{"id":40714,"date":"2025-12-22T22:06:09","date_gmt":"2025-12-22T14:06:09","guid":{"rendered":"https:\/\/17aitech.com\/?p=40714"},"modified":"2025-12-22T22:06:09","modified_gmt":"2025-12-22T14:06:09","slug":"%e8%ae%a1%e7%ae%97%e6%89%80%e4%b8%a5%e6%98%8e%e7%8e%89%e5%9b%a2%e9%98%9f%e6%96%b0%e4%bd%9c%ef%bc%9a-attention-%e5%b9%b6%e9%9d%9e%e6%b0%b8%e8%bf%9c%e6%98%af%e7%93%b6%e9%a2%88%ef%bc%8c%e5%a4%9a-gpu","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=40714","title":{"rendered":"\u8ba1\u7b97\u6240\u4e25\u660e\u7389\u56e2\u961f\u65b0\u4f5c\uff1a Attention \u5e76\u975e\u6c38\u8fdc\u662f\u74f6\u9888\uff0c\u591a GPU \u5e76\u4e0d\u4e00\u5b9a\u66f4\u5feb"},"content":{"rendered":"<p>\u968f\u7740\u5927\u8bed\u8a00\u6a21\u578b\u9010\u6e10\u8d70\u5411\u771f\u5b9e\u5e94\u7528\uff0c\u63a8\u7406\u9636\u6bb5\u7684\u6027\u80fd\u95ee\u9898\u6b63\u5728\u6210\u4e3a\u5236\u7ea6\u843d\u5730\u7684\u5173\u952e\u56e0\u7d20\u3002<\/p>\n<p>\u6a21\u578b\u89c4\u6a21\u4e0d\u65ad\u589e\u5927\u3001\u4e0a\u4e0b\u6587\u6301\u7eed\u62c9\u957f\uff0c\u518d\u52a0\u4e0a RAG\u3001MoE \u7b49\u65b0\u7528\u6cd5\u7684\u5f15\u5165\uff0c\u4f7f\u5f97\u5ef6\u8fdf\u3001\u541e\u5410\u548c\u80fd\u8017\u4e0d\u518d\u53ea\u662f\u201c\u8c03\u4e00\u8c03\u53c2\u6570\u201d\u7684\u95ee\u9898\uff0c\u800c\u662f\u76f4\u63a5\u5f71\u54cd\u7cfb\u7edf\u67b6\u6784\u548c\u7b97\u529b\u6210\u672c\u7684\u6838\u5fc3\u6311\u6218\u3002<\/p>\n<p>\u56f4\u7ed5\u8fd9\u4e9b\u95ee\u9898\uff0c\u6765\u81ea\u4e2d\u56fd\u79d1\u5b66\u9662\u8ba1\u7b97\u6240\u7684\u4e25\u660e\u7389\u56e2\u961f\uff0c\u8054\u5408\u4e2d\u56fd\u7535\u4fe1\u4e91\u8ba1\u7b97\u7814\u7a76\u9662\u3001\u6d59\u6c5f\u5b9e\u9a8c\u5ba4\u548c\u5317\u4eac\u5927\u5b66\u7684\u7814\u7a76\u8005\uff0c\u5728\u8bba\u6587\u300aA Systematic Characterization of LLM Inference on GPUs\u300b\u4e2d\uff0c\u5bf9\u5927\u8bed\u8a00\u6a21\u578b\u5728 GPU \u4e0a\u7684\u63a8\u7406\u884c\u4e3a\u8fdb\u884c\u4e86\u7cfb\u7edf\u7814\u7a76\u3002<\/p>\n<p>\u4e0d\u540c\u4e8e\u4ee5\u5f80\u4fa7\u91cd\u5355\u4e00\u6a21\u578b\u3001\u5355\u4e00\u7b97\u5b50\u6216\u5c40\u90e8\u4f18\u5316\u7684\u7814\u7a76\u8def\u5f84\uff0c\u4e25\u660e\u7389\u56e2\u961f\u4ece\u7cfb\u7edf\u548c\u786c\u4ef6\u534f\u540c\u7684\u89c6\u89d2\u51fa\u53d1\uff0c\u901a\u8fc7\u5927\u89c4\u6a21\u5b9e\u9a8c\uff0c\u5bf9\u5927\u8bed\u8a00\u6a21\u578b\u5728 GPU \u4e0a\u7684\u63a8\u7406\u884c\u4e3a\u8fdb\u884c\u4e86\u5168\u9762\u523b\u753b\uff0c\u8bd5\u56fe\u4ece\u6839\u672c\u4e0a\u56de\u7b54\u4e00\u4e2a\u957f\u671f\u56f0\u6270\u5de5\u7a0b\u5b9e\u8df5\u7684\u95ee\u9898\uff1a\u5927\u6a21\u578b\u63a8\u7406\u4e3a\u4ec0\u4e48\u4f1a\u5448\u73b0\u51fa\u73b0\u5728\u8fd9\u6837\u7684\u6027\u80fd\u7279\u5f81\u3002<\/p>\n<p>\u8fd9\u9879\u5de5\u4f5c\u5e76\u6ca1\u6709\u76f4\u63a5\u7ed9\u51fa\u5982\u4f55\u4f18\u5316\u7684\u73b0\u6210\u7b54\u6848\uff0c\u800c\u662f\u9996\u5148\u5efa\u7acb\u4e86\u4e00\u5957\u7edf\u4e00\u3001\u53ef\u89e3\u91ca\u7684\u6027\u80fd\u8ba4\u77e5\u6846\u67b6\uff0c\u5c06 Prefill \u4e0e Decode \u7684\u5dee\u5f02\u4e0a\u5347\u4e3a\u7cfb\u7edf\u7ea7\u7684\u57fa\u672c\u89c4\u5f8b\uff0c\u5e76\u5728\u4e0d\u540c\u6a21\u578b\u89c4\u6a21\u3001\u4e0d\u540c\u786c\u4ef6\u5e73\u53f0\u4ee5\u53ca MoE\u3001RAG \u7b49\u65b0\u578b\u63a8\u7406\u8303\u5f0f\u4e0b\u8fdb\u884c\u4e86\u9a8c\u8bc1\u3002<\/p>\n<p>\u5728\u8fd9\u4e00\u6846\u67b6\u4e0b\uff0c\u63a8\u7406\u5ef6\u8fdf\u3001\u8d44\u6e90\u5229\u7528\u7387\u548c\u80fd\u8017\u4e0d\u518d\u662f\u5b64\u7acb\u7684\u6307\u6807\uff0c\u800c\u662f\u968f\u7740\u5de5\u4f5c\u8d1f\u8f7d\u548c\u7cfb\u7edf\u914d\u7f6e\u53d1\u751f\u6709\u89c4\u5f8b\u7684\u53d8\u5316\u3002\u8fd9\u79cd\u4ece\u73b0\u8c61\u51fa\u53d1\u3001\u6700\u7ec8\u56de\u5230\u7cfb\u7edf\u6839\u56e0\u7684\u7814\u7a76\u65b9\u5f0f\uff0c\u4f7f\u5f97\u8fd9\u9879\u5de5\u4f5c\u66f4\u50cf\u662f\u5728\u4e3a\u5927\u6a21\u578b\u63a8\u7406\u5efa\u7acb\u4e00\u5f20\u53ef\u7406\u89e3\u3001\u53ef\u63a8\u6f14\u7684\u6027\u80fd\u5168\u666f\u56fe\uff0c\u800c\u4e0d\u662f\u63d0\u4f9b\u67d0\u4e2a\u573a\u666f\u4e0b\u7684\u7ecf\u9a8c\u7ed3\u8bba\u3002<\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-de193fbc1814016a93e693cf91e20ef6.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-de193fbc1814016a93e693cf91e20ef6.png\"><\/a><\/p>\n<p>\u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/pdf\/2512.01644v1<\/p>\n<div id=\"ez-toc-container\" class=\"ez-toc-v2_0_78 ez-toc-wrap-left-text counter-hierarchy ez-toc-counter ez-toc-light-blue ez-toc-container-direction\">\n<div class=\"ez-toc-title-container\">\n<p class=\"ez-toc-title\" style=\"cursor:inherit\">\u6587\u7ae0\u76ee\u5f55<\/p>\n<span class=\"ez-toc-title-toggle\"><a href=\"#\" class=\"ez-toc-pull-right ez-toc-btn ez-toc-btn-xs ez-toc-btn-default ez-toc-toggle\" aria-label=\"Toggle Table of Content\"><span class=\"ez-toc-js-icon-con\"><span class=\"\"><span class=\"eztoc-hide\" style=\"display:none;\">Toggle<\/span><span class=\"ez-toc-icon-toggle-span\"><svg style=\"fill: #999;color:#999\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" class=\"list-377408\" width=\"20px\" height=\"20px\" viewBox=\"0 