{"id":840,"date":"2026-07-12T01:15:28","date_gmt":"2026-07-11T17:15:28","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%9104-%e7%b5%90%e5%90%88-lmcache%ef%bc%9a%e6%8a%8a-kv-cache-%e5%bf%ab%e5%8f%96%e8%b5%b7%e4%be%86%ef%bc%8c%e9%95%b7%e5%b0%8d%e8%a9%b1\/"},"modified":"2026-07-12T01:15:28","modified_gmt":"2026-07-11T17:15:28","slug":"%e3%80%90vllm-%e5%b7%a5%e7%a8%8b%e5%b8%ab%e5%af%a6%e6%88%b0%e3%80%9104-%e7%b5%90%e5%90%88-lmcache%ef%bc%9a%e6%8a%8a-kv-cache-%e5%bf%ab%e5%8f%96%e8%b5%b7%e4%be%86%ef%bc%8c%e9%95%b7%e5%b0%8d%e8%a9%b1","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%9104-%e7%b5%90%e5%90%88-lmcache%ef%bc%9a%e6%8a%8a-kv-cache-%e5%bf%ab%e5%8f%96%e8%b5%b7%e4%be%86%ef%bc%8c%e9%95%b7%e5%b0%8d%e8%a9%b1\/","title":{"rendered":"\u3010vLLM \u5de5\u7a0b\u5e2b\u5be6\u6230\u3011#04 \u7d50\u5408 LMCache\uff1a\u628a KV Cache \u5feb\u53d6\u8d77\u4f86\uff0c\u9577\u5c0d\u8a71\u8207 RAG \u518d\u52a0\u901f"},"content":{"rendered":"\n<div id=\"quiz-ef733362\">\n<style>\n#quiz-ef733362 .question { background: white; border-radius: 8px; padding: 20px; margin: 20px 0; box-shadow: 0 2px 4px rgba(0,0,0,0.1); }#quiz-ef733362 .question h3 { margin-top: 0; color: #333; }#quiz-ef733362 .options { list-style: none; padding: 0; }#quiz-ef733362 .options li { padding: 12px 15px; margin: 8px 0; border: 2px solid #ddd; border-radius: 6px; cursor: pointer; transition: all 0.2s; }#quiz-ef733362 .options li:hover:not(.disabled) { border-color: #4caf50; background: #f0fff0; }#quiz-ef733362 .options li.correct { border-color: #4caf50; background: #e8f5e9; }#quiz-ef733362 .options li.wrong { border-color: #f44336; background: #ffebee; }#quiz-ef733362 .options li.disabled { cursor: default; opacity: 0.7; }#quiz-ef733362 .options li.show-answer { border-color: #4caf50; background: #e8f5e9; }#quiz-ef733362 .feedback { margin-top: 15px; padding: 12px; border-radius: 6px; display: none; }#quiz-ef733362 .feedback.correct { background: #e8f5e9; color: #2e7d32; display: block; }#quiz-ef733362 .feedback.wrong { background: #ffebee; color: #c62828; display: block; }#quiz-ef733362 .score { text-align: center; font-size: 1.5em; padding: 20px; background: white; border-radius: 8px; margin-top: 20px; }#quiz-ef733362 .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\u7d50\u5408 LMCache \u52a0\u901f\u9577\u5c0d\u8a71\u8207 RAG<\/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. LMCache \u4e3b\u8981\u89e3\u6c7a\u7684\u554f\u984c\u662f\u4ec0\u9ebc\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(1, 'A', 'B')\">A. \u8b93\u6a21\u578b\u751f\u6210\u7684 token \u54c1\u8cea\u66f4\u9ad8<\/li>\n        <li onclick=\"checkAnswer(1, 'B', 'B')\">B. \u8b93\u76f8\u540c\u524d\u7db4\u7684 KV Cache \u80fd\u8de8\u8acb\u6c42\u91cd\u7528\uff0c\u4e0d\u7528\u6bcf\u6b21\u91cd\u7b97<\/li>\n        <li onclick=\"checkAnswer(1, 'C', 'B')\">C. \u58d3\u7e2e\u6a21\u578b\u6b0a\u91cd\u8b93\u986f\u5b58\u9700\u6c42\u8b8a\u5c0f<\/li>\n        <li onclick=\"checkAnswer(1, 'D', 'B')\">D. \u81ea\u52d5\u5e6b\u4f60\u8abf\u6574 &#8211;max-model-len \u53c3\u6578<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb1\"><\/div>\n    <\/div>\n\n    <div class=\"question\" id=\"q2\">\n      <h3>2. \u5728 vLLM \u4e2d\u958b\u555f\u5167\u5efa\u7684 automatic prefix caching\uff0c\u8981\u52a0\u54ea\u500b flag\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(2, 'A', 'C')\">A. &#8211;gpu-memory-utilization<\/li>\n        <li onclick=\"checkAnswer(2, 'B', 'C')\">B. &#8211;enable-lmcache<\/li>\n        <li onclick=\"checkAnswer(2, 'C', 'C')\">C. &#8211;enable-prefix-caching<\/li>\n        <li onclick=\"checkAnswer(2, 'D', 'C')\">D. &#8211;kv-cache-offload<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb2\"><\/div>\n    <\/div>\n\n    <div class=\"question\" id=\"q3\">\n      <h3>3. \u76f8\u8f03\u65bc vLLM \u5167\u5efa\u7684 prefix caching\uff0cLMCache \u591a\u63d0\u4f9b\u4e86\u4ec0\u9ebc\u80fd\u529b\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(3, 'A', 'A')\">A. \u628a KV Cache \u5378\u8f09\u5230 CPU\/\u78c1\u789f\/\u9060\u7aef\uff0c\u5bb9\u91cf\u653e\u5927\u4e14\u80fd\u8de8\u6703\u8a71\u91cd\u7528<\/li>\n        <li onclick=\"checkAnswer(3, 'B', 'A')\">B. \u628a KV Cache \u5168\u90e8\u585e\u9032 GPU \u986f\u5b58\u52a0\u901f<\/li>\n        <li onclick=\"checkAnswer(3, 'C', 'A')\">C. \u8b93\u6a21\u578b\u4e0d\u9700\u8981\u7b97 prefill \u968e\u6bb5<\/li>\n        <li onclick=\"checkAnswer(3, 'D', 'A')\">D. \u81ea\u52d5\u6539\u5beb\u4f7f\u7528\u8005\u7684 prompt \u8b93\u524d\u7db4\u4e00\u81f4<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb3\"><\/div>\n    <\/div>\n\n    <div class=\"question\" id=\"q4\">\n      <h3>4. \u4e0b\u5217\u54ea\u500b\u5834\u666f\u6700\u80fd\u5f9e LMCache \u5f97\u5230\u9ad8\u6548\u76ca\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(4, 'A', 'C')\">A. \u6bcf\u6b21 prompt \u90fd\u5b8c\u5168\u4e0d\u540c\u7684\u4e00\u6b21\u6027\u4efb\u52d9<\/li>\n        <li onclick=\"checkAnswer(4, 'B', 'C')\">B. \u524d\u7db4\u53ea\u6709\u5e7e\u5341\u500b token \u7684\u77ed\u8acb\u6c42<\/li>\n        <li onclick=\"checkAnswer(4, 'C', 'C')\">C. RAG \u53cd\u8986\u5c0d\u540c\u4e00\u6279\u6587\u4ef6 context \u63d0\u554f<\/li>\n        <li onclick=\"checkAnswer(4, 'D', 'C')\">D. \u53ea\u751f\u6210\u55ae\u4e00 token \u7684\u5206\u985e\u4efb\u52d9<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb4\"><\/div>\n    <\/div>\n\n    <div class=\"question\" id=\"q5\">\n      <h3>5. \u4f5c\u70ba Vibe Coder\uff0c\u8981\u78ba\u8a8d KV Cache \u771f\u7684\u6709\u547d\u4e2d\uff0c\u6700\u5be6\u5728\u7684\u9a57\u6536\u65b9\u5f0f\u662f\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(5, 'A', 'D')\">A. \u770b AI \u6709\u6c92\u6709\u8aaa\u300c\u5feb\u53d6\u5df2\u555f\u7528\u300d<\/li>\n        <li onclick=\"checkAnswer(5, 'B', 