{"id":32470,"date":"2026-05-16T08:00:34","date_gmt":"2026-05-16T00:00:34","guid":{"rendered":"https:\/\/www.varidata.com\/uncategorized-zh-cn\/how-to-measure-real-llm-server-throughput\/"},"modified":"2026-05-15T15:21:46","modified_gmt":"2026-05-15T07:21:46","slug":"how-to-measure-real-llm-server-throughput","status":"publish","type":"post","link":"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-measure-real-llm-server-throughput\/","title":{"rendered":"\u5982\u4f55\u8861\u91cf LLM \u670d\u52a1\u5668\u7684\u771f\u5b9e\u541e\u5410\u91cf"},"content":{"rendered":"\n<p>\u5982\u679c\u4f60\u5728\u73b0\u4ee3\u670d\u52a1\u5668\u79df\u7528\u57fa\u7840\u8bbe\u65bd\u4e0a\u8fd0\u884c\u63a8\u7406\u5de5\u4f5c\u8d1f\u8f7d\uff0c\u90a3\u4e48<a href=\"https:\/\/www.varidata.com\/zh-cn\/blog-en\/llm-fine-tuning-gpu-hours-explained\/\" target=\"_blank\">LLM\u670d\u52a1\u5668<\/a>\u541e\u5410\u91cf\u8fd9\u4e2a\u8bcd\u5c31\u4e0d\u8be5\u53ea\u662f\u4f9b\u5e94\u5546\u5e7b\u706f\u7247\u4e0a\u7684\u8bf4\u6cd5\uff0c\u6216\u5b9e\u9a8c\u5ba4\u6f14\u793a\u4e2d\u7684\u6570\u5b57\u3002\u771f\u5b9e\u5bb9\u91cf\u53d6\u51b3\u4e8e\u6392\u961f\u884c\u4e3a\u3001token \u6d41\u52a8\u3001\u8bf7\u6c42\u7c7b\u578b\u7ec4\u5408\uff0c\u4ee5\u53ca\u7cfb\u7edf\u5728\u8d1f\u8f7d\u4e0b\u7684\u5b9e\u9645\u8868\u73b0\u3002\u5bf9\u4e8e\u5728<a href=\"https:\/\/www.varidata.com\/zh-cn\/server\/los-angeles\/\" target=\"_blank\">\u7f8e\u56fd\u670d\u52a1\u5668\u73af\u5883<\/a>\u4e2d\u6784\u5efa\u7cfb\u7edf\u7684\u6280\u672f\u56e2\u961f\u6765\u8bf4\uff0c\u76ee\u6807\u5176\u5b9e\u5f88\u76f4\u63a5\uff1a\u8861\u91cf\u8fd9\u5957\u6280\u672f\u6808\u5728\u5e76\u53d1\u7528\u6237\u3001\u957f\u63d0\u793a\u8bcd\u548c\u6301\u7eed\u6d41\u91cf\u540c\u65f6\u51fa\u73b0\u65f6\uff0c\u7a76\u7adf\u80fd\u771f\u6b63\u4ea4\u4ed8\u4ec0\u4e48\u3002<\/p>\n<p>\u5f88\u591a\u5173\u4e8e\u541e\u5410\u91cf\u7684\u8ba8\u8bba\u4e4b\u6240\u4ee5\u5931\u771f\uff0c\u662f\u56e0\u4e3a\u5b83\u4eec\u628a\u8bed\u8a00\u6a21\u578b\u670d\u52a1\u5f53\u6210\u666e\u901a Web \u6d41\u91cf\u6765\u770b\u5f85\u3002\u8fd9\u79cd\u7b80\u5316\u5f88\u5feb\u5c31\u4f1a\u5931\u6548\u3002\u751f\u6210\u5f0f\u5de5\u4f5c\u8d1f\u8f7d\u81f3\u5c11\u5305\u542b\u4e24\u4e2a\u4e0d\u540c\u9636\u6bb5\uff1a\u63d0\u793a\u8bcd\u5904\u7406\u548c token \u751f\u6210\u3002\u5b98\u65b9\u57fa\u51c6\u6d4b\u8bd5\u8d44\u6599\u901a\u5e38\u5c06\u5176\u79f0\u4e3a prefill \u548c decode\uff0c\u800c\u5b83\u4eec\u5bf9\u5ef6\u8fdf\u7684\u5f71\u54cd\u65b9\u5f0f\u5b8c\u5168\u4e0d\u540c\u3002\u9996 token \u65f6\u95f4\u4e0d\u4ec5\u5305\u542b\u6392\u961f\u3001\u63d0\u793a\u8bcd\u5904\u7406\uff0c\u8fd8\u5305\u62ec\u7f51\u7edc\u5f00\u9500\uff1b\u800c\u8f93\u51fa token \u541e\u5410\u91cf\u5219\u53cd\u6620\u6d41\u5f0f\u8f93\u51fa\u5f00\u59cb\u540e\u7cfb\u7edf\u6301\u7eed\u751f\u6210\u7684\u6548\u7387\u3002\u4e0e\u5355\u7eaf\u7684\u8bf7\u6c42\u6570\u76f8\u6bd4\uff0c\u8fd9\u4e9b\u6307\u6807\u91cd\u8981\u5f97\u591a\u3002<\/p>\n<h2><strong>\u4e3a\u4ec0\u4e48\u201c\u771f\u5b9e\u541e\u5410\u91cf\u201d\u4e0d\u540c\u4e8e\u7406\u8bba\u541e\u5410\u91cf<\/strong><\/h2>\n<p>\u7406\u8bba\u6027\u80fd\u603b\u662f\u5f88\u6574\u9f50\uff0c\u771f\u5b9e\u6027\u80fd\u5374\u5f80\u5f80\u5145\u6ee1\u566a\u58f0\u3002\u67d0\u53f0\u670d\u52a1\u5668\u5728\u7eb8\u9762\u53c2\u6570\u4e0a\u53ef\u80fd\u76f8\u5f53\u6f02\u4eae\uff0c\u4f46\u5f53\u63d0\u793a\u8bcd\u957f\u5ea6\u589e\u52a0\u3001\u5185\u5b58\u538b\u529b\u4e0a\u5347\uff0c\u6216\u591a\u4e2a\u4f1a\u8bdd\u540c\u65f6\u8fdb\u5165 prefill \u9636\u6bb5\u65f6\uff0c\u5b83\u7684\u8868\u73b0\u53ef\u80fd\u4f1a\u5b8c\u5168\u4e0d\u540c\u3002\u5b98\u65b9\u57fa\u51c6\u6d4b\u8bd5\u6307\u5357\u6307\u51fa\uff0c\u66f4\u957f\u7684\u8f93\u5165\u5e8f\u5217\u4f1a\u63d0\u9ad8 prefill \u6210\u672c\uff0c\u5e76\u53ef\u80fd\u62c9\u957f\u9996 token \u65f6\u95f4\uff1b\u800c\u66f4\u957f\u7684\u8f93\u51fa\u5219\u4f1a\u7ed9 decode \u548c token \u95f4\u5ef6\u8fdf\u5e26\u6765\u66f4\u5927\u538b\u529b\u3002\u6362\u53e5\u8bdd\u8bf4\uff0c\u4e00\u4e2a\u57fa\u51c6\u6570\u5b57\u6839\u672c\u65e0\u6cd5\u6982\u62ec\u6574\u4e2a LLM \u670d\u52a1\u3002<\/p>\n<p>\u56e0\u6b64\uff0c\u5de5\u7a0b\u5e08\u4e0d\u8be5\u518d\u95ee\u201c\u541e\u5410\u91cf\u662f\u591a\u5c11\u201d\uff0c\u800c\u5e94\u8be5\u63d0\u51fa\u66f4\u6709\u4ef7\u503c\u7684\u95ee\u9898\uff1a<\/p>\n<ul>\n<li>\u5728\u4f4e\u3001\u4e2d\u3001\u9ad8\u5e76\u53d1\u4e0b\uff0ctoken \u901f\u7387\u5206\u522b\u662f\u591a\u5c11\uff1f<\/li>\n<li>\u5728\u6392\u961f\u538b\u529b\u4e0b\uff0c\u9996 token \u65f6\u95f4\u7684\u7a33\u5b9a\u6027\u5982\u4f55\uff1f<\/li>\n<li>\u5ef6\u8fdf\u4ece\u54ea\u4e2a\u70b9\u5f00\u59cb\u4e0a\u5347\u5f97\u6bd4\u541e\u5410\u91cf\u66f4\u5feb\uff1f<\/li>\n<li>\u63d0\u793a\u8bcd\u957f\u5ea6\u4f1a\u5982\u4f55\u6539\u53d8\u6574\u4f53\u66f2\u7ebf\u5f62\u6001\uff1f<\/li>\n<li>\u7cfb\u7edf\u5728\u6301\u7eed\u8d1f\u8f7d\u4e0b\u662f\u5426\u80fd\u591f\u5e73\u6ed1\u9000\u5316\uff0c\u800c\u4e0d\u662f\u7a81\u7136\u5931\u7a33\uff1f<\/li>\n<\/ul>\n<p>\u8fd9\u4e9b\u95ee\u9898\u63ed\u793a\u7684\u662f\u5b9e\u9645\u8fd0\u884c\u72b6\u6001\uff0c\u800c\u4e0d\u662f\u5ba3\u4f20\u6750\u6599\u91cc\u7684\u7406\u60f3\u5316\u7b97\u672f\u3002<\/p>\n<h2><strong>\u4f60\u771f\u6b63\u5e94\u8be5\u8ddf\u8e2a\u7684\u6838\u5fc3\u6307\u6807<\/strong><\/h2>\n<p>\u5728\u5199\u4efb\u4f55\u57fa\u51c6\u6d4b\u8bd5\u811a\u672c\u4e4b\u524d\uff0c\u5148\u628a\u6307\u6807\u5b9a\u4e49\u6e05\u695a\u3002\u5982\u679c\u8861\u91cf\u541e\u5410\u91cf\u7684\u8bed\u8a00\u672c\u8eab\u662f\u6a21\u7cca\u7684\uff0c\u90a3\u4e48\u6700\u7ec8\u7ed3\u679c\u4e5f\u4e00\u5b9a\u662f\u6a21\u7cca\u7684\u3002<\/p>\n<ol>\n<li><strong>\u9996 