0 24 24\" fill=\"none\"><path d=\"M6 6H4v2h2V6zm14 0H8v2h12V6zM4 11h2v2H4v-2zm16 0H8v2h12v-2zM4 16h2v2H4v-2zm16 0H8v2h12v-2z\" fill=\"currentColor\"><\/path><\/svg><svg style=\"fill: #999;color:#999\" class=\"arrow-unsorted-368013\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" width=\"10px\" height=\"10px\" viewBox=\"0 0 24 24\" version=\"1.2\" baseProfile=\"tiny\"><path d=\"M18.2 9.3l-6.2-6.3-6.2 6.3c-.2.2-.3.4-.3.7s.1.5.3.7c.2.2.4.3.7.3h11c.3 0 .5-.1.7-.3.2-.2.3-.5.3-.7s-.1-.5-.3-.7zM5.8 14.7l6.2 6.3 6.2-6.3c.2-.2.3-.5.3-.7s-.1-.5-.3-.7c-.2-.2-.4-.3-.7-.3h-11c-.3 0-.5.1-.7.3-.2.2-.3.5-.3.7s.1.5.3.7z\"\/><\/svg><\/span><\/span><\/span><\/a><\/span><\/div>\n<nav><ul class='ez-toc-list ez-toc-list-level-1 ' ><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-1\" href=\"https:\/\/17aitech.com\/?p=40714\/#%E4%B8%80%E6%AC%A1%E5%AF%B9%E5%A4%A7%E6%A8%A1%E5%9E%8B%E6%8E%A8%E7%90%86%E6%80%A7%E8%83%BD%E7%9A%84%E7%B3%BB%E7%BB%9F%E6%80%A7%E6%8B%86%E8%A7%A3\" >\u4e00\u6b21\u5bf9\u5927\u6a21\u578b\u63a8\u7406\u6027\u80fd\u7684\u7cfb\u7edf\u6027\u62c6\u89e3<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-2\" href=\"https:\/\/17aitech.com\/?p=40714\/#%E5%B7%A5%E4%BD%9C%E8%83%8C%E5%90%8E%E7%9A%84%E7%A0%94%E7%A9%B6%E8%80%85\" >\u5de5\u4f5c\u80cc\u540e\u7684\u7814\u7a76\u8005<\/a><\/li><\/ul><\/nav><\/div>\n<h2><span class=\"ez-toc-section\" id=\"%E4%B8%80%E6%AC%A1%E5%AF%B9%E5%A4%A7%E6%A8%A1%E5%9E%8B%E6%8E%A8%E7%90%86%E6%80%A7%E8%83%BD%E7%9A%84%E7%B3%BB%E7%BB%9F%E6%80%A7%E6%8B%86%E8%A7%A3\"><\/span>\u4e00\u6b21\u5bf9\u5927\u6a21\u578b\u63a8\u7406\u6027\u80fd\u7684\u7cfb\u7edf\u6027\u62c6\u89e3<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>\u4e25\u660e\u7389\u56e2\u961f\u7684\u8fd9\u9879\u5de5\u4f5c\u901a\u8fc7\u5927\u91cf\u6709\u7ec4\u7ec7\u7684\u5b9e\u9a8c\uff0c\u7cfb\u7edf\u5730\u603b\u7ed3\u4e86\u5927\u8bed\u8a00\u6a21\u578b\u5728\u63a8\u7406\u9636\u6bb5\u7684\u6027\u80fd\u89c4\u5f8b\uff0c\u800c\u4e0d\u662f\u505c\u7559\u5728\u96f6\u6563\u7684\u7ecf\u9a8c\u89c2\u5bdf\u4e0a\u3002\u7814\u7a76\u56e2\u961f\u53d1\u73b0\uff0cLLM \u7684\u63a8\u7406\u8fc7\u7a0b\u5728\u672c\u8d28\u4e0a\u53ef\u4ee5\u5206\u6210\u4e24\u4e2a\u5b8c\u5168\u4e0d\u540c\u7684\u9636\u6bb5\uff0c\u800c\u4e14\u8fd9\u79cd\u5dee\u5f02\u5e76\u4e0d\u662f\u9760\u4f18\u5316\u5c31\u80fd\u6d88\u9664\u7684\u3002<\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-fa76f920c7aa757976ef5b13229ec692.