'D')\">B. \u78ba\u8a8d\u555f\u52d5\u6307\u4ee4\u88e1\u6709 &#8211;enable-prefix-caching \u5c31\u597d<\/li>\n        <li onclick=\"checkAnswer(5, 'C', 'D')\">C. \u6aa2\u67e5 GPU \u986f\u5b58\u7528\u91cf\u6709\u6c92\u6709\u8b8a\u9ad8<\/li>\n        <li onclick=\"checkAnswer(5, 'D', 'D')\">D. \u5c0d\u540c\u4e00\u500b\u9577 prompt \u6253\u5169\u6b21\uff0c\u6bd4\u8f03\u7b2c\u4e8c\u6b21 TTFT \u6709\u6c92\u6709\u660e\u986f\u4e0b\u964d<\/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: \"LMCache \u662f KV Cache \u7684\u5916\u63a5\u5009\u5eab\uff0c\u628a\u7b97\u904e\u7684\u524d\u7db4 KV Cache \u5b58\u8d77\u4f86\u8de8\u8acb\u6c42\u91cd\u7528\uff0c\u7701\u4e0b\u91cd\u7b97\u7684 prefill \u6210\u672c\u3001\u964d\u4f4e TTFT\u3002\",\n        2: \"vLLM \u7528 --enable-prefix-caching \u958b\u555f\u5167\u5efa\u524d\u7db4\u5feb\u53d6\uff1bLMCache \u5efa\u7acb\u5728\u5b83\u4e4b\u4e0a\u3002--gpu-memory-utilization \u662f\u8abf\u986f\u5b58\u6bd4\u4f8b\u7684\u3002\",\n        3: \"\u5167\u5efa prefix caching \u7684 block \u53ea\u6d3b\u5728 GPU \u986f\u5b58\uff0c\u91cf\u5c0f\u3001\u91cd\u555f\u5c31\u6c92\u3002LMCache \u591a\u63a5\u4e86 CPU\/\u78c1\u789f\/\u9060\u7aef\u5c64\uff0c\u5bb9\u91cf\u5927\u4e14\u80fd\u8de8\u6703\u8a71\u3001\u8de8\u5be6\u4f8b\u91cd\u7528\u3002\",\n        4: \"RAG \u53cd\u8986\u554f\u540c\u4e00\u6279\u6587\u4ef6\uff0c\u524d\u7db4 context \u4e00\u6a21\u4e00\u6a23\u3001\u53c8\u9577\uff0c\u91cd\u8907\u7387\u6700\u9ad8\uff0c\u6548\u76ca\u6700\u5927\u3002\u524d\u7db4\u4e0d\u91cd\u8907\u6216\u5f88\u77ed\u7684\u5834\u666f\u6548\u76ca\u4f4e\uff0c\u4e0d\u503c\u5f97\u5c0e\u5165\u3002\",\n        5: \"flag \u6709\u958b\u4e0d\u4ee3\u8868\u771f\u7684\u547d\u4e2d\uff08\u524d\u7db4\u5dee\u4e00\u500b token \u5c31\u6574\u6bb5 miss\uff09\u3002\u6700\u5be6\u5728\u7684\u662f\u540c\u4e00\u500b prompt \u6253\u5169\u6b21\uff0c\u770b\u7b2c\u4e8c\u6b21 TTFT \u6709\u6c92\u6709\u660e\u986f\u4e0b\u964d\u3002\"\n      };\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) => {\n            if (opt.textContent.startsWith(selected + \".\")) opt.classList.add(\"wrong\");\n          });\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 +\n            (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 4 \u7bc7\uff08\u5171 4 \u7bc7\uff09\uff5c\u96e3\u5ea6\uff1aL3-\u719f\u7df4<br>\u524d\u7f6e\u77e5\u8b58\uff1a\u5df2\u8b80\u7b2c 1\u30012\u30013 \u7bc7\uff0c\u719f\u6089 KV Cache\u3001PagedAttention \u8207 vLLM \u8abf\u53c3<\/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>LMCache \u5c31\u662f\u300cKV Cache \u7684\u5916\u63a5\u5009\u5eab\u300d<\/strong>\uff1a\u628a\u7b97\u904e\u7684\u524d\u7db4\uff08system prompt\u3001RAG \u6587\u4ef6\u3001\u5c0d\u8a71\u6b77\u53f2\uff09\u7684 KV Cache \u5b58\u5230 CPU \u8a18\u61b6\u9ad4\u3001\u672c\u5730\u78c1\u789f\u751a\u81f3\u9060\u7aef\uff0c\u4e0b\u6b21\u9047\u5230\u4e00\u6a23\u7684\u524d\u7db4\u5c31\u76f4\u63a5\u642c\u56de\u4f86\u7528\uff0c\u4e0d\u7528\u91cd\u7b97\uff0cTTFT\uff08\u7b2c\u4e00\u500b token \u51fa\u4f86\u7684\u6642\u9593\uff09\u76f4\u63a5\u780d\u6389\u4e00\u5927\u622a\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5982\u679c\u4f60\u662f\u7528 AI \u5e6b\u4f60\u63a5 vLLM \u7684 Vibe Coder\uff0c\u9019\u7bc7\u6559\u4f60<strong>\u770b\u61c2 AI \u7522\u7684\u8a2d\u5b9a\u6a94\u5728\u5e79\u561b\u3001\u600e\u9ebc\u78ba\u8a8d\u5feb\u53d6\u771f\u7684\u6709\u547d\u4e2d\u3001\u54ea\u4e9b\u5751\u8981\u76ef\u8457<\/strong>\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\u61c2\u554f\u984c\uff1a\u70ba\u4ec0\u9ebc\u8981\u5feb\u53d6 KV Cache<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u56de\u9867\u4e00\u4e0b\u524d\u5e7e\u7bc7\uff1a\u6a21\u578b\u6bcf\u6536\u5230\u4e00\u6bb5 prompt\uff0c\u6703\u628a\u6bcf\u500b token \u7b97\u6210\u4e00\u5806\u4e2d\u9593\u72c0\u614b\uff08\u5c31\u662f KV Cache\uff09\u5b58\u5728 GPU \u986f\u5b58\u88e1\uff0c\u5f8c\u9762\u751f\u6210\u624d\u7528\u5f97\u5230\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u554f\u984c\u4f86\u4e86\u2014\u2014\u5f88\u591a\u8acb\u6c42\u7684\u300c\u958b\u982d\u300d\u5176\u5be6\u4e00\u6a21\u4e00\u6a23\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\" aria-describedby=\"shcb-language-1\" data-shcb-language-name=\"CSS\" data-shcb-language-slug=\"css\"><span><code class=\"hljs language-css\">\u8acb\u6c42 <span class=\"hljs-selector-tag\">A<\/span>: <span class=\"hljs-selector-attr\">&#91;3000 \u5b57\u7684 system prompt]<\/span> + \u300c\u5e6b\u6211\u5beb\u4e00\u5c01\u4fe1\u300d\n\u8acb\u6c42 <span class=\"hljs-selector-tag\">B<\/span>: <span class=\"hljs-selector-attr\">&#91;3000 \u5b57\u7684 system prompt]<\/span> + \u300c\u5e6b\u6211\u7ffb\u8b6f\u9019\u6bb5\u300d\n\u8acb\u6c42 <span class=\"hljs-selector-tag\">C<\/span>: <span class=\"hljs-selector-attr\">&#91;3000 \u5b57\u7684 system prompt]<\/span> + \u300c\u7e3d\u7d50\u4e00\u4e0b\u91cd\u9ede\u300d\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-1\"><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\">\u9019 3000 \u5b57\u7684 system prompt\uff0c\u6bcf\u4e00\u6b21\u8acb\u6c42\u90fd\u8981\u91cd\u65b0\u7b97\u4e00\u904d KV Cache\u3002\u9019\u53eb <strong>prefill\uff08\u9810\u586b\u5145\uff09<\/strong>\uff0c\u662f TTFT \u7684\u4e3b\u8981\u6210\u672c\u3002\u524d\u7db4\u8d8a\u9577\uff0c\u4f7f\u7528\u8005\u7b49\u8d8a\u4e45\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">RAG \u66f4\u8a87\u5f35\uff1a\u4f60\u628a\u540c\u4e00\u4efd\u6587\u4ef6\u585e\u9032 context \u554f\u5341\u500b\u554f\u984c\uff0c\u90a3\u4efd\u6587\u4ef6\u7684 KV Cache \u5c31\u88ab\u767d\u7b97\u4e86\u5341\u6b21\u3002\u591a\u8f2a\u5c0d\u8a71\u4e5f\u4e00\u6a23\uff0c\u7b2c 5 \u8f2a\u8981\u628a\u524d 4 \u8f2a\u7684\u6b77\u53f2\u5168\u90e8\u91cd\u7b97\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u6838\u5fc3\u6d1e\u5bdf\uff1a\u76f8\u540c\u7684\u524d\u7db4 \u2192 \u76f8\u540c\u7684 KV Cache \u2192 \u61c9\u8a72\u53ea\u7b97\u4e00\u6b21\u3002<\/strong><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u7b2c\u4e00\u5c64\uff1avLLM \u5167\u5efa\u7684 Automatic Prefix Caching<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">vLLM \u81ea\u5df1\u5c31\u6709\u4e00\u500b\u8f15\u91cf\u7248\u89e3\u6cd5\uff0c\u4e00\u500b flag \u5c31\u958b\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 meta-llama\/Llama-3.1-8B-Instruct \\\n  --enable-prefix-caching\n<\/code><\/span><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u9019\u5728\u5e79\u561b<\/strong>\uff1a\u958b\u555f\u5f8c\uff0cvLLM \u6703\u8a18\u4f4f\u300c\u5df2\u7d93\u7b97\u904e\u7684\u524d\u7db4\u300d\u5c0d\u61c9\u7684 KV Cache block\uff08\u9084\u8a18\u5f97\u7b2c 2 \u7bc7 PagedAttention \u628a KV Cache \u5207\u6210\u4e00\u584a\u584a block \u55ce\uff1f\u5c31\u662f\u90a3\u500b\uff09\u3002\u4e0b\u4e00\u500b\u8acb\u6c42\u5982\u679c\u958b\u982d token \u5b8c\u5168\u4e00\u6a23\uff0c\u5c31\u76f4\u63a5\u91cd\u7528\u90a3\u4e9b block\uff0c\u8df3\u904e\u91cd\u7b97\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\">\u65b0\u7248 vLLM\uff080.6+\uff09\u7684 V1 \u5f15\u64ce\u9810\u8a2d\u5c31\u958b prefix caching\uff0c\u4f60\u53ef\u80fd\u9023 flag \u90fd\u4e0d\u7528\u52a0\u3002\u770b\u5230 AI \u6c92\u52a0\u9019\u500b flag \u5225\u6025\u8457\u8aaa\u5b83\u6f0f\u4e86\uff0c\u5148\u78ba\u8a8d\u7248\u672c\u3002<\/p>\n<\/blockquote>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u4f46\u5b83\u6709\u500b\u5929\u82b1\u677f<\/strong>\uff1a\u9019\u4e9b\u91cd\u7528\u7684 block \u53ea\u6d3b\u5728 <strong>GPU \u986f\u5b58<\/strong>\u88e1\u3002\u986f\u5b58\u5f88\u8cb4\u5f88\u5c0f\uff0c\u4e00\u65e6\u88dd\u4e0d\u4e0b\u5c31\u6703\u88ab\u64e0\u6389\uff08evict\uff09\u3002\u6240\u4ee5\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n\n<li>\u524d\u7db4\u591a\u5230 GPU \u653e\u4e0d\u4e0b \u2192 \u820a\u7684\u88ab\u8e22\u6389 \u2192 \u4e0b\u6b21\u53c8\u8981\u91cd\u7b97<\/li>\n\n\n<li>vLLM \u91cd\u555f \u2192 \u5feb\u53d6\u5168\u6c92\u4e86 \u2192 \u91cd\u982d\u4f86\u904e<\/li>\n\n\n<li>\u591a\u53f0\u6a5f\u5668 \u2192 \u5404\u81ea\u70ba\u653f\uff0cA \u6a5f\u5668\u7b97\u904e\u7684 B \u6a5f\u5668\u7528\u4e0d\u5230<\/li>\n\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u5c31\u662f LMCache \u8981\u88dc\u7684\u6d1e\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u7b2c\u4e8c\u5c64\uff1aLMCache \u662f\u4ec0\u9ebc<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">LMCache \u662f\u4e00\u500b<strong>\u5916\u639b\u7684 KV Cache \u5206\u5c64\u5132\u5b58\u7cfb\u7d71<\/strong>\u3002\u5b83\u628a GPU \u653e\u4e0d\u4e0b\u7684 KV Cache\u300c\u5378\u8f09\uff08offload\uff09\u300d\u5230\u66f4\u5927\u66f4\u4fbf\u5b9c\u7684\u5730\u65b9\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\">        \u5feb              \u2192              \u6162\n        \u5c0f              \u2192              \u5927\n  \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510  \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510  \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510  \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n  \u2502   GPU     \u2502  \u2502   CPU     \u2502  \u2502  \u672c\u5730\u78c1\u789f  \u2502  \u2502  \u9060\u7aef     \u2502\n  \u2502  \u986f\u5b58     \u2502  \u2502  \u8a18\u61b6\u9ad4    \u2502  \u2502  (SSD)    \u2502  \u2502 (Redis\u7b49) \u2502\n  \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518  \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518  \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518  \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n   vLLM \u539f\u751f      \u2190 LMCache \u5e6b\u4f60\u591a\u63a5\u7684\u9019\u5e7e\u5c64 \u2192\n<\/code><\/span><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u7528\u767d\u8a71\u8b1b<\/strong>\uff1aGPU \u986f\u5b58\u50cf\u4f60\u684c\u4e0a\u7684\u4e00\u5c0f\u584a\u7a7a\u9593\uff0c\u53ea\u653e\u6700\u5e38\u7528\u7684\u6771\u897f\uff1bLMCache \u5e6b\u4f60\u591a\u958b\u4e86\u62bd\u5c5c\uff08CPU\uff09\u3001\u6ac3\u5b50\uff08\u78c1\u789f\uff09\u3001\u751a\u81f3\u5009\u5eab\uff08\u9060\u7aef\uff09\uff0c\u6771\u897f\u653e\u4e0d\u4e0b\u5c31\u5f80\u5916\u632a\uff0c\u8981\u7528\u518d\u62ff\u56de\u4f86\u3002\u5c31\u7b97\u300c\u62ff\u56de\u4f86\u300d\u4e5f\u6bd4\u300c\u91cd\u65b0\u7b97\u4e00\u904d\u300d\u5feb\u5f97\u591a\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5b83\u5e36\u4f86\u4e09\u500b GPU \u539f\u751f\u505a\u4e0d\u5230\u7684\u80fd\u529b\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n\n<li><strong>\u8de8\u6703\u8a71\u91cd\u7528<\/strong>\uff1a\u4f7f\u7528\u8005\u95dc\u6389\u5c0d\u8a71\u660e\u5929\u518d\u4f86\uff0c\u524d\u7db4 KV Cache \u9084\u5728\u78c1\u789f\u4e0a\uff0c\u4e0d\u7528\u91cd\u7b97\u3002<\/li>\n\n\n<li><strong>\u5bb9\u91cf\u653e\u5927<\/strong>\uff1aCPU \u8a18\u61b6\u9ad4\u901a\u5e38\u662f\u986f\u5b58\u7684\u597d\u5e7e\u500d\uff0c\u80fd\u5feb\u53d6\u7684\u524d\u7db4\u591a\u975e\u5e38\u591a\u3002<\/li>\n\n\n<li><strong>\u8de8\u5be6\u4f8b\u5171\u4eab<\/strong>\uff08\u9032\u968e\uff09\uff1a\u591a\u500b vLLM \u5be6\u4f8b\u53ef\u4ee5\u5171\u7528\u4e00\u500b\u9060\u7aef KV Cache \u6c60\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\">\u5be6\u4f5c\uff1a\u628a LMCache \u63a5\u4e0a vLLM<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">\u6b65\u9a5f 1\uff1a\u5b89\u88dd<\/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\">pip install lmcache\n<\/code><\/span><\/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\">LMCache \u5c0d vLLM \u7248\u672c\u5f88\u654f\u611f\u3002\u52d9\u5fc5\u78ba\u8a8d\u5169\u8005\u7248\u672c\u76f8\u5bb9\uff08\u898b\u5b98\u65b9\u76f8\u5bb9\u8868\uff09\uff0c\u7248\u672c\u5c0d\u4e0d\u4e0a\u662f\u6700\u5e38\u898b\u7684\u7ffb\u8eca\u9ede\u3002<\/p>\n<\/blockquote>\n\n\n\n<h3 