token \u65f6\u95f4\uff08TTFT, Time to First Token\uff09\uff1a<\/strong>\u4ece\u53d1\u9001\u8bf7\u6c42\u5230\u6536\u5230\u7b2c\u4e00\u4e2a\u751f\u6210 token \u4e4b\u95f4\u7684\u5ef6\u8fdf\u3002\u8fd9\u662f\u4ea4\u4e92\u5f0f\u573a\u666f\u4e2d\u6700\u76f4\u89c2\u7684\u54cd\u5e94\u6307\u6807\u3002\u5b98\u65b9\u6587\u6863\u5c06\u5176\u89c6\u4e3a\u9762\u5411\u7528\u6237\u7684\u54cd\u5e94\u6027\u6307\u6807\uff0c\u5e76\u6307\u51fa\u5176\u4e2d\u5305\u542b\u7f51\u7edc\u5ef6\u8fdf\u3001\u6392\u961f\u548c\u63d0\u793a\u8bcd\u5904\u7406\u3002<\/li>\n<li><strong>Token \u95f4\u5ef6\u8fdf\uff08Inter-Token Latency\uff09\uff1a<\/strong>\u6d41\u5f0f\u8f93\u51fa\u5f00\u59cb\u540e\uff0c\u76f8\u90bb\u751f\u6210 token \u4e4b\u95f4\u7684\u65f6\u95f4\u95f4\u9694\u3002\u8fd9\u4e2a\u6307\u6807\u80fd\u591f\u63ed\u793a decode \u9636\u6bb5\u7684\u8868\u73b0\uff0c\u5e76\u5e2e\u52a9\u5224\u65ad\u8f93\u51fa\u662f\u987a\u6ed1\u8fd8\u662f\u5361\u987f\u3002<\/li>\n<li><strong>\u8f93\u51fa token \u541e\u5410\u91cf\uff08Output Token Throughput\uff09\uff1a<\/strong>\u6574\u4e2a\u7cfb\u7edf\u6bcf\u79d2\u751f\u6210\u7684 token \u6570\u91cf\u3002\u8fd9\u662f\u8861\u91cf\u751f\u6210\u5f0f\u670d\u52a1\u541e\u5410\u80fd\u529b\u6700\u76f4\u63a5\u7684\u6307\u6807\u3002<\/li>\n<li><strong>\u8bf7\u6c42\u541e\u5410\u91cf\uff08Request Throughput\uff09\uff1a<\/strong>\u6bcf\u79d2\u5b8c\u6210\u7684\u8bf7\u6c42\u6570\u3002\u5b83\u5f53\u7136\u6709\u7528\uff0c\u4f46\u53ea\u6709\u5728\u8f93\u5165\u548c\u8f93\u51fa\u957f\u5ea6\u88ab\u4e25\u683c\u63a7\u5236\u65f6\u624d\u771f\u6b63\u6709\u53c2\u8003\u610f\u4e49\u3002<\/li>\n<li><strong>\u5c3e\u90e8\u5ef6\u8fdf\uff08Tail Latency\uff09\uff1a<\/strong>p95 \u6216 p99 \u5f80\u5f80\u6bd4\u4e00\u4e2a\u597d\u770b\u7684\u5e73\u5747\u503c\u66f4\u91cd\u8981\uff0c\u56e0\u4e3a\u751f\u4ea7\u73af\u5883\u4e2d\u7684\u771f\u5b9e\u95ee\u9898\u901a\u5e38\u90fd\u51fa\u73b0\u5728\u5c3e\u90e8\u3002<\/li>\n<li><strong>\u5931\u8d25\u7387\uff08Failure Rate\uff09\uff1a<\/strong>\u8d85\u65f6\u3001\u7a7a\u54cd\u5e94\u3001\u91cd\u8bd5\u4ee5\u53ca\u4e0e\u5185\u5b58\u76f8\u5173\u7684\u5d29\u6e83\uff0c\u90fd\u5e94\u8be5\u88ab\u7eb3\u5165\u6700\u7ec8\u62a5\u544a\u3002<\/li>\n<\/ol>\n<p>\u5bf9\u4e8e\u6280\u672f\u8bfb\u8005\u6765\u8bf4\uff0c\u6700\u5173\u952e\u7684\u539f\u5219\u662f\uff1a\u4e0d\u8981\u5728\u6ca1\u6709\u5ef6\u8fdf\u6570\u636e\u7684\u60c5\u51b5\u4e0b\u8c08\u541e\u5410\u91cf\uff0c\u4e5f\u4e0d\u8981\u5728\u4e0d\u8bf4\u660e\u8d1f\u8f7d\u5f62\u6001\u7684\u524d\u63d0\u4e0b\u8c08\u5ef6\u8fdf\u3002<\/p>\n<h2><strong>\u5728\u5f00\u59cb\u6d4b\u8bd5\u4e4b\u524d\uff0c\u5148\u7406\u89e3 prefill \u548c decode<\/strong><\/h2>\n<p>\u5927\u591a\u6570\u57fa\u51c6\u6d4b\u8bd5\u9519\u8bef\uff0c\u6839\u6e90\u90fd\u5728\u4e8e\u5ffd\u89c6\u4e86\u9636\u6bb5\u884c\u4e3a\u3002Prefill \u662f\u6a21\u578b\u5904\u7406\u8f93\u5165\u4e0a\u4e0b\u6587\u7684\u9636\u6bb5\uff0cdecode \u5219\u662f\u6a21\u578b\u9010\u6b65\u8f93\u51fa token \u7684\u9636\u6bb5\u3002\u591a\u4efd\u5b98\u65b9\u8d44\u6599\u90fd\u5f3a\u8c03\uff0c\u8fd9\u4e24\u4e2a\u9636\u6bb5\u5bf9\u7cfb\u7edf\u7684\u538b\u529b\u6765\u6e90\u5e76\u4e0d\u76f8\u540c\uff0c\u800c\u4e14 prefill \u9636\u6bb5\u7684\u5e76\u53d1\u60c5\u51b5\u4f1a\u663e\u8457\u5f71\u54cd\u6392\u961f\u6df1\u5ea6\u4e0e\u9996 token \u5ef6\u8fdf\u3002<\/p>\n<p>\u8fd9\u4e4b\u6240\u4ee5\u91cd\u8981\uff0c\u662f\u56e0\u4e3a\u201c\u77ed\u63d0\u793a\u8bcd + \u957f\u8f93\u51fa\u201d\u548c\u201c\u957f\u63d0\u793a\u8bcd + \u77ed\u8f93\u51fa\u201d\u53ef\u80fd\u5448\u73b0\u5b8c\u5168\u4e0d\u540c\u7684\u7cfb\u7edf\u74f6\u9888\u3002\u5982\u679c\u4f60\u7684\u5de5\u4f5c\u8d1f\u8f7d\u6d89\u53ca\u68c0\u7d22\u3001\u4ee3\u7801\u4e0a\u4e0b\u6587\u6216\u591a\u8f6e\u8bb0\u5fc6\uff0c\u90a3\u4e48 prefill \u5f88\u53ef\u80fd\u4e3b\u5bfc\u7528\u6237\u4f53\u9a8c\uff1b\u5982\u679c\u4f60\u7684\u5de5\u4f5c\u8d1f\u8f7d\u9700\u8981\u6301\u7eed\u751f\u6210\u957f\u6587\u672c\uff0c\u90a3\u4e48 decode \u6548\u7387\u5c31\u53ef\u80fd\u6210\u4e3a\u771f\u6b63\u7684\u4e0a\u9650\u3002<\/p>\n<p>\u4e00\u4e2a\u5b9e\u7528\u7684\u601d\u7ef4\u6a21\u578b\u53ef\u4ee5\u8fd9\u6837\u7406\u89e3\uff1a<\/p>\n<ul>\n<li><strong>Prefill \u4e3b\u5bfc\u578b\u8d1f\u8f7d\uff1a<\/strong>\u5bf9\u63d0\u793a\u8bcd\u957f\u5ea6\u3001\u6392\u961f\u4e0e\u51c6\u5165\u63a7\u5236\u66f4\u654f\u611f<\/li>\n<li><strong>Decode \u4e3b\u5bfc\u578b\u8d1f\u8f7d\uff1a<\/strong>\u5bf9 token \u8c03\u5ea6\u3001\u5185\u5b58\u5e26\u5bbd\u548c\u6279\u5904\u7406\u6548\u7387\u66f4\u654f\u611f<\/li>\n<li><strong>\u6df7\u5408\u578b\u8d1f\u8f7d\uff1a<\/strong>\u4e24\u8005\u90fd\u654f\u611f\uff0c\u800c\u8fd9\u5176\u5b9e\u624d\u662f\u5f88\u591a\u751f\u4ea7\u7cfb\u7edf\u7684\u771f\u5b9e\u72b6\u6001<\/li>\n<\/ul>\n<h2><strong>\u5982\u4f55\u8bbe\u8ba1\u4e00\u4e2a\u8d34\u8fd1\u751f\u4ea7\u73af\u5883\u7684\u57fa\u51c6\u6d4b\u8bd5<\/strong><\/h2>\n<p>\u53ea\u6709\u5f53\u57fa\u51c6\u6d4b\u8bd5\u63a5\u8fd1\u4f60\u8ba1\u5212\u8fd0\u884c\u7684\u771f\u5b9e\u670d\u52a1\u65f6\uff0c\u5b83\u624d\u6709\u4ef7\u503c\u3002\u8fd9\u5e76\u4e0d\u610f\u5473\u7740\u8981\u590d\u5236\u6bcf\u4e00\u4e2a\u8fb9\u7f18\u573a\u666f\uff0c\u800c\u662f\u8981\u63a7\u5236\u90a3\u4e9b\u771f\u6b63\u5f71\u54cd\u7cfb\u7edf\u884c\u4e3a\u7684\u5173\u952e\u53d8\u91cf\u3002<\/p>\n<ol>\n<li><strong>\u56fa\u5b9a\u6a21\u578b\u4e0e\u8fd0\u884c\u8def\u5f84\u3002<\/strong>\u9664\u975e\u67d0\u9879\u53d8\u66f4\u672c\u8eab\u5c31\u662f\u5b9e\u9a8c\u76ee\u6807\uff0c\u5426\u5219\u4e0d\u8981\u5728\u91cf\u5316\u65b9\u5f0f\u3001\u8c03\u5ea6\u53c2\u6570\u6216\u5e76\u884c\u5e03\u5c40\u53d8\u5316\u540e\u6a2a\u5411\u6bd4\u8f83\u7ed3\u679c\u3002<\/li>\n<li><strong>\u56fa\u5b9a\u63d0\u793a\u8bcd\u4e0e\u8f93\u51fa\u8303\u56f4\u3002<\/strong>\u6ca1\u6709 