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-fa76f920c7aa757976ef5b13229ec692.png\"><\/a><\/p>\n<p>\u5177\u4f53\u6765\u8bf4\uff0c\u7b2c\u4e00\u4e2a\u9636\u6bb5\u662f Prefill \u9636\u6bb5\uff0c\u4e3b\u8981\u4efb\u52a1\u662f\u4e00\u6b21\u6027\u5904\u7406\u7528\u6237\u8f93\u5165\u7684 prompt\u3002\u8fd9\u4e2a\u9636\u6bb5\u53ef\u4ee5\u5e76\u884c\u6267\u884c\uff0c\u6838\u5fc3\u8ba1\u7b97\u662f\u5927\u89c4\u6a21\u77e9\u9635\u4e58\u6cd5\uff0c\u56e0\u6b64\u8ba1\u7b97\u91cf\u5927\u3001\u7b97\u5f97\u5f88\u6ee1\uff0cGPU \u7684\u8ba1\u7b97\u5355\u5143\u5229\u7528\u7387\u5f88\u9ad8\uff0c\u6574\u4f53\u6027\u80fd\u4e3b\u8981\u53d7\u9650\u4e8e\u7b97\u529b\u672c\u8eab\u3002<\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-1958b01fa9574fb6727299fc6d719838.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-1958b01fa9574fb6727299fc6d719838.png\"><\/a><\/p>\n<p>\u7b2c\u4e8c\u4e2a\u9636\u6bb5\u662f Decode \u9636\u6bb5\uff0c\u7528\u6765\u9010\u4e2a\u751f\u6210\u8f93\u51fa token\u3002\u7531\u4e8e\u751f\u6210\u8fc7\u7a0b\u662f\u4e00\u6b65\u4e00\u6b65\u8fdb\u884c\u7684\uff0c\u6bcf\u4e00\u6b65\u90fd\u9700\u8981\u8bbf\u95ee\u5df2\u7ecf\u7f13\u5b58\u7684\u4e0a\u4e0b\u6587\u4fe1\u606f\uff08KV Cache\uff09\uff0c\u5b9e\u9645\u8ba1\u7b97\u91cf\u4e0d\u5927\uff0c\u4f46\u5185\u5b58\u8bbf\u95ee\u975e\u5e38\u9891\u7e41\uff0c\u56e0\u6b64\u6027\u80fd\u74f6\u9888\u4ece\u7b97\u529b\u8f6c\u79fb\u5230\u4e86\u5185\u5b58\u5e26\u5bbd\u548c\u8bbf\u95ee\u5ef6\u8fdf\u4e0a\u3002<\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-c3e6676e1743c38039c8d18fa0aa5d37.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-c3e6676e1743c38039c8d18fa0aa5d37.png\"><\/a><\/p>\n<p>\u5728\u6b64\u57fa\u7840\u4e0a\uff0c\u8bba\u6587\u8fdb\u4e00\u6b65\u6307\u51fa\uff0c\u63a8\u7406\u8fc7\u7a0b\u4e2d\u5230\u5e95\u662f Prefill \u6162\u8fd8\u662f Decode \u6162\uff0c\u5e76\u4e0d\u662f\u56fa\u5b9a\u7684\uff0c\u800c\u662f\u53d6\u51b3\u4e8e\u5177\u4f53\u7684\u8f93\u5165\u548c\u8f93\u51fa\u60c5\u51b5\u3002\u5f53\u8f93\u5165\u8f83\u77ed\u65f6\uff0c\u867d\u7136\u6bcf\u4e00\u6b65 Decode \u7684\u8ba1\u7b97\u4e0d\u591a\uff0c\u4f46\u9700\u8981\u6267\u884c\u5f88\u591a\u6b65\uff0c\u56e0\u6b64 Decode \u5f80\u5f80\u6210\u4e3a\u4e3b\u8981\u8017\u65f6\u90e8\u5206\u3002<\/p>\n<p>\u800c\u5f53\u8f93\u5165\u53d8\u957f\u65f6\uff0cPrefill \u9636\u6bb5\u9700\u8981\u5904\u7406\u7684 token \u6570\u8fc5\u901f\u589e\u52a0\uff0c\u5176\u8ba1\u7b97\u91cf\u589e\u957f\u66f4\u5feb\uff0c\u5728\u8d85\u8fc7\u67d0\u4e2a\u957f\u5ea6\u4e4b\u540e\u5c31\u4f1a\u53cd\u8fc7\u6765\u6210\u4e3a\u6574\u4f53\u5ef6\u8fdf\u7684\u4e3b\u8981\u6765\u6e90\u3002\u8fd9\u79cd\u4ece