class=\"wp-block-heading\">\u6b65\u9a5f 2\uff1a\u7528 KV Connector \u63a5\u4e0a vLLM<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u65b0\u7248 vLLM \u900f\u904e <strong>KV Connector<\/strong> \u9019\u500b\u63d2\u69fd\u6a5f\u5236\u63a5\u5916\u90e8\u5feb\u53d6\u3002\u4f60\u6703\u770b\u5230 AI \u7522\u51fa\u985e\u4f3c\u9019\u6a23\u7684\u555f\u52d5\u6307\u4ee4\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\" aria-describedby=\"shcb-language-2\" data-shcb-language-name=\"JavaScript\" data-shcb-language-slug=\"javascript\"><span><code class=\"hljs language-javascript\">LMCACHE_CONFIG_FILE=lmcache.yaml \\\nvllm serve meta-llama\/Llama<span class=\"hljs-number\">-3.1<\/span><span class=\"hljs-number\">-8<\/span>B-Instruct \\\n  --enable-prefix-caching \\\n  --kv-transfer-config \\\n  <span class=\"hljs-string\">'{\"kv_connector\":\"LMCacheConnectorV1\",\"kv_role\":\"kv_both\"}'<\/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\">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\"><strong>\u9010\u884c\u7ffb\u8b6f<\/strong>\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table>\n<thead>\n<tr>\n<th>\u9019\u4e00\u6bb5<\/th>\n<th>\u5728\u5e79\u561b<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><code>LMCACHE_CONFIG_FILE=lmcache.yaml<\/code><\/td>\n<td>\u7528\u74b0\u5883\u8b8a\u6578\u6307\u5b9a LMCache \u7684\u8a2d\u5b9a\u6a94\u4f4d\u7f6e<\/td>\n<\/tr>\n<tr>\n<td><code>--enable-prefix-caching<\/code><\/td>\n<td>\u5148\u628a vLLM \u81ea\u5df1\u7684\u524d\u7db4\u5feb\u53d6\u6253\u958b\uff08LMCache \u5efa\u5728\u5b83\u4e4b\u4e0a\uff09<\/td>\n<\/tr>\n<tr>\n<td><code>--kv-transfer-config '{...}'<\/code><\/td>\n<td>\u544a\u8a34 vLLM\u300c\u7528\u4e00\u500b\u5916\u90e8 connector \u4f86\u642c KV Cache\u300d<\/td>\n<\/tr>\n<tr>\n<td><code>\"kv_connector\":\"LMCacheConnectorV1\"<\/code><\/td>\n<td>\u6307\u5b9a\u7528 LMCache \u7684 connector<\/td>\n<\/tr>\n<tr>\n<td><code>\"kv_role\":\"kv_both\"<\/code><\/td>\n<td>\u9019\u500b\u5be6\u4f8b<strong>\u540c\u6642<\/strong>\u80fd\u5b58 KV\u3001\u4e5f\u80fd\u8b80 KV\uff08\u65e2\u5beb\u53c8\u8b80\uff09<\/td>\n<\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">\u6b65\u9a5f 3\uff1aLMCache \u8a2d\u5b9a\u6a94\uff08lmcache.yaml\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\"><span class=\"hljs-comment\"># \u6bcf\u500b KV chunk \u7684 token \u6578\uff0c\u8981\u8ddf vLLM \u5c0d\u9f4a<\/span>\nchunk_size: <span class=\"hljs-number\">256<\/span>\n\n<span class=\"hljs-comment\"># \u958b\u555f CPU \u8a18\u61b6\u9ad4\u9019\u4e00\u5c64\uff0c\u7d66\u5b83 20GB<\/span>\nlocal_cpu: <span class=\"hljs-keyword\">true<\/span>\nmax_local_cpu_size: <span class=\"hljs-number\">20<\/span>      <span class=\"hljs-comment\"># \u55ae\u4f4d GB<\/span>\n\n<span class=\"hljs-comment\"># \u958b\u555f\u672c\u5730\u78c1\u789f\u9019\u4e00\u5c64\uff0c\u7d66\u5b83 100GB<\/span>\nlocal_disk: <span class=\"hljs-string\">\"file:\/\/\/tmp\/lmcache\"<\/span>\nmax_local_disk_size: <span class=\"hljs-number\">100<\/span>    <span class=\"hljs-comment\"># \u55ae\u4f4d GB<\/span>\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\u4efd\u8a2d\u5b9a\u5728\u5e79\u561b<\/strong>\uff1a\u544a\u8a34 LMCache\u300cGPU \u653e\u4e0d\u4e0b\u7684 KV Cache\uff0c\u5148\u5f80 CPU \u632a\uff08\u6700\u591a 20GB\uff09\uff0cCPU \u4e5f\u6eff\u4e86\u5c31\u5beb\u5230\u78c1\u789f \/tmp\/lmcache\uff08\u6700\u591a 100GB\uff09\u300d\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e5f\u53ef\u4ee5\u5b8c\u5168\u7528\u74b0\u5883\u8b8a\u6578\u8a2d\u5b9a\uff0cAI \u6709\u6642\u6703\u9019\u6a23\u5beb\uff0c\u770b\u5230\u5225\u56f0\u60d1\uff0c\u662f\u540c\u4e00\u4ef6\u4e8b\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\" aria-describedby=\"shcb-language-4\" data-shcb-language-name=\"JavaScript\" data-shcb-language-slug=\"javascript\"><span><code class=\"hljs language-javascript\"><span class=\"hljs-keyword\">export<\/span> LMCACHE_CHUNK_SIZE=<span class=\"hljs-number\">256<\/span>\n<span class=\"hljs-keyword\">export<\/span> LMCACHE_LOCAL_CPU=True\n<span class=\"hljs-keyword\">export<\/span> LMCACHE_MAX_LOCAL_CPU_SIZE=<span class=\"hljs-number\">20<\/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\">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<h3 class=\"wp-block-heading\">\u2705 \u5fc5\u770b\u61c2 \/ \ud83d\udccc \u77e5\u9053\u5c31\u597d<\/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\">\u2705 \u5fc5\u770b\u61c2\uff08\u5224\u65b7 AI \u6709\u6c92\u6709\u63a5\u5c0d\uff09\uff1a\n- --enable-prefix-caching \u6709\u6c92\u6709\u958b\n- --kv-transfer-config \u88e1\u6709\u6c92\u6709 LMCacheConnector\n- lmcache.yaml \u6709\u6c92\u6709\u771f\u7684\u5206\u914d CPU \/ \u78c1\u789f\u7a7a\u9593\n\n\ud83d\udccc \u77e5\u9053\u5c31\u597d\uff08\u9047\u5230\u518d\u67e5\uff09\uff1a\n- kv_role \u7684 kv_producer \/ kv_consumer \u62c6\u5206\uff08\u505a P\/D \u5206\u96e2\u624d\u9700\u8981\uff09\n- chunk_size \u600e\u9ebc\u8abf\uff08\u9810\u8a2d\u80fd\u8dd1\u5c31\u5148\u5225\u52d5\uff09\n\n\ud83d\udd70\ufe0f \u8a8d\u5f97\u5c31\u597d\uff08\u820a\u5beb\u6cd5\uff0c\u7dad\u8b77\u820a\u5c08\u6848\u6703\u9047\u5230\uff09\uff1a\n- LMCacheConnector\uff08V1 \u4e4b\u524d\u7684\u820a connector \u540d\u7a31\uff09\n- \u820a\u7248\u7528 --kv-transfer-config \u4ee5\u5916\u7684\u63a5\u6cd5\n<\/code><\/span><\/pre><\/div>\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u4ec0\u9ebc\u5834\u666f\u503c\u5f97\u5c0e\u5165\uff1f\u6548\u76ca\u600e\u9ebc\u5224\u65b7<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">LMCache \u4e0d\u662f\u842c\u9748\u4e39\u3002\u5224\u65b7\u503c\u4e0d\u503c\u5f97\uff0c\u5c31\u770b\u4e00\u500b\u554f\u984c\uff1a<strong>\u4f60\u7684\u8acb\u6c42\u524d\u7db4\u91cd\u8907\u7387\u9ad8\u4e0d\u9ad8\uff1f<\/strong><\/p>\n\n\n\n<figure