token \u6570\u91cf\u7ea6\u675f\u7684\u541e\u5410\u91cf\u6570\u636e\uff0c\u51e0\u4e4e\u6ca1\u6709\u89e3\u91ca\u4ef7\u503c\u3002<\/li>\n<li><strong>\u5148\u5b8c\u6210\u9884\u70ed\u3002<\/strong>\u9996\u6b21\u8fd0\u884c\u7684\u989d\u5916\u5f00\u9500\u53ef\u80fd\u6765\u81ea\u60f0\u6027\u521d\u59cb\u5316\u3001\u56fe\u6355\u83b7\u6216\u7f13\u5b58\u586b\u5145\uff0c\u8fd9\u4f1a\u6c61\u67d3\u7ed3\u679c\u3002<\/li>\n<li><strong>\u5c06\u5355\u7528\u6237\u6d4b\u8bd5\u548c\u5e76\u53d1\u6d4b\u8bd5\u5206\u5f00\u3002<\/strong>\u524d\u8005\u7528\u4e8e\u89c2\u5bdf\u57fa\u7840\u54cd\u5e94\uff0c\u540e\u8005\u7528\u4e8e\u8bc4\u4f30\u670d\u52a1\u5bb9\u91cf\u3002<\/li>\n<li><strong>\u6d4b\u8bd5\u65f6\u95f4\u8981\u8db3\u591f\u957f\u3002<\/strong>\u8fc7\u77ed\u7684\u7a81\u53d1\u6d4b\u8bd5\u4f1a\u63a9\u76d6\u6e29\u5ea6\u3001\u6392\u961f\u6216\u5185\u5b58\u5c42\u9762\u7684\u95ee\u9898\u3002<\/li>\n<li><strong>\u91c7\u96c6\u7cfb\u7edf\u9065\u6d4b\u6570\u636e\u3002<\/strong>GPU \u5229\u7528\u7387\u3001\u5185\u5b58\u5360\u7528\u3001CPU \u538b\u529b\u548c\u7f51\u7edc\u8868\u73b0\u4e0d\u662f\u53ef\u6709\u53ef\u65e0\u7684\u9644\u6ce8\uff0c\u5b83\u4eec\u662f\u89e3\u91ca\u7ed3\u679c\u7684\u5173\u952e\u5c42\u3002<\/li>\n<\/ol>\n<p>\u5b98\u65b9\u7684\u751f\u6210\u5f0f\u7cfb\u7edf\u57fa\u51c6\u6d4b\u8bd5\u5de5\u5177\u901a\u5e38\u652f\u6301\u5408\u6210\u8f93\u5165\u6216\u6570\u636e\u96c6\u9a71\u52a8\u8f93\u5165\u3001\u57fa\u4e8e\u5e76\u53d1\u7684\u538b\u6d4b\u3001\u57fa\u4e8e\u8bf7\u6c42\u901f\u7387\u7684\u538b\u6d4b\uff0c\u4ee5\u53ca\u5bfc\u51fa\u65e5\u5fd7\u4f9b\u540e\u7eed\u5206\u6790\u3002\u8fd9\u4f7f\u5f97\u5de5\u7a0b\u56e2\u961f\u53ef\u4ee5\u6784\u5efa\u53ef\u91cd\u590d\u7684\u6d4b\u8bd5\u6d41\u7a0b\uff0c\u800c\u4e0d\u662f\u53ea\u622a\u51e0\u5f20\u7ec8\u7aef\u622a\u56fe\u4e86\u4e8b\u3002<\/p>\n<h2><strong>\u4e00\u5957\u9002\u5408\u6781\u5ba2\u8fd0\u7ef4\u89c6\u89d2\u7684\u5b9e\u7528\u6d4b\u8bd5\u77e9\u9635<\/strong><\/h2>\n<p>\u5982\u679c\u4f60\u7684\u7f51\u7ad9\u9762\u5411\u7f8e\u56fd\u57fa\u7840\u8bbe\u65bd\u91c7\u8d2d\u8005\uff0c\u90a3\u4e48\u8bfb\u8005\u591a\u534a\u662f\u5728\u505a\u67b6\u6784\u9009\u62e9\uff0c\u800c\u4e0d\u662f\u9605\u8bfb\u7eaf\u7406\u8bba\u3002\u56e0\u6b64\uff0c\u7ed9\u4ed6\u4eec\u4e00\u5957\u80fd\u76f4\u63a5\u8dd1\u8d77\u6765\u7684\u6d4b\u8bd5\u77e9\u9635\uff0c\u8fdc\u6bd4\u62bd\u8c61\u89e3\u91ca\u66f4\u6709\u4ef7\u503c\u3002<\/p>\n<ul>\n<li><strong>\u573a\u666f A\uff1a<\/strong>\u77ed\u63d0\u793a\u8bcd\u3001\u77ed\u8f93\u51fa\u3001\u4f4e\u5e76\u53d1<\/li>\n<li><strong>\u573a\u666f B\uff1a<\/strong>\u77ed\u63d0\u793a\u8bcd\u3001\u957f\u8f93\u51fa\u3001\u4e2d\u7b49\u5e76\u53d1<\/li>\n<li><strong>\u573a\u666f C\uff1a<\/strong>\u957f\u63d0\u793a\u8bcd\u3001\u77ed\u8f93\u51fa\u3001\u4e2d\u7b49\u5e76\u53d1<\/li>\n<li><strong>\u573a\u666f D\uff1a<\/strong>\u957f\u63d0\u793a\u8bcd\u3001\u957f\u8f93\u51fa\u3001\u9010\u7ea7\u63d0\u5347\u5e76\u53d1<\/li>\n<li><strong>\u573a\u666f E\uff1a<\/strong>\u6df7\u5408\u63d0\u793a\u8bcd\u957f\u5ea6\uff0c\u5e76\u53e0\u52a0\u6301\u7eed\u80cc\u666f\u6d41\u91cf<\/li>\n<\/ul>\n<p>\u7136\u540e\u4ee5\u9636\u68af\u65b9\u5f0f\u9010\u6b65\u63d0\u9ad8\u5e76\u53d1\uff0c\u800c\u4e0d\u662f\u4e00\u5f00\u59cb\u5c31\u628a\u5e76\u53d1\u503c\u62c9\u5230\u6781\u9650\u3002\u8fd9\u6837\u505a\u53ef\u4ee5\u627e\u51fa\u66f2\u7ebf\u7684\u62d0\u70b9\uff0c\u4e5f\u5c31\u662f\u8f93\u51fa token \u541e\u5410\u91cf\u4e0d\u518d\u8db3\u4ee5\u62b5\u6d88\u5ef6\u8fdf\u4ee3\u4ef7\u7684\u90a3\u4e2a\u4f4d\u7f6e\u3002\u5b98\u65b9\u7684\u751f\u6210\u5f0f\u670d\u52a1\u57fa\u51c6\u6d4b\u8bd5\u5efa\u8bae\u4e5f\u5f3a\u8c03\uff0c\u5e94\u4ed4\u7ec6\u626b\u63cf\u5e76\u53d1\u5ea6\u4ee5\u53ca\u8f93\u5165\u8f93\u51fa\u957f\u5ea6\uff0c\u56e0\u4e3a\u8fd9\u4e9b\u53c2\u6570\u4f1a\u76f4\u63a5\u51b3\u5b9a\u6d4b\u8bd5\u7ed3\u679c\u662f\u5426\u6709\u5b9e\u9645\u610f\u4e49\u3002<\/p>\n<p>\u5bf9\u5f88\u591a\u56e2\u961f\u800c\u8a00\uff0c\u6700\u6709\u4ef7\u503c\u7684\u56fe\u8868\u5e76\u4e0d\u662f\u201c\u6700\u9ad8\u541e\u5410\u91cf\u201d\uff0c\u800c\u662f\u201c\u670d\u52a1\u4ece\u54cd\u5e94\u987a\u7545\u53d8\u5f97\u8fdf\u6ede\u7684\u8f6c\u6298\u70b9\u201d\u3002<\/p>\n<h2><strong>\u6d4b\u8bd5\u8fc7\u7a0b\u4e2d\u5e94\u8be5\u8bb0\u5f55\u4ec0\u4e48<\/strong><\/h2>\n<p>\u5355\u7eaf\u7684\u57fa\u51c6\u6d4b\u8bd5\u8f93\u51fa\u6c38\u8fdc\u4e0d\u591f\u3002\u4f60\u9700\u8981\u628a\u8bf7\u6c42\u4fa7\u89c6\u89d2\u548c\u7cfb\u7edf\u4fa7\u89c6\u89d2\u914d\u5bf9\u8d77\u6765\u770b\u3002<\/p>\n<ul>\n<li>\u8bf7\u6c42\u5f00\u59cb\u65f6\u95f4\u548c\u5b8c\u6210\u65f6\u95f4<\/li>\n<li>\u9996 token \u65f6\u95f4\u6233<\/li>\n<li>\u8f93\u5165 token \u6570\u548c\u8f93\u51fa token \u6570<\/li>\n<li>\u6bcf\u4e2a\u8bf7\u6c42\u7684\u72b6\u6001\uff0c\u5305\u62ec\u5931\u8d25\u6761\u4ef6<\/li>\n<li>\u6b63\u5728\u5904\u7406\u4e2d\u7684\u5e76\u53d1\u8bf7\u6c42\u6570<\/li>\n<li>GPU \u5229\u7528\u7387\u548c\u663e\u5b58\u5360\u7528<\/li>\n<li>\u4e3b\u673a\u4e0a\u7684 CPU \u4e0e RAM \u538b\u529b<\/li>\n<li>\u5982\u679c\u5ba2\u6237\u7aef\u4f4d\u4e8e\u8fdc\u7aef\uff0c\u8fd8\u8981\u8bb0\u5f55\u7f51\u7edc\u5ef6\u8fdf<\/li>\n<\/ul>\n<p>\u5f53\u6570\u5b57\u770b\u8d77\u6765\u4e0d\u5bf9\u52b2\u65f6\uff0c\u8fd9\u4e9b\u65e5\u5fd7\u80fd\u5e2e\u52a9\u4f60\u5224\u65ad\u95ee\u9898\u7a76\u7adf\u6765\u81ea\u6392\u961f\u5806\u79ef\u3001\u63d0\u793a\u8bcd\u81a8\u80c0\u3001decode \u53d8\u6162\uff0c\u8fd8\u662f\u4e3b\u673a\u5c42\u9762\u7684\u74f6\u9888\u3002\u6ca1\u6709\u8fd9\u4e9b\u4fe1\u606f\uff0c\u5de5\u7a0b\u5e08\u5f80\u5f80\u4f1a\u628a\u95ee\u9898\u5f52\u56e0\u5230\u9519\u8bef\u7684\u5c42\u3002<\/p>\n<h2><strong>\u90a3\u4e9b\u4f1a\u8ba9\u7ed3\u679c\u5931\u53bb\u610f\u4e49\u7684\u5e38\u89c1\u6d4b\u8bd5\u9677\u9631<\/strong><\/h2>\n<p>\u5728 LLM \u541e\u5410\u91cf\u6d4b\u8bd5\u4e2d\uff0c\u6709\u4e00\u4e9b\u9519\u8bef\u51e0\u4e4e\u53cd\u590d\u51fa\u73b0\u3002<\/p>\n<ol>\n<li><strong>\u53ea\u6d4b\u4e00\u4e2a\u5e76\u53d1\u503c\u3002<\/strong>\u8fd9\u6837\u6839\u672c\u770b\u4e0d\u5230\u7cfb\u7edf\u66f2\u7ebf\u7684\u5f62\u72b6\u3002<\/li>\n<li><strong>\u5ffd\u89c6\u63d0\u793a\u8bcd\u957f\u5ea6\u3002<\/strong>\u5fae\u5c0f\u7684\u5408\u6210\u63d0\u793a\u8bcd\u51e0\u4e4e\u65e0\u6cd5\u4ee3\u8868\u68c0\u7d22\u578b\u6d41\u91cf\u3002<\/li>\n<li><strong>\u53ea\u770b\u5e73\u5747\u5ef6\u8fdf\u3002<\/strong>\u7528\u6237\u771f\u6b63\u4f1a\u62b1\u6028\u7684\u662f\u5c3e\u90e8\u5ef6\u8fdf\u3002<\/li>\n<li><strong>\u628a\u51b7\u542f\u52a8\u548c\u70ed\u542f\u52a8\u6df7\u5728\u4e00\u8d77\u3002<\/strong>\u521d\u59cb\u5316\u5f00\u9500\u4f1a\u6c61\u67d3\u6837\u672c\u3002<\/li>\n<li><strong>\u5728\u4e0d\u540c token \u9884\u7b97\u4e0b\u6bd4\u8f83\u8bf7\u6c42\u541e\u5410\u91cf\u3002<\/strong>\u8fd9\u6839\u672c\u4e0d\u662f\u540c\u4e00\u7ef4\u5ea6\u7684\u5bf9\u6bd4\u3002<\/li>\n<li><strong>\u62a5\u544a tokens per second \u65f6\uff0c\u4e0d\u8bf4\u660e\u662f\u7cfb\u7edf\u603b\u541e\u5410\u8fd8\u662f\u5355\u7528\u6237\u541e\u5410\u3002<\/strong>\u5b98\u65b9\u8d44\u6599\u4e4b\u6240\u4ee5\u533a\u5206\u8fd9\u4e24\u79cd\u89c6\u89d2\uff0c\u662f\u6709\u5145\u5206\u539f\u56e0\u7684\u3002<\/li>\n<li><strong>\u8ba9\u5ba2\u6237\u7aef\u5148\u6210\u4e3a\u74f6\u9888\u3002<\/strong>\u6027\u80fd\u4e0d\u8db3\u7684\u538b\u6d4b\u7aef\u4f1a\u4f2a\u9020\u670d\u52a1\u5668\u4e0a\u9650\u3002<\/li>\n<\/ol>\n<p>\u8fd8\u6709\u4e00\u4e2a\u5bb9\u6613\u88ab\u5ffd\u7565\u7684\u9677\u9631\u5c24\u5176\u503c\u5f97\u5355\u72ec\u63d0\u51fa\uff1a\u534f\u8c03\u9057\u6f0f\u3002\u5173\u4e8e prefill \u5e76\u53d1\u7684\u5b98\u65b9\u6587\u6863\u63d0\u5230\uff0c\u5982\u679c\u57fa\u51c6\u6d4b\u8bd5\u5de5\u5177\u5728\u7b49\u5f85\u7a7a\u95f2\u69fd\u4f4d\u65f6\u81ea\u52a8\u653e\u6162\u8282\u594f\uff0c\u90a3\u4e48\u6574\u4e2a\u6d4b\u8bd5\u53ef\u80fd\u4f1a\u65e0\u610f\u4e2d\u628a\u81ea\u8eab\u8282\u6d41\u5230\u4e0e\u670d\u52a1\u5668\u5bb9\u91cf\u4e00\u81f4\u3002\u8fd9\u4f1a\u8ba9\u5ef6\u8fdf\u8868\u73b0\u770b\u8d77\u6765\u6bd4\u771f\u5b9e\u5f00\u653e\u6d41\u91cf\u573a\u666f\u4e0b\u66f4\u597d\u3002<\/p>\n<h2><strong>\u5982\u4f55\u50cf\u8fd0\u7ef4\u5de5\u7a0b\u5e08\u4e00\u6837\u89e3\u8bfb\u7ed3\u679c<\/strong><\/h2>\n<p>\u5f53\u6d4b\u8bd5\u7ed3\u675f\u540e\uff0c\u4e0d\u8981\u6025\u7740\u628a\u6700\u5927\u7684\u6570\u5b57\u5f53\u6210\u8d62\u5bb6\u3002\u771f\u6b63\u6709\u4ef7\u503c\u7684\u89e3\u8bfb\uff0c\u6c38\u8fdc\u662f\u5728\u5206\u6790\u6743\u8861\u5173\u7cfb\u3002<\/p>\n<ul>\n<li>\u5982\u679c TTFT \u4e00\u76f4\u8f83\u4f4e\uff0c\u4f46 token \u6d41\u53d8\u5f97\u65ad\u65ad\u7eed\u7eed\uff0c\u90a3\u4e48\u9650\u5236\u56e0\u7d20\u5f88\u53ef\u80fd\u51fa\u5728 decode \u9636\u6bb5\u3002<\/li>\n<li>\u5982\u679c\u957f\u63d0\u793a\u8bcd\u8ba9 TTFT \u6025\u5267\u4e0a\u5347\uff0c\u90a3\u4e48\u95ee\u9898\u53ef\u80fd\u6765\u81ea prefill \u6216\u6392\u961f\u51c6\u5165\u3002<\/li>\n<li>\u5982\u679c\u541e\u5410\u91cf\u4e0a\u5347\u4e86\uff0c\u4f46 p99 \u5ef6\u8fdf\u4e5f\u66b4\u6da8\uff0c\u90a3\u4e48\u4f60\u5f88\u53ef\u80fd\u5df2\u7ecf\u8d85\u51fa\u4e86\u53ef\u7528\u8fd0\u884c\u533a\u95f4\u3002<\/li>\n<li>\u5982\u679c\u7ed3\u679c\u5f88\u5dee\u65f6 GPU \u5229\u7528\u7387\u5374\u5e76\u4e0d\u9ad8\uff0c\u90a3\u4e48\u74f6\u9888\u53ef\u80fd\u5728\u8c03\u5ea6\u3001CPU \u5904\u7406\u6d41\u7a0b\u6216\u5ba2\u6237\u7aef\u8def\u5f84\u4e0a\u3002<\/li>\n<li>\u5982\u679c\u5931\u8d25\u53ea\u5728\u6301\u7eed\u8d1f\u8f7d\u4e0b\u51fa\u73b0\uff0c\u90a3\u4e48\u4f60\u7684\u6808\u66f4\u53ef\u80fd\u662f\u7a33\u5b9a\u6027\u51fa\u4e86\u95ee\u9898\uff0c\u800c\u4e0d\u662f\u7b97\u529b\u672c\u8eab\u4e0d\u591f\u3002<\/li>\n<\/ul>\n<p>\u4f18\u79c0\u7684\u8fd0\u7ef4\u4eba\u5458\u5173\u6ce8\u7684\u4e0d\u662f\u201c\u6700\u5927\u6570\u5b57\u201d\uff0c\u800c\u662f\u201c\u9ad8\u6548\u533a\u95f4\u201d\u3002\u6240\u8c13\u9ad8\u6548\u533a\u95f4\uff0c\u5c31\u662f\u541e\u5410\u91cf\u8db3\u591f\u5f3a\u3001\u5ef6\u8fdf\u4ecd\u7136\u53ef\u9884\u6d4b\u3001\u5c3e\u90e8\u8868\u73b0\u4e5f\u4ecd\u7136\u7b26\u5408\u5e94\u7528\u8981\u6c42\u7684\u90a3\u4e00\u6bb5\u5de5\u4f5c\u533a\u95f4\u3002<\/p>\n<h2><strong>\u4e3a\u4ec0\u4e48\u8fd9\u5bf9\u7f8e\u56fd\u670d\u52a1\u5668\u79df\u7528\u548c\u670d\u52a1\u5668\u6258\u7ba1\u89c4\u5212\u5f88\u91cd\u8981<\/strong><\/h2>\n<p>