8cea\uff0c\n\u53ea\u6709\u4f60\u4e00\u500b\u4eba\u7528\uff0c\u4e0d\u9700\u8981\u5c0d\u5916\u63d0\u4f9b\u670d\u52d9\u3002<\/div>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(3, 'A', 'A')\">A. \u7528 Ollama\uff1aCPU\/GGUF \u53cb\u5584\u3001\u55ae\u6a5f\u958b\u7bb1\u5373\u7528\uff0c\u6b63\u597d\u7b26\u5408\u55ae\u4eba\u8a66\u73a9\u60c5\u5883<\/li>\n        <li onclick=\"checkAnswerAdv(3, 'B', 'A')\">B. \u7528 vLLM\uff1a\u5373\u4f7f\u6c92 GPU \u4e5f\u662f\u6700\u4f73\u9078\u64c7<\/li>\n        <li onclick=\"checkAnswerAdv(3, 'C', 'A')\">C. \u4e00\u5b9a\u8981\u5148\u79df\u4e00\u5f35\u96f2\u7aef GPU \u8dd1 vLLM<\/li>\n        <li onclick=\"checkAnswerAdv(3, 'D', 'A')\">D. \u5169\u8005\u90fd\u4e0d\u9069\u5408\uff0c\u53ea\u80fd\u7528\u7dda\u4e0a API<\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv3\"><\/div>\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv4\">\n      <h3>4. \u932f\u8aa4\u8a3a\u65b7 <span class=\"badge debug\">\u932f\u8aa4\u8a3a\u65b7<\/span><\/h3>\n      <div class=\"code-block\">$ vllm serve Qwen\/Qwen2.5-7B-Instruct &#8211;max-model-len 32768\n&#8230;\nValueError: User-specified max_model_len (32768) is greater than\nthe derived max_model_len (8192) &#8230;<\/div>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(4, 'A', 'C')\">A. \u986f\u5b58\u4e0d\u5920\uff0c\u8981\u8abf\u4f4e &#8211;gpu-memory-utilization<\/li>\n        <li onclick=\"checkAnswerAdv(4, 'B', 'C')\">B. 8000 \u57e0\u88ab\u5360\u7528\u4e86\uff0c\u63db\u500b\u57e0\u5373\u53ef<\/li>\n        <li onclick=\"checkAnswerAdv(4, 'C', 'C')\">C. &#8211;max-model-len \u8d85\u904e\u9019\u500b\u6a21\u578b\u672c\u8eab\u652f\u63f4\u7684\u4e0a\u4e0b\u6587\u4e0a\u9650\uff0c\u8a2d\u5b9a\u503c\u4e0d\u5408\u6cd5\uff0c\u8981\u8abf\u56de 8192 \u4ee5\u5167<\/li>\n        <li onclick=\"checkAnswerAdv(4, 'D', 'C')\">D. \u6a21\u578b\u540d\u7a31\u6253\u932f\uff0cHugging Face \u4e0a\u627e\u4e0d\u5230<\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv4\"><\/div>\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv5\">\n      <h3>5. \u932f\u8aa4\u8a3a\u65b7 <span class=\"badge debug\">\u932f\u8aa4\u8a3a\u65b7<\/span><\/h3>\n      <div class=\"code-block\">$ vllm serve Qwen\/Qwen2.5-7B-Instruct\n&#8230;\n[Errno 98] Address already in use<\/div>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(5, 'A', 'D')\">A. \u6a21\u578b\u4e0b\u8f09\u5931\u6557\uff0c\u8981\u91cd\u65b0 pip install vllm<\/li>\n        <li onclick=\"checkAnswerAdv(5, 'B', 'D')\">B. \u986f\u5b58\u7206\u4e86\uff0c\u8981\u8abf\u5c0f &#8211;max-model-len<\/li>\n        <li onclick=\"checkAnswerAdv(5, 'C', 'D')\">C. CUDA \u9a45\u52d5\u6c92\u88dd\u597d\uff0cnvidia-smi \u6703\u5931\u6557<\/li>\n        <li onclick=\"checkAnswerAdv(5, 'D', 'D')\">D. 8000 \u57e0\u5df2\u88ab\u5360\u7528\uff1b\u6539\u7528 &#8211;port 8001 \u6216\u5148\u95dc\u6389\u5360\u7528\u8a72\u57e0\u7684\u7a0b\u5e8f<\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv5\"><\/div>\n    <\/div>\n\n    <div class=\"score-adv\" id=\"scoreAdv\" style=\"display:none;\"><\/div>\n<script>\n\n      let answeredAdv = 0;\n      let correctAdv = 0;\n      const totalAdv = 5;\n      const explanationsAdv = {\n        1: \"OOM \u7684\u6eab\u548c\u6551\u6cd5\u9806\u5e8f\uff1a\u5148\u8abf\u4f4e --gpu-memory-utilization\uff08\u7559\u7a7a\u9593\u7d66\u7cfb\u7d71\uff09\uff0c\u518d\u8abf\u5c0f --max-model-len\uff08\u76f4\u63a5\u7701\u4e0b KV cache\uff09\uff0c\u6700\u5f8c\u624d\u8003\u616e\u63db\u66f4\u5c0f\u6216\u91cf\u5316\u6a21\u578b\u3002\u628a max-model-len \u8abf\u66f4\u5927\uff08C\uff09\u53cd\u800c\u66f4\u5403\u986f\u5b58\uff1btensor-parallel-size \u8a2d 4\uff08A\uff09\u5728\u53ea\u6709\u4e00\u5f35\u5361\u6642\u6703\u76f4\u63a5\u5931\u6557\u3002\",\n        2: \"\u670d\u52d9\u7d81 0.0.0.0 \u53c8\u6c92\u91d1\u9470\u6642\uff0c\u4efb\u4f55\u9023\u5f97\u5230\u7684\u4eba\u90fd\u80fd\u514d\u8cbb\u7528\u4f60\u7684 GPU\u3002\u5c0d\u5916\u958b\u653e\u524d\u52d9\u5fc5\u52a0 --api-key \u9a57\u8b49\u6216\u7528\u53cd\u5411\u4ee3\u7406\u64cb\u4e00\u5c64\uff0c\u9019\u662f\u6587\u7ae0\u660e\u5217\u7684\u5b89\u5168\u7d05\u65d7\u3002\",\n        3: \"\u6c92\u6709\u7368\u7acb GPU\u3001\u55ae\u4eba\u8a66\u73a9\u3001\u4e0d\u5c0d\u5916\u670d\u52d9\u2014\u2014\u6b63\u662f Ollama \u7684\u4e3b\u5834\uff1aCPU\/GGUF \u53cb\u5584\u3001\u958b\u7bb1\u5373\u7528\u3002vLLM \u5e7e\u4e4e\u7d81 GPU\uff0c\u9019\u60c5\u5883\u4e0d\u5212\u7b97\u3002\",\n        4: \"\u932f\u8aa4\u8a0a\u606f\u5df2\u7d93\u5beb\u660e\uff1a\u4f60\u8a2d\u7684 max_model_len \u5927\u65bc\u6a21\u578b derived \u51fa\u4f86\u7684\u4e0a\u9650\u3002\u9019\u4e0d\u662f\u986f\u5b58\u6216\u57e0\u7684\u554f\u984c\uff0c\u800c\u662f\u8a2d\u5b9a\u503c\u672c\u8eab\u4e0d\u5408\u6cd5\uff0c\u8981\u628a --max-model-len \u8abf\u56de\u6a21\u578b\u652f\u63f4\u7bc4\u570d\uff08\u8a0a\u606f\u88e1\u7684 8192\uff09\u4ee5\u5167\u3002\",\n        5: \"\u300eAddress already in use\u300f\u662f\u5178\u578b\u7684\u57e0\u88ab\u5360\u7528\uff0c\u5e38\u898b\u65bc\u4e0a\u4e00\u6b21 vLLM \u6c92\u95dc\u4e7e\u6de8\u3002\u7528 --port 8001 \u63db\u57e0\uff0c\u6216\u7528 lsof -i:8000 \u627e\u51fa\u4e26\u95dc\u6389\u5360\u7528\u7684\u7a0b\u5e8f\u3002\u8a18\u5f97\u63db\u57e0\u5f8c\u6253 API \u7684\u7db2\u5740\u4e5f\u8981\u8ddf\u8457\u6539\u3002\"\n      };\n      function checkAnswerAdv(qNum, selected, correctAns) {\n        const question = document.getElementById(\"qAdv\" + qNum);\n        if (question.classList.contains(\"answered\")) return;\n        question.classList.add(\"answered\");\n        answeredAdv++;\n        const options = question.querySelectorAll(\".options-adv li\");\n        options.forEach((opt) => {\n          opt.classList.add(\"disabled\");\n          if (opt.textContent.startsWith(correctAns + \".