class=\"wp-block-table\"><table>\n<thead>\n<tr>\n<th>\u5834\u666f<\/th>\n<th>\u524d\u7db4\u91cd\u8907\u60c5\u6cc1<\/th>\n<th>\u6548\u76ca<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>RAG\uff08\u540c\u4e00\u6279\u6587\u4ef6\u53cd\u8986\u554f\uff09<\/td>\n<td>\u6587\u4ef6 context \u4e00\u6a21\u4e00\u6a23<\/td>\n<td>\u2b50\u2b50\u2b50 \u5f88\u9ad8<\/td>\n<\/tr>\n<tr>\n<td>\u5171\u7528\u9577 system prompt<\/td>\n<td>\u6bcf\u500b\u8acb\u6c42\u958b\u982d\u90fd\u76f8\u540c<\/td>\n<td>\u2b50\u2b50\u2b50 \u5f88\u9ad8<\/td>\n<\/tr>\n<tr>\n<td>\u9577\u591a\u8f2a\u5c0d\u8a71<\/td>\n<td>\u6b77\u53f2\u8d8a\u7a4d\u8d8a\u9577\u4e14\u91cd\u8907<\/td>\n<td>\u2b50\u2b50 \u9ad8<\/td>\n<\/tr>\n<tr>\n<td>\u6bcf\u6b21 prompt \u90fd\u4e0d\u4e00\u6a23\u7684\u4e00\u6b21\u6027\u4efb\u52d9<\/td>\n<td>\u5e7e\u4e4e\u4e0d\u91cd\u8907<\/td>\n<td>\u2b50 \u4f4e\uff0c\u4e0d\u503c\u5f97<\/td>\n<\/tr>\n<tr>\n<td>\u524d\u7db4\u5f88\u77ed\uff08\u5e7e\u5341\u500b token\uff09<\/td>\n<td>\u91cd\u7b97\u6210\u672c\u672c\u4f86\u5c31\u4f4e<\/td>\n<td>\u2b50 \u4f4e\uff0c\u4e0d\u503c\u5f97<\/td>\n<\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u5224\u65b7\u5fc3\u6cd5<\/strong>\uff1a\u524d\u7db4\u8d8a\u9577\u3001\u91cd\u8907\u8d8a\u591a\u3001prefill \u4f54\u6bd4\u8d8a\u9ad8\uff0cLMCache \u8cfa\u8d8a\u591a\u3002\u53cd\u904e\u4f86\uff0c\u5982\u679c\u4f60\u7684\u8acb\u6c42\u53c8\u77ed\u53c8\u4e0d\u91cd\u8907\uff0c\u52a0\u4e86 LMCache \u53ea\u662f\u5f92\u589e\u8907\u96dc\u5ea6\u548c\u4e00\u5c64\u642c\u904b\u958b\u92b7\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e00\u500b\u7c97\u7565\u7684\u76f4\u89ba\uff1a\u5982\u679c\u4e00\u500b\u8acb\u6c42\u7684 TTFT \u4e3b\u8981\u82b1\u5728\u7b97\u90a3 3000 \u5b57\u7684\u56fa\u5b9a system prompt \u4e0a\uff0cLMCache \u547d\u4e2d\u5f8c\uff0cTTFT \u53ef\u80fd\u5f9e\u300c\u79d2\u7d1a\u300d\u6389\u5230\u300c\u767e\u6beb\u79d2\u7d1a\u300d\u3002\u524d\u7db4\u8d8a\u9577\uff0c\u9019\u500b\u5dee\u8ddd\u8d8a\u8a87\u5f35\u3002<\/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\">AI \u5e6b\u4f60\u63a5\u597d\u4e86\u8aaa\u300c\u641e\u5b9a\uff0c\u5feb\u53d6\u5df2\u555f\u7528\u300d\u2014\u2014\u5225\u4fe1\uff0c\u81ea\u5df1\u9a57\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u6aa2\u67e5\u9ede 1\uff1a\u78ba\u8a8d\u670d\u52d9\u6709\u639b\u4e0a LMCache<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u555f\u52d5 vLLM \u6642\u770b log\uff0c\u61c9\u8a72\u6709 LMCache \u521d\u59cb\u5316\u7684\u8a0a\u606f\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\">\u9810\u671f\u770b\u5230\uff1aLMCache \u76f8\u95dc\u7684 init log\uff0c\u4f8b\u5982\u8f09\u5165 connector\u3001\n         \u914d\u7f6e CPU\/disk backend \u7684\u884c\u3002\n\u6c92\u770b\u5230 \u2192 connector \u6c92\u63a5\u4e0a\uff0c\u591a\u534a\u662f --kv-transfer-config \u5beb\u932f\u6216\u7248\u672c\u4e0d\u5408\u3002\n<\/code><\/span><\/pre><\/div>\n\n\n<h3 class=\"wp-block-heading\">\u6aa2\u67e5\u9ede 2\uff1a\u8dd1\u5169\u6b21\u76f8\u540c\u524d\u7db4\uff0c\u6bd4 TTFT<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u662f\u6700\u5be6\u5728\u7684\u9a57\u6536\u2014\u2014<strong>\u540c\u4e00\u500b\u9577 prompt \u6253\u5169\u6b21<\/strong>\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\" aria-describedby=\"shcb-language-5\" data-shcb-language-name=\"CSS\" data-shcb-language-slug=\"css\"><span><code class=\"hljs language-css\">\u7b2c 1 \u6b21\u8acb\u6c42\uff1a\u51b7\u555f\u52d5\uff0c\u8981\u7b97 <span class=\"hljs-selector-tag\">KV<\/span> <span class=\"hljs-selector-tag\">Cache<\/span> \u2192 <span class=\"hljs-selector-tag\">TTFT<\/span> \u8f03\u9ad8\uff08\u4f8b\u5982 1<span class=\"hljs-selector-class\">.2s<\/span>\uff09\n\u7b2c 2 \u6b21\u8acb\u6c42\uff1a\u547d\u4e2d\u5feb\u53d6\uff0c\u76f4\u63a5\u642c \u2192 <span class=\"hljs-selector-tag\">TTFT<\/span> \u660e\u986f\u4e0b\u964d\uff08\u4f8b\u5982 0<span class=\"hljs-selector-class\">.2s<\/span>\uff09\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\">\u5982\u679c\u7b2c\u4e8c\u6b21\u6c92\u6709\u660e\u986f\u8b8a\u5feb\uff0c\u5c31\u662f<strong>\u5feb\u53d6\u6839\u672c\u6c92\u547d\u4e2d<\/strong>\uff0c\u63a5\u7684\u6709\u554f\u984c\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u6aa2\u67e5\u9ede 3\uff1a\u770b cache hit \u6307\u6a19 \/ \u65e5\u8a8c<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u89c0\u5bdf LMCache \u7684\u547d\u4e2d\u65e5\u8a8c\u6216 vLLM \u7684 metrics\uff08<code>\/metrics<\/code> \u7aef\u9ede\uff09\u88e1\u8ddf prefix cache hit rate \u76f8\u95dc\u7684\u6578\u5b57\u3002\u547d\u4e2d\u7387\u9ad8 = \u6709\u5728\u8cfa\uff1b\u4e00\u76f4\u662f 0 = \u767d\u63a5\u4e86\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\">\u4e0d\u78ba\u5b9a\u600e\u9ebc\u770b\u6307\u6a19\uff1f\u628a log \u8cbc\u7d66 AI \u554f\uff1a\u300c\u9019\u6bb5 vLLM\/LMCache \u7684 log \u88e1\uff0c\u600e\u9ebc\u5224\u65b7 KV Cache \u6709\u6c92\u6709\u547d\u4e2d\uff1f\u547d\u4e2d\u7387\u5728\u54ea\u4e00\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\">\ud83d\udea9 \u7d05\u65d7\uff1a\u9019\u4e9b\u5730\u65b9 AI \u7279\u5225\u5bb9\u6613\u8e29<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\ud83d\udea9 \u7d05\u65d7 1\uff1a\u53ea\u52a0\u4e86 flag \u537b\u6c92\u9a57\u8b49\u547d\u4e2d<\/strong><\/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\" aria-describedby=\"shcb-language-6\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">vllm serve ... --enable-prefix-caching   <span class=\"hljs-comment\"># AI\uff1a\u300c\u5feb\u53d6\u5df2\u958b\u555f\uff01\u300d<\/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\">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\">\u958b\u4e86 flag \u2260 \u771f\u7684\u6709\u547d\u4e2d\u3002\u524d\u7db4\u53ea\u8981\u5dee\u4e00\u500b token\uff08\u9023\u7a7a\u683c\u3001\u6a19\u9ede\u90fd\u7b97\uff09\u5c31\u6574\u6bb5 