Decode \u4e3b\u5bfc\u5230 Prefill \u4e3b\u5bfc\u7684\u8f6c\u53d8\u8bf4\u660e\uff0c\u6027\u80fd\u74f6\u9888\u66f4\u591a\u662f\u7531\u5de5\u4f5c\u8d1f\u8f7d\u51b3\u5b9a\u7684\uff0c\u800c\u4e0d\u662f\u6a21\u578b\u672c\u8eab\u5929\u751f\u6162\u5728\u54ea\u3002<\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-a036f115f6133d91be7818447ad298e3.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-a036f115f6133d91be7818447ad298e3.png\"><\/a><\/p>\n<p>\u5982\u679c\u8fdb\u4e00\u6b65\u62c6\u5230\u6a21\u578b\u5185\u90e8\u7684\u7b97\u5b50\u5c42\u9762\uff0c\u7814\u7a76\u56e2\u961f\u53d1\u73b0\u74f6\u9888\u540c\u6837\u4e0d\u662f\u56fa\u5b9a\u7684\u3002\u5728\u5e38\u89c1\u7684\u4e0a\u4e0b\u6587\u957f\u5ea6\u4e0b\uff0cPrefill \u9636\u6bb5\u7684\u4e3b\u8981\u65f6\u95f4\u5f80\u5f80\u82b1\u5728\u524d\u9988\u7f51\u7edc\uff08FFN\uff09\u4e0a\uff0c\u4f46\u5728\u4e0a\u4e0b\u6587\u7279\u522b\u957f\u7684\u60c5\u51b5\u4e0b\uff0c\u7531\u4e8e\u6ce8\u610f\u529b\u8ba1\u7b97\u7684\u590d\u6742\u5ea6\u589e\u957f\u66f4\u5feb\uff0cAttention \u4f1a\u9010\u6e10\u6210\u4e3a\u4e3b\u8981\u74f6\u9888\u3002<\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-5279966c2eba3b6b76e578ca633e3bad.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-5279966c2eba3b6b76e578ca633e3bad.png\"><\/a><\/p>\n<p>Decode \u9636\u6bb5\u7684\u60c5\u51b5\u5219\u548c\u6a21\u578b\u89c4\u6a21\u6709\u5173\uff1a\u5bf9\u4e8e\u8f83\u5c0f\u7684\u6a21\u578b\uff0c\u9891\u7e41\u8bbf\u95ee KV Cache \u7684 Attention \u66f4\u5bb9\u6613\u6210\u4e3a\u74f6\u9888\uff1b\u800c\u5bf9\u4e8e\u5927\u6a21\u578b\uff0c\u7531\u4e8e\u524d\u9988\u7f51\u7edc\u6743\u91cd\u66f4\u5927\uff0cFFN \u7684\u5185\u5b58\u52a0\u8f7d\u6210\u672c\u53cd\u800c\u66f4\u7a81\u51fa\u3002\u8fd9\u8bf4\u660e\uff0c\u5355\u7eaf\u5730\u8bf4 Attention \u662f\u74f6\u9888\u6216 FFN \u662f\u74f6\u9888\u90fd\u662f\u4e0d\u51c6\u786e\u7684\uff0c\u5fc5\u987b\u7ed3\u5408\u6240\u5904\u9636\u6bb5\u3001\u4e0a\u4e0b\u6587\u957f\u5ea6\u548c\u6a21\u578b\u89c4\u6a21\u6765\u5224\u65ad\u3002\u96f7\u5cf0\u7f51<span style=\"color:#FD5D3C\">(\u516c\u4f17\u53f7\uff1a\u96f7\u5cf0\u7f51)<\/span><\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-4235c15c18331e40e067cacf5b6b82d1.