\")) { opt.classList.add(\"show-answer\"); }\n        });\n        const feedback = document.getElementById(\"fbAdv\" + qNum);\n        if (selected === correctAns) {\n          correctAdv++;\n          feedback.className = \"feedback-adv correct\";\n          feedback.innerHTML = \"\u2713 \u6b63\u78ba\uff01\" + explanationsAdv[qNum];\n        } else {\n          feedback.className = \"feedback-adv wrong\";\n          feedback.innerHTML = \"\u2717 \u932f\u8aa4\u3002\u6b63\u78ba\u7b54\u6848\u662f \" + correctAns + \"\u3002\" + explanationsAdv[qNum];\n          options.forEach((opt) => { if (opt.textContent.startsWith(selected + \".\")) { opt.classList.add(\"wrong\"); } });\n        }\n        if (answeredAdv === totalAdv) {\n          const scoreDiv = document.getElementById(\"scoreAdv\");\n          scoreDiv.style.display = \"block\";\n          let msg = \"\";\n          if (correctAdv === totalAdv) msg = \"\ud83c\udfaf \u5b8c\u7f8e\uff01\u4f60\u5df2\u7d93\u6e96\u5099\u597d\u5be6\u6230\u4e86\uff01\";\n          else if (correctAdv >= Math.ceil(totalAdv * 0.8)) msg = \"\ud83d\udcaa \u5f88\u68d2\uff01\u518d\u8907\u7fd2\u4e00\u4e0b\u932f\u7684\u984c\u76ee\";\n          else if (correctAdv >= Math.ceil(totalAdv * 0.6)) msg = \"\ud83d\udcd6 \u5efa\u8b70\u91cd\u8b80\u76f8\u95dc\u6bb5\u843d\u5f8c\u518d\u8a66\u4e00\u6b21\";\n          else msg = \"\ud83d\udd04 \u5efa\u8b70\u91cd\u65b0\u95b1\u8b80\u6587\u4ef6\uff0c\u7279\u5225\u6ce8\u610f\u64cd\u4f5c\u6d41\u7a0b\";\n          scoreDiv.innerHTML = \"\u6e2c\u9a57\u5b8c\u6210\uff01\u5f97\u5206\uff1a\" + correctAdv + \" \/ \" + totalAdv + \"<br>\" + msg;\n        }\n      }\n    \n<\/script>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>\u6e2c\u9a57\uff1a\u5f9e Ollama \u5230 vLLM \u5171 5 \u984c\uff0c\u9ede\u9078\u7b54\u6848\u5f8c\u6703\u7acb\u5373\u986f\u793a\u7d50\u679c 1. \u6839\u64da\u6587\u7ae0\uff0cvLLM \u548c Ol [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"rolpb_likes":0,"footnotes":""},"categories":[858],"tags":[862,860,859,861,847],"class_list":["post-837","post","type-post","status-publish","format-standard","hentry","category-vllm-","tag-gpu","tag-llm","tag-ollama","tag-openaiapi","tag-vllm"],"blocksy_meta":{"styles_descriptor":{"styles":{"desktop":"","tablet":"","mobile":""},"google_fonts":[],"version":7}},"_links":{"self":[{"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/posts\/837","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/comments?post=837"}],"version-history":[{"count":0,"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/posts\/837\/revisions"}],"wp:attachment":[{"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/media?parent=837"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/categories?post=837"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/tags?post=837"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}