miss\u3002\u4e00\u5b9a\u8981\u7528\u300c\u6aa2\u67e5\u9ede 2\u300d\u5be6\u6e2c TTFT\uff0c\u4e0d\u8981\u770b\u5230 flag \u5c31\u6536\u8ca8\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\ud83d\udea9 \u7d05\u65d7 2\uff1a\u7248\u672c\u6c92\u5c0d\u5c31\u76f4\u63a5\u88dd<\/strong><\/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\" aria-describedby=\"shcb-language-7\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">pip install lmcache   <span class=\"hljs-comment\"># \u6c92\u6307\u5b9a\u7248\u672c\uff0c\u4e5f\u6c92\u5c0d\u7167 vLLM \u7248\u672c<\/span>\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-7\"><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\">LMCache \u8ddf vLLM \u7d81\u5f88\u7dca\uff0c\u7248\u672c\u4e0d\u5408\u6703\u51fa\u73fe\u300cconnector \u8f09\u5165\u5931\u6557\u300d\u6216\u66f4\u96b1\u6666\u7684\u300c\u670d\u52d9\u8d77\u5f97\u4f86\u4f46\u5feb\u53d6\u6c38\u9060 miss\u300d\u3002\u770b\u5230 AI \u6c92\u63d0\u7248\u672c\u76f8\u5bb9\uff0c\u4e3b\u52d5\u554f\u5b83\uff1a\u300c\u6211\u7684 vLLM \u662f x.y.z\uff0c\u9019\u500b LMCache \u7248\u672c\u76f8\u5bb9\u55ce\uff1f\u300d<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\ud83d\udea9 \u7d05\u65d7 3\uff1a\u78c1\u789f\u5feb\u53d6\u7121\u4e0a\u9650\u6216\u585e\u5230\u7cfb\u7d71\u789f<\/strong><\/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\" aria-describedby=\"shcb-language-8\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">local_disk: <span class=\"hljs-string\">\"file:\/\/\/tmp\/lmcache\"<\/span>\n<span class=\"hljs-comment\"># \ud83d\udea9 \u6c92\u8a2d max_local_disk_size\uff0c\u6216\u6307\u5230\u6703\u88ab\u6e05\u7a7a\u7684 \/tmp<\/span>\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-8\"><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\">\u6c92\u8a2d\u4e0a\u9650\u7684\u78c1\u789f\u5feb\u53d6\u6703\u628a\u786c\u789f\u5403\u7206\uff1b\u6307\u5230 <code>\/tmp<\/code> \u53ef\u80fd\u91cd\u958b\u6a5f\u5c31\u6c92\u4e86\uff0c\u6216\u8ddf\u7cfb\u7d71\u6436\u7a7a\u9593\u3002\u8981\u78ba\u8a8d\u6709 <code>max_local_disk_size<\/code>\uff0c\u4e14\u8def\u5f91\u653e\u5728\u5920\u5927\u3001\u4e0d\u6703\u88ab\u6e05\u7684\u78c1\u789f\u4e0a\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\ud83d\udea9 \u7d05\u65d7 4\uff1a\u628a\u300c\u8a18\u61b6\u9ad4\u63db\u78c1\u789f\u300d\u7576\u514d\u8cbb<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5f80\u78c1\u789f\u5378\u8f09\u96d6\u7136\u6bd4\u91cd\u7b97\u5feb\uff0c\u4f46<strong>\u78c1\u789f\u8b80\u53d6 KV Cache \u4e5f\u662f\u6709\u5ef6\u9072\u7684<\/strong>\u3002\u5982\u679c\u4f60\u7684 SSD \u6162\u3001\u6216\u524d\u7db4\u5176\u5be6\u4e0d\u9577\uff0c\u300c\u5f9e\u78c1\u789f\u642c\u56de\u4f86\u300d\u53ef\u80fd\u6c92\u6bd4\u300c\u91cd\u7b97\u300d\u5feb\u591a\u5c11\u3002\u5225\u9810\u8a2d\u300c\u591a\u52a0\u5e7e\u5c64\u4e00\u5b9a\u66f4\u5feb\u300d\uff0c\u8981\u7528 TTFT \u5be6\u6e2c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\ud83d\udea9 \u7d05\u65d7 5\uff1a\u4ee5\u70ba\u5feb\u53d6\u6703\u81ea\u5df1\u300c\u66f4\u65b0\u300d<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">KV Cache \u662f\u7d81\u5b9a\u300c\u6a21\u578b + \u524d\u7db4\u5167\u5bb9\u300d\u7684\u3002\u63db\u4e86\u6a21\u578b\u3001\u6539\u4e86 system prompt \u5167\u5bb9\uff0c\u820a\u5feb\u53d6\u5c31\u8a72\u5931\u6548\u3002\u5982\u679c AI \u6c92\u8655\u7406\u5feb\u53d6\u5931\u6548\u908f\u8f2f\uff0c\u4f60\u53ef\u80fd\u5403\u5230\u300c\u4e0a\u4e00\u7248 prompt \u7b97\u51fa\u4f86\u7684\u820a\u5feb\u53d6\u300d\u3002\u6539\u52d5\u524d\u7db4\u5f8c\uff0c\u89c0\u5bdf TTFT \u662f\u5426\u56de\u5230\u51b7\u555f\u52d5\u6c34\u6e96\u4f86\u78ba\u8a8d\u820a\u5feb\u53d6\u6c92\u88ab\u8aa4\u7528\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u7cfb\u5217\u7e3d\u7d50<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u56db\u7bc7\u8d70\u4e0b\u4f86\uff0c\u4f60\u624b\u4e0a\u7684 vLLM \u5fc3\u667a\u6a21\u578b\u61c9\u8a72\u662f\u9019\u6a23\u7684\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n\n<li><strong>#01 KV Cache<\/strong>\uff1a\u641e\u61c2\u70ba\u4ec0\u9ebc\u63a8\u8ad6\u6703\u5403\u986f\u5b58\u3001KV Cache \u662f\u4ec0\u9ebc\u3002<\/li>\n\n\n<li><strong>#02 PagedAttention<\/strong>\uff1a\u986f\u5b58\u600e\u9ebc\u88ab\u5207\u6210 block \u9ad8\u6548\u7ba1\u7406\u3001\u524d\u7db4\u600e\u9ebc\u5171\u4eab\u3002<\/li>\n\n\n<li><strong>#03 \u8abf\u53c3<\/strong>\uff1a<code>--gpu-memory-utilization<\/code>\u3001<code>--max-model-len<\/code> \u9019\u4e9b\u65cb\u9215\u600e\u9ebc\u8f49\u3002<\/li>\n\n\n<li><strong>#04 LMCache\uff08\u672c\u7bc7\uff09<\/strong>\uff1a\u628a KV Cache \u5feb\u53d6\u5230 GPU \u4e4b\u5916\uff0c\u8b93\u91cd\u8907\u524d\u7db4\u8de8\u8acb\u6c42\u3001\u8de8\u6703\u8a71\u91cd\u7528\u3002<\/li>\n\n<\/ol>\n\n\n\n<p class=\"wp-block-paragraph\">\u4f5c\u70ba Vibe Coder\uff0c\u4f60\u4e0d\u9700\u8981\u81ea\u5df1\u523b\u9019\u4e9b\u7cfb\u7d71\uff0c\u4f46\u4f60\u8981\u80fd<strong>\u770b\u61c2 AI \u63a5\u7684\u8a2d\u5b9a\u3001\u5be6\u6e2c\u9a57\u6536\u3001\u6293\u51fa\u300c\u770b\u8d77\u4f86\u6709\u958b\u5176\u5be6\u6c92\u547d\u4e2d\u300d\u7684\u5047\u8c61<\/strong>\u3002\u9019\u5c31\u662f\u628a AI \u7576\u6210\u4e00\u500b\u8d85\u5feb\u4f46\u5076\u723e\u552c\u721b\u7684\u5de5\u7a0b\u5e2b\u6642\uff0c\u4f60\u9019\u500b\u9a57\u6536\u8005\u8a72\u6709\u7684\u624b\u611f\u3002<\/p>\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\u9019\u6bb5 vLLM \u555f\u52d5\u6307\u4ee4\u88e1\u7684 <code>--kv-transfer-config<\/code> \u5728\u5e79\u561b\uff0c\u6211\u662f Vibe Coder\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\u7684 vLLM \u7248\u672c\u662f ___\uff0c\u8981\u88dd\u54ea\u500b\u7248\u672c\u7684 LMCache \u624d\u76f8\u5bb9\uff1f\u5e6b\u6211\u5217\u51fa\u5c0d\u7167\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\u5e6b\u6211\u5beb\u4e00\u500b\u5c0f\u8173\u672c\uff0c\u5c0d\u540c\u4e00\u500b\u9577 prompt \u9023\u6253\u5169\u6b21\uff0c\u5370\u51fa\u5169\u6b21\u7684 TTFT\uff0c\u8b93\u6211\u78ba\u8a8d KV Cache \u6709\u6c92\u6709\u547d\u4e2d\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\u7684\u5834\u666f\u662f ___\uff08RAG\uff0f\u9577 system prompt\uff0f\u591a\u8f2a\u5c0d\u8a71\uff09\uff0c\u524d\u7db4\u5927\u6982 ___ \u500b token\uff0c\u5c0e\u5165 LMCache \u5927\u6982\u80fd\u7701\u591a\u5c11 TTFT\uff1f\u503c\u5f97\u55ce\uff1f\u300d<\/p>\n<\/blockquote>\n\n\n\n<div id=\"quiz-9e0c2e10\">\n<style>\n#quiz-9e0c2e10 .badge { display: inline-block; padding: 4px 10px; border-radius: 12px; font-size: 0.75em; font-weight: bold; margin-left: 8px; }#quiz-9e0c2e10 .badge.scenario { background: #ebf8ff; color: #2b6cb0; }#quiz-9e0c2e10 .badge.debug { background: #fef3c7; color: #92400e; }#quiz-9e0c2e10 .