\u5bf9\u4e8e\u6b63\u5728\u8bc4\u4f30<strong>\u670d\u52a1\u5668\u79df\u7528<\/strong>\u8fd8\u662f<strong>\u670d\u52a1\u5668\u6258\u7ba1<\/strong>\u7684\u56e2\u961f\u6765\u8bf4\uff0c\u541e\u5410\u91cf\u6d4b\u8bd5\u7edd\u4e0d\u53ea\u662f\u5b9e\u9a8c\u5ba4\u91cc\u7684\u7ec3\u4e60\u3002\u5b83\u4f1a\u76f4\u63a5\u5f71\u54cd\u5bb9\u91cf\u89c4\u5212\u3001\u8d44\u6e90\u5bc6\u5ea6\u8bbe\u8ba1\u548c\u670d\u52a1\u5f62\u6001\u9009\u62e9\u3002\u4e00\u4efd\u624e\u5b9e\u7684\u57fa\u51c6\u62a5\u544a\u80fd\u591f\u544a\u8bc9\u4f60\uff0c\u5f53\u524d\u7cfb\u7edf\u66f4\u9002\u5408\u56f4\u7ed5\u54cd\u5e94\u901f\u5ea6\u4f18\u5316\u3001\u56f4\u7ed5\u6279\u91cf\u6548\u7387\u4f18\u5316\uff0c\u8fd8\u662f\u56f4\u7ed5\u5185\u5b58\u4f59\u91cf\u4f18\u5316\u3002\u5b83\u4e5f\u80fd\u5e2e\u52a9\u4f60\u4f30\u7b97\uff0c\u5728\u670d\u52a1\u8d28\u91cf\u5f00\u59cb\u6076\u5316\u4e4b\u524d\uff0c\u4e00\u5957\u90e8\u7f72\u7a76\u7adf\u80fd\u652f\u6491\u591a\u5c11\u4f1a\u8bdd\u3002<\/p>\n<p>\u8fd9\u5728\u7f8e\u56fd\u670d\u52a1\u5668\u90e8\u7f72\u4e2d\u5c24\u5176\u91cd\u8981\uff0c\u56e0\u4e3a\u8fd9\u91cc\u5bf9\u5ef6\u8fdf\u7684\u9884\u671f\u901a\u5e38\u6bd4\u8f83\u4e25\u683c\uff0c\u800c\u751f\u4ea7\u6d41\u91cf\u53c8\u53ef\u80fd\u968f\u7740\u5730\u533a\u3001\u5de5\u4f5c\u8d1f\u8f7d\u6a21\u5f0f\u548c\u4e1a\u52a1\u65f6\u6bb5\u53d1\u751f\u53d8\u5316\u3002\u505a\u804a\u5929\u5f0f\u4ea4\u4e92\u7684\u56e2\u961f\uff0c\u4e5f\u8bb8\u66f4\u770b\u91cd\u9996 token \u901f\u5ea6\uff1b\u8fd0\u884c\u540e\u53f0\u6279\u91cf\u751f\u6210\u7684\u56e2\u961f\uff0c\u5219\u53ef\u80fd\u66f4\u5173\u6ce8\u603b\u4f53\u8f93\u51fa token \u541e\u5410\u91cf\u3002\u57fa\u51c6\u6d4b\u8bd5\u5e94\u8be5\u53cd\u6620\u8fd9\u79cd\u8fd0\u8425\u73b0\u5b9e\uff0c\u800c\u4e0d\u662f\u5f3a\u884c\u8ffd\u6c42\u67d0\u4e2a\u653e\u4e4b\u56db\u6d77\u800c\u7686\u51c6\u7684\u5355\u4e00\u76ee\u6807\u3002<\/p>\n<h2><strong>\u4e00\u5957\u53ef\u4ee5\u53cd\u590d\u590d\u7528\u7684\u7b80\u5355\u6d4b\u8bd5\u6d41\u7a0b<\/strong><\/h2>\n<ol>\n<li>\u8bb0\u5f55\u786c\u4ef6\u548c\u8f6f\u4ef6\u73af\u5883\u3002<\/li>\n<li>\u7528\u4e00\u7ec4\u56fa\u5b9a\u7684\u5c0f\u8bf7\u6c42\u5bf9\u670d\u52a1\u8fdb\u884c\u9884\u70ed\u3002<\/li>\n<li>\u5148\u8dd1\u57fa\u7ebf\u5355\u8bf7\u6c42\u6d4b\u8bd5\u3002<\/li>\n<li>\u9010\u6b65\u626b\u63cf\u5e76\u53d1\u5ea6\u3002<\/li>\n<li>\u7528\u66f4\u957f\u7684\u63d0\u793a\u8bcd\u91cd\u590d\u626b\u63cf\u3002<\/li>\n<li>\u518d\u7528\u66f4\u957f\u7684\u8f93\u51fa\u91cd\u590d\u4e00\u6b21\u3002<\/li>\n<li>\u6267\u884c\u6301\u7eed\u6027\u6d4b\u8bd5\uff0c\u89c2\u5bdf\u6f02\u79fb\u548c\u5931\u8d25\u60c5\u51b5\u3002<\/li>\n<li>\u5c06 TTFT\u3001token \u95f4\u5ef6\u8fdf\u3001\u5c3e\u90e8\u5ef6\u8fdf\u548c\u603b\u4f53 token \u541e\u5410\u91cf\u7ed3\u5408\u8d77\u6765\u5206\u6790\u3002<\/li>\n<li>\u6807\u8bb0\u51fa\u6700\u7b26\u5408\u76ee\u6807\u5e94\u7528\u7684\u8fd0\u884c\u533a\u95f4\u3002<\/li>\n<\/ol>\n<p>\u8fd9\u5957\u6d41\u7a0b\u8db3\u591f\u7b80\u5355\uff0c\u4fbf\u4e8e\u91cd\u590d\u6267\u884c\uff1b\u540c\u65f6\u4e5f\u8db3\u591f\u4e25\u683c\uff0c\u80fd\u591f\u4ea7\u51fa\u771f\u6b63\u6709\u5de5\u7a0b\u4ef7\u503c\u7684\u8bc1\u636e\u3002<\/p>\n<h2><strong>\u6700\u540e\u7684\u601d\u8003<\/strong><\/h2>\n<p><strong>LLM \u670d\u52a1\u5668\u541e\u5410\u91cf<\/strong>\u5e76\u4e0d\u662f\u4e00\u4e2a\u8425\u9500\u6570\u5b57\uff0c\u800c\u662f\u5728\u660e\u786e\u5de5\u4f5c\u8d1f\u8f7d\u4e0b\uff0cprefill\u3001decode\u3001\u6392\u961f\u3001\u5e76\u53d1\u548c\u7cfb\u7edf\u8fb9\u754c\u5171\u540c\u4f5c\u7528\u540e\u7684\u7ed3\u679c\u3002\u53ea\u8981\u4f60\u8ba4\u771f\u8861\u91cf\u8fd9\u4e9b\u52a8\u6001\u56e0\u7d20\uff0c\u5c31\u80fd\u5728\u67b6\u6784\u3001\u5bb9\u91cf\u548c\u7f8e\u56fd<strong>\u670d\u52a1\u5668\u79df\u7528<\/strong>\u7b56\u7565\u4e0a\u505a\u51fa\u66f4\u660e\u667a\u7684\u51b3\u5b9a\u3002\u5982\u679c\u8df3\u8fc7\u8fd9\u4e00\u6b65\u9aa4\uff0c\u57fa\u51c6\u6d4b\u8bd5\u5c31\u53ea\u5269\u4e0b\u8868\u6f14\u610f\u4e49\u3002\u5bf9\u4e8e\u6280\u672f\u56e2\u961f\u6765\u8bf4\uff0c\u771f\u6b63\u7684\u76ee\u6807\u4e0d\u662f\u8ffd\u9010\u6700\u54cd\u4eae\u7684\u6570\u5b57\uff0c\u800c\u662f\u627e\u5230\u4e00\u4e2a\u7a33\u5b9a\u8fd0\u884c\u70b9\uff1a\u5728\u90a3\u91cc\uff0c\u5ef6\u8fdf\u4ecd\u7136\u5408\u7406\uff0ctoken \u6d41\u4fdd\u6301\u987a\u6ed1\uff0c\u57fa\u7840\u8bbe\u65bd\u6548\u7387\u4e5f\u8db3\u591f\u9ad8\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u5982\u679c\u4f60\u5728\u73b0\u4ee3\u670d\u52a1\u5668\u79df\u7528\u57fa\u7840\u8bbe\u65bd\u4e0a\u8fd0\u884c\u63a8\u7406\u5de5\u4f5c\u8d1f\u8f7d\uff0c\u90a3\u4e48LLM\u670d\u52a1\u5668\u541e\u5410\u91cf\u8fd9\u4e2a\u8bcd\u5c31\u4e0d\u8be5\u53ea\u662f\u4f9b\u5e94\u5546\u5e7b\u706f\u7247\u4e0a\u7684\u8bf4\u6cd5\uff0c\u6216\u5b9e\u9a8c\u5ba4\u6f14\u793a\u4e2d\u7684\u6570\u5b57\u3002\u771f\u5b9e\u5bb9\u91cf\u53d6\u51b3\u4e8e\u6392\u961f\u884c\u4e3a\u3001token \u6d41\u52a8\u3001\u8bf7\u6c42\u7c7b\u578b\u7ec4\u5408\uff0c\u4ee5\u53ca\u7cfb\u7edf\u5728\u8d1f\u8f7d\u4e0b\u7684\u5b9e\u9645\u8868\u73b0\u3002\u5bf9\u4e8e\u5728\u7f8e\u56fd\u670d\u52a1\u5668\u73af\u5883\u4e2d\u6784\u5efa\u7cfb\u7edf\u7684\u6280\u672f\u56e2\u961f\u6765\u8bf4\uff0c\u76ee\u6807\u5176\u5b9e\u5f88\u76f4\u63a5\uff1a\u8861\u91cf\u8fd9\u5957\u6280\u672f\u6808\u5728\u5e76\u53d1\u7528\u6237\u3001\u957f\u63d0\u793a\u8bcd\u548c\u6301\u7eed\u6d41\u91cf\u540c\u65f6\u51fa\u73b0\u65f6\uff0c\u7a76\u7adf\u80fd\u771f\u6b63\u4ea4\u4ed8\u4ec0\u4e48\u3002 \u5f88\u591a\u5173\u4e8e\u541e\u5410\u91cf\u7684\u8ba8\u8bba\u4e4b\u6240\u4ee5\u5931\u771f\uff0c\u662f\u56e0\u4e3a\u5b83\u4eec\u628a\u8bed\u8a00\u6a21\u578b\u670d\u52a1\u5f53\u6210\u666e\u901a Web \u6d41\u91cf\u6765\u770b\u5f85\u3002\u8fd9\u79cd\u7b80\u5316\u5f88\u5feb\u5c31\u4f1a\u5931\u6548\u3002\u751f\u6210\u5f0f\u5de5\u4f5c\u8d1f\u8f7d\u81f3\u5c11\u5305\u542b\u4e24\u4e2a\u4e0d\u540c\u9636\u6bb5\uff1a\u63d0\u793a\u8bcd\u5904\u7406\u548c token \u751f\u6210\u3002\u5b98\u65b9\u57fa\u51c6\u6d4b\u8bd5\u8d44\u6599\u901a\u5e38\u5c06\u5176\u79f0\u4e3a prefill \u548c decode\uff0c\u800c\u5b83\u4eec\u5bf9\u5ef6\u8fdf\u7684\u5f71\u54cd\u65b9\u5f0f\u5b8c\u5168\u4e0d\u540c\u3002\u9996 token \u65f6\u95f4\u4e0d\u4ec5\u5305\u542b\u6392\u961f\u3001\u63d0\u793a\u8bcd\u5904\u7406\uff0c\u8fd8\u5305\u62ec\u7f51\u7edc\u5f00\u9500\uff1b\u800c\u8f93\u51fa token \u541e\u5410\u91cf\u5219\u53cd\u6620\u6d41\u5f0f\u8f93\u51fa\u5f00\u59cb\u540e\u7cfb\u7edf\u6301\u7eed\u751f\u6210\u7684\u6548\u7387\u3002\u4e0e\u5355\u7eaf\u7684\u8bf7\u6c42\u6570\u76f8\u6bd4\uff0c\u8fd9\u4e9b\u6307\u6807\u91cd\u8981\u5f97\u591a\u3002 \u4e3a\u4ec0\u4e48\u201c\u771f\u5b9e\u541e\u5410\u91cf\u201d\u4e0d\u540c\u4e8e\u7406\u8bba\u541e\u5410\u91cf \u7406\u8bba\u6027\u80fd\u603b\u662f\u5f88\u6574\u9f50\uff0c\u771f\u5b9e\u6027\u80fd\u5374\u5f80\u5f80\u5145\u6ee1\u566a\u58f0\u3002\u67d0\u53f0\u670d\u52a1\u5668\u5728\u7eb8\u9762\u53c2\u6570\u4e0a\u53ef\u80fd\u76f8\u5f53\u6f02\u4eae\uff0c\u4f46\u5f53\u63d0\u793a\u8bcd\u957f\u5ea6\u589e\u52a0\u3001\u5185\u5b58\u538b\u529b\u4e0a\u5347\uff0c\u6216\u591a\u4e2a\u4f1a\u8bdd\u540c\u65f6\u8fdb\u5165 prefill \u9636\u6bb5\u65f6\uff0c\u5b83\u7684\u8868\u73b0\u53ef\u80fd\u4f1a\u5b8c\u5168\u4e0d\u540c\u3002\u5b98\u65b9\u57fa\u51c6\u6d4b\u8bd5\u6307\u5357\u6307\u51fa\uff0c\u66f4\u957f\u7684\u8f93\u5165\u5e8f\u5217\u4f1a\u63d0\u9ad8 prefill \u6210\u672c\uff0c\u5e76\u53ef\u80fd\u62c9\u957f\u9996 token \u65f6\u95f4\uff1b\u800c\u66f4\u957f\u7684\u8f93\u51fa\u5219\u4f1a\u7ed9 decode \u548c token \u95f4\u5ef6\u8fdf\u5e26\u6765\u66f4\u5927\u538b\u529b\u3002\u6362\u53e5\u8bdd\u8bf4\uff0c\u4e00\u4e2a\u57fa\u51c6\u6570\u5b57\u6839\u672c\u65e0\u6cd5\u6982\u62ec\u6574\u4e2a LLM \u670d\u52a1\u3002 \u56e0\u6b64\uff0c\u5de5\u7a0b\u5e08\u4e0d\u8be5\u518d\u95ee\u201c\u541e\u5410\u91cf\u662f\u591a\u5c11\u201d\uff0c\u800c\u5e94\u8be5\u63d0\u51fa\u66f4\u6709\u4ef7\u503c\u7684\u95ee\u9898\uff1a \u5728\u4f4e\u3001\u4e2d\u3001\u9ad8\u5e76\u53d1\u4e0b\uff0ctoken \u901f\u7387\u5206\u522b\u662f\u591a\u5c11\uff1f \u5728\u6392\u961f\u538b\u529b\u4e0b\uff0c\u9996 token \u65f6\u95f4\u7684\u7a33\u5b9a\u6027\u5982\u4f55\uff1f \u5ef6\u8fdf\u4ece\u54ea\u4e2a\u70b9\u5f00\u59cb\u4e0a\u5347\u5f97\u6bd4\u541e\u5410\u91cf\u66f4\u5feb\uff1f \u63d0\u793a\u8bcd\u957f\u5ea6\u4f1a\u5982\u4f55\u6539\u53d8\u6574\u4f53\u66f2\u7ebf\u5f62\u6001\uff1f \u7cfb\u7edf\u5728\u6301\u7eed\u8d1f\u8f7d\u4e0b\u662f\u5426\u80fd\u591f\u5e73\u6ed1\u9000\u5316\uff0c\u800c\u4e0d\u662f\u7a81\u7136\u5931\u7a33\uff1f \u8fd9\u4e9b\u95ee\u9898\u63ed\u793a\u7684\u662f\u5b9e\u9645\u8fd0\u884c\u72b6\u6001\uff0c\u800c\u4e0d\u662f\u5ba3\u4f20\u6750\u6599\u91cc\u7684\u7406\u60f3\u5316\u7b97\u672f\u3002 \u4f60\u771f\u6b63\u5e94\u8be5\u8ddf\u8e2a\u7684\u6838\u5fc3\u6307\u6807 \u5728\u5199\u4efb\u4f55\u57fa\u51c6\u6d4b\u8bd5\u811a\u672c\u4e4b\u524d\uff0c\u5148\u628a\u6307\u6807\u5b9a\u4e49\u6e05\u695a\u3002\u5982\u679c\u8861\u91cf\u541e\u5410\u91cf\u7684\u8bed\u8a00\u672c\u8eab\u662f\u6a21\u7cca\u7684\uff0c\u90a3\u4e48\u6700\u7ec8\u7ed3\u679c\u4e5f\u4e00\u5b9a\u662f\u6a21\u7cca\u7684\u3002 \u9996 token \u65f6\u95f4\uff08TTFT, Time to First Token\uff09\uff1a\u4ece\u53d1\u9001\u8bf7\u6c42\u5230\u6536\u5230\u7b2c\u4e00\u4e2a\u751f\u6210 token \u4e4b\u95f4\u7684\u5ef6\u8fdf\u3002\u8fd9\u662f\u4ea4\u4e92\u5f0f\u573a\u666f\u4e2d\u6700\u76f4\u89c2\u7684\u54cd\u5e94\u6307\u6807\u3002\u5b98\u65b9\u6587\u6863\u5c06\u5176\u89c6\u4e3a\u9762\u5411\u7528\u6237\u7684\u54cd\u5e94\u6027\u6307\u6807\uff0c\u5e76\u6307\u51fa\u5176\u4e2d\u5305\u542b\u7f51\u7edc\u5ef6\u8fdf\u3001\u6392\u961f\u548c\u63d0\u793a\u8bcd\u5904\u7406\u3002 Token \u95f4\u5ef6\u8fdf\uff08Inter-Token Latency\uff09\uff1a\u6d41\u5f0f\u8f93\u51fa\u5f00\u59cb\u540e\uff0c\u76f8\u90bb\u751f\u6210 token \u4e4b\u95f4\u7684\u65f6\u95f4\u95f4\u9694\u3002\u8fd9\u4e2a\u6307\u6807\u80fd\u591f\u63ed\u793a decode [&#8230;]<\/p>\n<p><a class=\"btn btn-secondary understrap-read-more-link\" href=\"https:\/\/www.varidata.com\/zh-cn\/blog\/how-to-measure-real-llm-server-throughput\/\">Read More&#8230;<\/a><\/p>\n","protected":false},"author":12,"featured_media":32467,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[76],"tags":[],"class_list":["post-32470","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-blog"],"acf":[],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v26.3 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>\u5982\u4f55\u8861\u91cf LLM \u670d\u52a1\u5668\u7684\u771f\u5b9e\u541e\u5410\u91cf<\/title>\n<meta name=\"description\" content=\"\u4e86\u89e3\u5982\u4f55\u901a\u8fc7\u5b9e\u7528\u65b9\u6cd5\u5bf9 LLM \u670d\u52a1\u5668\u7684\u771f\u5b9e\u541e\u5410\u91cf\u8fdb\u884c\u57fa\u51c6\u6d4b\u8bd5\uff0c\u6db5\u76d6 TTFT\u3001\u5ef6\u8fdf\u3001\u5e76\u53d1\u4e0e token \u6d41\u8868\u73b0\uff0c\u9002\u7528\u4e8e\u7f8e\u56fd\u670d\u52a1\u5668\u79df\u7528\u57fa\u7840\u8bbe\u65bd\u3002\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u5982\u4f55\u8861\u91cf LLM \u670d\u52a1\u5668\u7684\u771f\u5b9e\u541e\u5410\u91cf\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/\" \/>\n<meta property=\"og:site_name\" content=\"Varidata Limited\" \/>\n<meta property=\"article:published_time\" content=\"2026-05-16T00:00:34+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.varidata.com\/wp-content\/uploads\/2026\/05\/Screenshot-2026-05-15-150435.