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-9e0c2e10 .question-adv h3 { margin-top: 0; color: #2d3748; }#quiz-9e0c2e10 .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-9e0c2e10 .options-adv { list-style: none; padding: 0; }#quiz-9e0c2e10 .options-adv li { padding: 14px 16px; margin: 10px 0; border: 2px solid #e2e8f0; border-radius: 8px; cursor: pointer; transition: all 0.2s; }#quiz-9e0c2e10 .options-adv li:hover:not(.disabled) { border-color: #ed8936; background: #fffaf0; }#quiz-9e0c2e10 .options-adv li.disabled { cursor: default; opacity: 0.7; }#quiz-9e0c2e10 .options-adv li.show-answer { border-color: #48bb78; background: #f0fff4; }#quiz-9e0c2e10 .options-adv li.wrong { border-color: #fc8181; background: #fff5f5; }#quiz-9e0c2e10 .feedback-adv { margin-top: 16px; padding: 14px; border-radius: 8px; display: none; line-height: 1.6; }#quiz-9e0c2e10 .feedback-adv.correct { background: #f0fff4; color: #276749; display: block; }#quiz-9e0c2e10 .feedback-adv.wrong { background: #fff5f5; color: #c53030; display: block; }#quiz-9e0c2e10 .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-9e0c2e10 .intro { background: #ebf8ff; padding: 16px; border-radius: 8px; margin-bottom: 20px; color: #2c5282; }\n<\/style>\n<h1>\u9032\u968e\u6e2c\u9a57\uff1a\u7d50\u5408 LMCache \u52a0\u901f\u9577\u5c0d\u8a71\u8207 RAG<\/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. \u4f60\u8981\u90e8\u7f72\u4e00\u500b\u5ba2\u670d bot\uff0c\u6bcf\u500b\u8acb\u6c42\u90fd\u5e36\u540c\u4e00\u6bb5 4000 \u5b57\u7684 system prompt\uff0c\u4f7f\u7528\u8005\u5e38\u9694\u5929\u56de\u4f86\u5ef6\u7e8c\u5c0d\u8a71\u3002\u4f60\u60f3\u964d\u4f4e TTFT \u4e14\u8b93\u5feb\u53d6\u8de8\u6703\u8a71\u5b58\u6d3b\u3002\u6700\u9069\u5408\u7684\u505a\u6cd5\u662f\uff1f <span class=\"badge scenario\">\u60c5\u5883\u984c<\/span><\/h3>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(1, 'A', 'C')\">A. \u53ea\u52a0 &#8211;enable-prefix-caching \u5c31\u5920\u4e86\uff0c\u4e0d\u9700\u8981\u5225\u7684<\/li>\n        <li onclick=\"checkAnswerAdv(1, 'B', 'C')\">B. \u8abf\u5927 &#8211;gpu-memory-utilization \u8b93\u986f\u5b58\u585e\u66f4\u591a\u5feb\u53d6<\/li>\n        <li onclick=\"checkAnswerAdv(1, 'C', 'C')\">C. \u958b &#8211;enable-prefix-caching \u4e26\u63a5\u4e0a LMCache\uff0c\u628a KV Cache \u5378\u8f09\u5230 CPU\/\u78c1\u789f\u4ee5\u8de8\u6703\u8a71\u91cd\u7528<\/li>\n        <li onclick=\"checkAnswerAdv(1, 'D', 'C')\">D. \u7e2e\u77ed system prompt \u5230\u5e7e\u5341\u500b token<\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv1\"><\/div>\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv2\">\n      <h3>2. \u4f60\u7684\u670d\u52d9\u53ea\u505a\u300c\u6bcf\u6b21 prompt \u90fd\u5b8c\u5168\u4e0d\u540c\u3001\u524d\u7db4\u5f88\u77ed\u300d\u7684\u4e00\u6b21\u6027\u6458\u8981\u4efb\u52d9\uff0c\u540c\u4e8b\u5efa\u8b70\u5c0e\u5165 LMCache\u3002\u4f60\u8a72\u600e\u9ebc\u5224\u65b7\uff1f <span class=\"badge scenario\">\u60c5\u5883\u984c<\/span><\/h3>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(2, 'A', 'B')\">A. \u4e00\u5b9a\u8981\u5c0e\u5165\uff0c\u5feb\u53d6\u6c38\u9060\u662f\u597d\u4e8b<\/li>\n        <li onclick=\"checkAnswerAdv(2, 'B', 'B')\">B. \u524d\u7db4\u4e0d\u91cd\u8907\u53c8\u77ed\uff0c\u91cd\u7b97\u6210\u672c\u672c\u4f86\u5c31\u4f4e\uff0c\u5c0e\u5165\u6548\u76ca\u4f4e\u3001\u5f92\u589e\u8907\u96dc\u5ea6\uff0c\u5148\u4e0d\u505a<\/li>\n        <li onclick=\"checkAnswerAdv(2, 'C', 'B')\">C. \u5c0e\u5165\u5f8c\u628a chunk_size \u8abf\u5230\u6700\u5927\u5c31\u6703\u8b8a\u5feb<\/li>\n        <li onclick=\"checkAnswerAdv(2, 'D', 'B')\">D. \u53ea\u8981 GPU \u5920\u5927\u5c0e\u5165\u4e00\u5b9a\u66f4\u5feb<\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv2\"><\/div>\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv3\">\n      <h3>3. \u4f60\u60f3\u5728\u591a\u53f0 vLLM \u5be6\u4f8b\u4e4b\u9593\u5171\u4eab\u540c\u4e00\u500b KV Cache \u6c60\u3002\u95dc\u65bc kv_role \u8a2d\u5b9a\uff0c\u4e0b\u5217\u6558\u8ff0\u4f55\u8005\u6b63\u78ba\uff1f <span class=\"badge scenario\">\u60c5\u5883\u984c<\/span><\/h3>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(3, 'A', 'A')\">A. kv_role \u53ef\u8a2d kv_both\uff08\u65e2\u5beb\u53c8\u8b80\uff09\uff0c\u6216\u62c6\u6210 producer\/consumer \u5c0d\u61c9\u4e0d\u540c\u89d2\u8272<\/li>\n        <li onclick=\"checkAnswerAdv(3, 'B', 'A')\">B. kv_role \u53ea\u80fd\u662f true \u6216 false<\/li>\n        <li onclick=\"checkAnswerAdv(3, 'C', 'A')\">C. kv_role \u6c7a\u5b9a\u6a21\u578b\u8981\u751f\u6210\u5e7e\u500b token<\/li>\n        <li onclick=\"checkAnswerAdv(3, 'D', 'A')\">D. kv_role \u662f\u7528\u4f86\u8a2d\u5b9a\u986f\u5b58\u6bd4\u4f8b\u7684<\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv3\"><\/div>\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv4\">\n      <h3>4. \u5718\u968a\u53cd\u6620\u300c\u670d\u52d9\u8d77\u5f97\u4f86\uff0c\u4f46\u5feb\u53d6\u4e00\u76f4 miss\u3001TTFT \u6c92\u4e0b\u964d\u300d\uff0c\u555f\u52d5\u6307\u4ee4\u8207\u8a2d\u5b9a\u770b\u8d77\u4f86\u90fd\u6b63\u78ba\u3002\u6700\u53ef\u80fd\u7684\u6839\u56e0\u662f\uff1f <span class=\"badge debug\">\u932f\u8aa4\u8a3a\u65b7<\/span><\/h3>\n      <div class=\"code-block\">$ pip install lmcache   # \u76f4\u63a5\u88dd\uff0c\u6c92\u5c0d\u7167 vLLM \u7248\u672c\n$ vllm serve &#8230; &#8211;enable-prefix-caching \\\n    &#8211;kv-transfer-config &#8216;{&#8220;kv_connector&#8221;:&#8221;LMCacheConnectorV1&#8243;,&#8221;kv_role&#8221;:&#8221;kv_both&#8221;}&#8217;<\/div>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(4, 'A', 'B')\">A. GPU \u986f\u5b58\u4e0d\u5920\u5927<\/li>\n        <li onclick=\"checkAnswerAdv(4, 'B', 'B')\">B. LMCache \u8207 vLLM \u7248\u672c\u4e0d\u76f8\u5bb9\uff0cconnector \u6c92\u771f\u6b63\u751f\u6548\u5c0e\u81f4\u6c38\u9060 miss<\/li>\n        <li onclick=\"checkAnswerAdv(4, 'C', 'B')\">C. \u6c92\u6709\u5148\u57f7\u884c git pull<\/li>\n        <li onclick=\"checkAnswerAdv(4, 'D', 'B')\">D. system prompt \u592a\u9577\u8d85\u904e\u6a21\u578b\u4e0a\u9650<\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv4\"><\/div>\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv5\">\n      <h3>5. AI \u7522\u51fa\u7684 LMCache \u78c1\u789f\u5feb\u53d6\u8a2d\u5b9a\u5982\u4e0b\uff0c\u9019\u4efd\u8a2d\u5b9a\u6709\u4ec0\u9ebc\u96b1\u60a3\uff1f <span class=\"badge debug\">\u932f\u8aa4\u8a3a\u65b7<\/span><\/h3>\n      <div class=\"code-block\">local_disk: &#8220;file:\/\/\/tmp\/lmcache&#8221;\n# \u6c92\u6709\u8a2d\u5b9a max_local_disk_size<\/div>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(5, 'A', 'D')\">A. \u6c92\u554f\u984c\uff0c\u78c1\u789f\u5feb\u53d6\u6703\u81ea\u5df1\u7ba1\u7406\u5bb9\u91cf<\/li>\n        <li onclick=\"checkAnswerAdv(5, 'B', 'D')\">B. \u8def\u5f91\u5beb\u932f\u4e86\uff0c\u61c9\u8a72\u7528 http:\/\/<\/li>\n        <li onclick=\"checkAnswerAdv(5, 'C', 'D')\">C. \u78c1\u789f\u5feb\u53d6\u4e00\u5b9a\u6bd4 CPU \u5feb\uff0c\u61c9\u8a72\u53ea\u7559\u78c1\u789f<\/li>\n        <li onclick=\"checkAnswerAdv(5, 'D', 'D')\">D. \u6c92\u8a2d\u5bb9\u91cf\u4e0a\u9650\u53ef\u80fd\u628a\u786c\u789f\u5403\u7206\uff0c\u4e14 \/tmp \u53ef\u80fd\u91cd\u958b\u6a5f\u88ab\u6e05\u7a7a\u6216\u8ddf\u7cfb\u7d71\u6436\u7a7a\u9593<\/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: \"\u5167\u5efa prefix caching \u7684 block \u53ea\u6d3b\u5728 GPU \u986f\u5b58\uff0c\u91cd\u555f\u6216\u88ab evict \u5c31\u6c92\u4e86\uff0c\u8de8\u6703\u8a71\u5b58\u6d3b\u4e0d\u4e86\u3002\u8981\u8de8\u6703\u8a71\u91cd\u7528\u9577\u524d\u7db4\uff0c\u5fc5\u9808\u63a5 LMCache \u628a KV Cache \u5378\u8f09\u5230 CPU\/\u78c1\u789f\u3002\u8abf\u5927\u986f\u5b58\u6bd4\u4f8b\u6cbb\u6a19\u4e0d\u6cbb\u672c\uff0c\u7e2e\u77ed prompt \u5247\u72a7\u7272\u529f\u80fd\u3002\",\n        2: \"\u5224\u65b7 LMCache \u503c\u4e0d\u503c\u5f97\uff0c\u95dc\u9375\u770b\u524d\u7db4\u91cd\u8907\u7387\u3002\u6bcf\u6b21 prompt \u90fd\u4e0d\u540c\u3001\u524d\u7db4\u53c8\u77ed\uff0cprefill \u6210\u672c\u672c\u4f86\u5c31\u4f4e\uff0c\u5feb\u53d6\u5e7e\u4e4e miss\uff0c\u5c0e\u5165\u53ea\u662f\u589e\u52a0\u642c\u904b\u958b\u92b7\u8207\u8907\u96dc\u5ea6\u3002\u61c9\u4ee5\u524d\u7db4\u9577\u5ea6\u8207\u91cd\u8907\u7387\u3001\u4e26\u7528 TTFT \u5be6\u6e2c\u4f86\u6c7a\u5b9a\u3002\",\n        3: \"kv_role \u63cf\u8ff0\u9019\u500b\u5be6\u4f8b\u5728 KV \u50b3\u8f38\u4e2d\u7684\u89d2\u8272\uff1akv_both \u4ee3\u8868\u65e2\u80fd\u5beb\u5165\u4e5f\u80fd\u8b80\u53d6\uff1b\u505a P\/D \u5206\u96e2\u6216\u591a\u5be6\u4f8b\u5171\u4eab\u6642\u53ef\u62c6\u6210 producer\/consumer\u3002\u5b83\u8ddf\u751f\u6210 token \u6578\u3001\u986f\u5b58\u6bd4\u4f8b\u7121\u95dc\u3002\",\n        4: \"LMCache \u8207 vLLM \u7248\u672c\u7d81\u5f88\u7dca\u3002\u6c92\u5c0d\u7167\u76f8\u5bb9\u8868\u76f4\u63a5 pip install\uff0c\u5178\u578b\u75c7\u72c0\u5c31\u662f\u300c\u670d\u52d9\u8d77\u5f97\u4f86\u4f46\u5feb\u53d6\u6c38\u9060 miss\u300d\u6216 connector \u8f09\u5165\u5931\u6557\u3002\u4fee\u6b63\u65b9\u5f0f\u662f\u4f9d\u5b98\u65b9\u76f8\u5bb9\u8868\u5b89\u88dd\u5c0d\u61c9\u7248\u672c\uff1b\u986f\u5b58\u3001git pull\u3001prompt \u9577\u5ea6\u90fd\u4e0d\u662f\u9019\u88e1\u7684\u6839\u56e0\u3002\",\n        5: \"\u6c92\u6709 max_local_disk_size \u7684\u78c1\u789f\u5feb\u53d6\u53ef\u80fd\u7121\u9650\u6210\u9577\u628a\u786c\u789f\u5403\u7206\uff1b\u6307\u5230 \/tmp \u53ef\u80fd\u91cd\u958b\u6a5f\u88ab\u6e05\u7a7a\u3001\u6216\u8ddf\u7cfb\u7d71\u6436\u7a7a\u9593\u3002\u61c9\u8a2d\u5b9a\u5bb9\u91cf\u4e0a\u9650\u4e26\u653e\u5728\u5920\u5927\u3001\u4e0d\u6703\u88ab\u6e05\u7684\u78c1\u789f\u8def\u5f91\u3002\u8def\u5f91\u7528 file:\/\/ \u662f\u6b63\u78ba\u7684\uff0c\u78c1\u789f\u4e5f\u4e0d\u4e00\u5b9a\u6bd4 CPU \u5feb\u3002\"\n      };\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) => {\n            if (opt.textContent.startsWith(selected + \".\")) opt.classList.add(\"wrong\");\n          });\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\u7d50\u5408 LMCache \u52a0\u901f\u9577\u5c0d\u8a71\u8207 RAG \u5171 5 \u984c\uff0c\u9ede\u9078\u7b54\u6848\u5f8c\u6703\u7acb\u5373\u986f\u793a\u7d50\u679c 1. LMCache  [&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":[863,870,871,594,847],"class_list":["post-840","post","type-post","status-publish","format-standard","hentry","category-vllm-","tag-kv-cache","tag-lmcache","tag-prefix-caching","tag-rag","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\/840","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=840"}],"version-history":[{"count":0,"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/posts\/840\/revisions"}],"wp:attachment":[{"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/media?parent=840"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/categories?post=840"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/tags?post=840"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}