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"615\" \/>\n\t<meta property=\"og:image:height\" content=\"346\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/#article\",\"isPartOf\":{\"@id\":\"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/\"},\"author\":\"Varidata\",\"headline\":\"\u5982\u4f55\u8861\u91cf LLM \u670d\u52a1\u5668\u7684\u771f\u5b9e\u541e\u5410\u91cf\",\"datePublished\":\"2026-05-16T00:00:34+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/\"},\"wordCount\":93,\"publisher\":{\"@id\":\"https:\/\/www.varidata.com\/zh-cn\/#organization\"},\"image\":{\"@id\":\"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/www.varidata.com\/wp-content\/uploads\/2026\/05\/Screenshot-2026-05-15-150435.jpg\",\"articleSection\":[\"Varidata \u5b98\u65b9\u535a\u5ba2\"],\"inLanguage\":\"zh-SC\"},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/\",\"url\":\"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/\",\"name\":\"\u5982\u4f55\u8861\u91cf LLM \u670d\u52a1\u5668\u7684\u771f\u5b9e\u541e\u5410\u91cf\",\"isPartOf\":{\"@id\":\"https:\/\/www.varidata.com\/zh-cn\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/#primaryimage\"},\"image\":{\"@id\":\"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/www.varidata.com\/wp-content\/uploads\/2026\/05\/Screenshot-2026-05-15-150435.jpg\",\"datePublished\":\"2026-05-16T00:00:34+00:00\",\"description\":\"\u4e86\u89e3\u5982\u4f55\u901a\u8fc7\u5b9e\u7528\u65b9\u6cd5\u5bf9 LLM \u670d\u52a1\u5668\u7684\u771f\u5b9e\u541e\u5410\u91cf\u8fdb\u884c\u57fa\u51c6\u6d4b\u8bd5\uff0c\u6db5\u76d6 TTFT\u3001\u5ef6\u8fdf\u3001\u5e76\u53d1\u4e0e token \u6d41\u8868\u73b0\uff0c\u9002\u7528\u4e8e\u7f8e\u56fd\u670d\u52a1\u5668\u79df\u7528\u57fa\u7840\u8bbe\u65bd\u3002\",\"breadcrumb\":{\"@id\":\"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/#breadcrumb\"},\"inLanguage\":\"zh-SC\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-SC\",\"@id\":\"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/#primaryimage\",\"url\":\"https:\/\/www.varidata.com\/wp-content\/uploads\/2026\/05\/Screenshot-2026-05-15-150435.jpg\",\"contentUrl\":\"https:\/\/www.varidata.com\/wp-content\/uploads\/2026\/05\/Screenshot-2026-05-15-150435.jpg\",\"width\":615,\"height\":346,\"caption\":\"Diagram of LLM server throughput testing\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/www.varidata.com\/zh-cn\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u5982\u4f55\u8861\u91cf LLM \u670d\u52a1\u5668\u7684\u771f\u5b9e\u541e\u5410\u91cf\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.varidata.com\/zh-cn\/#website\",\"url\":\"https:\/\/www.varidata.com\/zh-cn\/\",\"name\":\"Varidata Limited\",\"description\":\"\",\"publisher\":{\"@id\":\"https:\/\/www.varidata.com\/zh-cn\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.varidata.com\/zh-cn\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"zh-SC\"},{\"@type\":\"Organization\",\"@id\":\"https:\/\/www.varidata.com\/zh-cn\/#organization\",\"name\":\"Varidata\",\"url\":\"https:\/\/www.varidata.com\/zh-cn\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-SC\",\"@id\":\"https:\/\/www.varidata.com\/zh-cn\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/www.varidata.com\/wp-content\/uploads\/2021\/09\/varidata_logo_white_-748x480_hor_web-1.png\",\"contentUrl\":\"https:\/\/www.varidata.com\/wp-content\/uploads\/2021\/09\/varidata_logo_white_-748x480_hor_web-1.png\",\"width\":248,\"height\":94,\"caption\":\"Varidata\"},\"image\":{\"@id\":\"https:\/\/www.varidata.com\/zh-cn\/#\/schema\/logo\/image\/\"}},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.varidata.com\/zh-cn\/#\/schema\/person\/8fe12b0dd65fd2ba7ccf0dea7d14de6c\",\"name\":\"Amy Yan\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-SC\",\"@id\":\"https:\/\/www.varidata.com\/zh-cn\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/087e2e5704ab9b17ce8a5ebfda282b5f9d46ea62ff6c00f7b19ef4d6ce4e83aa?s=96&d=mm&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/087e2e5704ab9b17ce8a5ebfda282b5f9d46ea62ff6c00f7b19ef4d6ce4e83aa?s=96&d=mm&r=g\",\"caption\":\"Amy Yan\"}}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u5982\u4f55\u8861\u91cf LLM \u670d\u52a1\u5668\u7684\u771f\u5b9e\u541e\u5410\u91cf","description":"\u4e86\u89e3\u5982\u4f55\u901a\u8fc7\u5b9e\u7528\u65b9\u6cd5\u5bf9 LLM \u670d\u52a1\u5668\u7684\u771f\u5b9e\u541e\u5410\u91cf\u8fdb\u884c\u57fa\u51c6\u6d4b\u8bd5\uff0c\u6db5\u76d6 TTFT\u3001\u5ef6\u8fdf\u3001\u5e76\u53d1\u4e0e token \u6d41\u8868\u73b0\uff0c\u9002\u7528\u4e8e\u7f8e\u56fd\u670d\u52a1\u5668\u79df\u7528\u57fa\u7840\u8bbe\u65bd\u3002","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/","og_locale":"zh_CN","og_type":"article","og_title":"\u5982\u4f55\u8861\u91cf LLM \u670d\u52a1\u5668\u7684\u771f\u5b9e\u541e\u5410\u91cf","og_url":"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/","og_site_name":"Varidata Limited","article_published_time":"2026-05-16T00:00:34+00:00","og_image":[{"width":615,"height":346,"url":"https:\/\/www.varidata.com\/wp-content\/uploads\/2026\/05\/Screenshot-2026-05-15-150435.jpg","type":"image\/jpeg"}],"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/#article","isPartOf":{"@id":"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/"},"author":"Varidata","headline":"\u5982\u4f55\u8861\u91cf LLM \u670d\u52a1\u5668\u7684\u771f\u5b9e\u541e\u5410\u91cf","datePublished":"2026-05-16T00:00:34+00:00","mainEntityOfPage":{"@id":"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/"},"wordCount":93,"publisher":{"@id":"https:\/\/www.varidata.com\/zh-cn\/#organization"},"image":{"@id":"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/#primaryimage"},"thumbnailUrl":"https:\/\/www.varidata.com\/wp-content\/uploads\/2026\/05\/Screenshot-2026-05-15-150435.jpg","articleSection":["Varidata \u5b98\u65b9\u535a\u5ba2"],"inLanguage":"zh-SC"},{"@type":"WebPage","@id":"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/","url":"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/","name":"\u5982\u4f55\u8861\u91cf LLM \u670d\u52a1\u5668\u7684\u771f\u5b9e\u541e\u5410\u91cf","isPartOf":{"@id":"https:\/\/www.varidata.com\/zh-cn\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/#primaryimage"},"image":{"@id":"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/#primaryimage"},"thumbnailUrl":"https:\/\/www.varidata.com\/wp-content\/uploads\/2026\/05\/Screenshot-2026-05-15-150435.jpg","datePublished":"2026-05-16T00:00:34+00:00","description":"\u4e86\u89e3\u5982\u4f55\u901a\u8fc7\u5b9e\u7528\u65b9\u6cd5\u5bf9 LLM \u670d\u52a1\u5668\u7684\u771f\u5b9e\u541e\u5410\u91cf\u8fdb\u884c\u57fa\u51c6\u6d4b\u8bd5\uff0c\u6db5\u76d6 TTFT\u3001\u5ef6\u8fdf\u3001\u5e76\u53d1\u4e0e token \u6d41\u8868\u73b0\uff0c\u9002\u7528\u4e8e\u7f8e\u56fd\u670d\u52a1\u5668\u79df\u7528\u57fa\u7840\u8bbe\u65bd\u3002","breadcrumb":{"@id":"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/#breadcrumb"},"inLanguage":"zh-SC","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/"]}]},{"@type":"ImageObject","inLanguage":"zh-SC","@id":"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/#primaryimage","url":"https:\/\/www.varidata.com\/wp-content\/uploads\/2026\/05\/Screenshot-2026-05-15-150435.jpg","contentUrl":"https:\/\/www.varidata.com\/wp-content\/uploads\/2026\/05\/Screenshot-2026-05-15-150435.jpg","width":615,"height":346,"caption":"Diagram of LLM server throughput testing"},{"@type":"BreadcrumbList","@id":"https:\/\/www.varidata.com\/blog-en\/how-to-measure-real-llm-server-throughput\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.varidata.com\/zh-cn\/"},{"@type":"ListItem","position":2,"name":"\u5982\u4f55\u8861\u91cf LLM \u670d\u52a1\u5668\u7684\u771f\u5b9e\u541e\u5410\u91cf"}]},{"@type":"WebSite","@id":"https:\/\/www.varidata.com\/zh-cn\/#website","url":"https:\/\/www.varidata.com\/zh-cn\/","name":"Varidata Limited","description":"","publisher":{"@id":"https:\/\/www.varidata.com\/zh-cn\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.varidata.com\/zh-cn\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"zh-SC"},{"@type":"Organization","@id":"https:\/\/www.varidata.com\/zh-cn\/#organization","name":"Varidata","url":"https:\/\/www.varidata.com\/zh-cn\/","logo":{"@type":"ImageObject","inLanguage":"zh-SC","@id":"https:\/\/www.varidata.com\/zh-cn\/#\/schema\/logo\/image\/","url":"https:\/\/www.varidata.com\/wp-content\/uploads\/2021\/09\/varidata_logo_white_-748x480_hor_web-1.png","contentUrl":"https:\/\/www.varidata.com\/wp-content\/uploads\/2021\/09\/varidata_logo_white_-748x480_hor_web-1.png","width":248,"height":94,"caption":"Varidata"},"image":{"@id":"https:\/\/www.varidata.com\/zh-cn\/#\/schema\/logo\/image\/"}},{"@type":"Person","@id":"https:\/\/www.varidata.com\/zh-cn\/#\/schema\/person\/8fe12b0dd65fd2ba7ccf0dea7d14de6c","name":"Amy Yan","image":{"@type":"ImageObject","inLanguage":"zh-SC","@id":"https:\/\/www.varidata.com\/zh-cn\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/087e2e5704ab9b17ce8a5ebfda282b5f9d46ea62ff6c00f7b19ef4d6ce4e83aa?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/087e2e5704ab9b17ce8a5ebfda282b5f9d46ea62ff6c00f7b19ef4d6ce4e83aa?s=96&d=mm&r=g","caption":"Amy Yan"}}]}},"_links":{"self":[{"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/posts\/32470","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/users\/12"}],"replies":[{"embeddable":true,"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/comments?post=32470"}],"version-history":[{"count":3,"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/posts\/32470\/revisions"}],"predecessor-version":[{"id":32475,"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/posts\/32470\/revisions\/32475"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/media\/32467"}],"wp:attachment":[{"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/media?parent=32470"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/categories?post=32470"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.varidata.com\/zh-cn\/wp-json\/wp\/v2\/tags?post=32470"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}