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-4235c15c18331e40e067cacf5b6b82d1.png\"><\/a><\/p>\n<p>\u5728\u6027\u80fd\u53ef\u9884\u6d4b\u6027\u65b9\u9762\uff0c\u8bba\u6587\u53d1\u73b0 Prefill \u9636\u6bb5\u7684\u884c\u4e3a\u975e\u5e38\u89c4\u5f8b\u3002\u5b83\u7684\u6267\u884c\u65f6\u95f4\u51e0\u4e4e\u53ea\u7531\u771f\u6b63\u9700\u8981\u8ba1\u7b97\u7684\u8f93\u5165 token \u6570\u91cf\u51b3\u5b9a\uff0c\u800c\u4e14\u4e8c\u8005\u4e4b\u95f4\u5448\u73b0\u975e\u5e38\u7a33\u5b9a\u7684\u7ebf\u6027\u5173\u7cfb\u3002\u8fd9\u610f\u5473\u7740\uff0c\u53ea\u8981\u77e5\u9053\u8f93\u5165\u957f\u5ea6\u548c\u7f13\u5b58\u547d\u4e2d\u60c5\u51b5\uff0c\u5c31\u53ef\u4ee5\u6bd4\u8f83\u51c6\u786e\u5730\u9884\u6d4b Prefill \u7684\u5ef6\u8fdf\uff0c\u8fd9\u5bf9\u7cfb\u7edf\u8c03\u5ea6\u548c\u8d44\u6e90\u89c4\u5212\u975e\u5e38\u6709\u7528\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0cDecode \u9636\u6bb5\u7531\u4e8e\u662f\u9010\u6b65\u751f\u6210\uff0c\u5e76\u4e14\u53d7\u5230\u91c7\u6837\u548c\u4e32\u884c\u4f9d\u8d56\u7684\u5f71\u54cd\uff0c\u6027\u80fd\u6ce2\u52a8\u66f4\u5927\uff0c\u4e5f\u66f4\u96be\u63d0\u524d\u9884\u6d4b\u3002<\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-e229a632e0a9af4577fcf020cdb58db0.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-e229a632e0a9af4577fcf020cdb58db0.png\"><\/a><\/p>\n<p>\u5728\u80fd\u8017\u5206\u6790\u4e2d\uff0c\u8bba\u6587\u7ed9\u51fa\u4e86\u4e00\u4e2a\u975e\u5e38\u76f4\u89c2\u4f46\u91cd\u8981\u7684\u7ed3\u8bba\uff1a\u6574\u4e2a\u63a8\u7406\u8fc7\u7a0b\u6d88\u8017\u7684\u80fd\u91cf\uff0c\u51e0\u4e4e\u5168\u90e8\u6765\u81ea Decode \u9636\u6bb5\u3002\u8f93\u5165\u6709\u591a\u957f\uff0c\u5bf9\u603b\u80fd\u8017\u5f71\u54cd\u5f88\u5c0f\uff0c\u800c\u8f93\u51fa\u751f\u6210\u4e86\u591a\u5c11 token\uff0c\u51e0\u4e4e\u76f4\u63a5\u51b3\u5b9a\u4e86\u80fd\u8017\u5927\u5c0f\u3002\u540c\u65f6\uff0c\u6a21\u578b\u53c2\u6570\u8d8a\u591a\uff0c\u603b\u80fd\u8017\u4e5f\u4f1a\u968f\u4e4b\u589e\u52a0\u3002\u8fd9\u8bf4\u660e\uff0c\u5728\u771f\u5b9e\u7cfb\u7edf\u4e2d\uff0c\u5982\u679c\u60f3\u8981\u964d\u4f4e\u63a8\u7406\u80fd\u8017\uff0c\u9650\u5236\u8f93\u51fa\u957f\u5ea6\u5f80\u5f80\u6bd4\u4f18\u5316 Prefill \u66f4\u6709\u6548\u3002<\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-82d75e62b6ab71b5c7d4e22043ecece8.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-82d75e62b6ab71b5c7d4e22043ecece8.png\"><\/a><\/p>\n<p>\u5728\u591a GPU \u6269\u5c55\u5b9e\u9a8c\u4e2d\uff0c\u7814\u7a76\u56e2\u961f\u53d1\u73b0\u5e76\u884c\u5316\u5e76\u4e0d\u662f\u5728\u6240\u6709\u60c5\u51b5\u4e0b\u90fd\u6709\u6548\u3002Prefill \u9636\u6bb5\u7531\u4e8e\u8ba1\u7b97\u91cf\u5927\uff0c\u628a\u8ba1\u7b97\u5206\u644a\u5230\u591a\u5f20 GPU \u4e0a\u901a\u5e38\u80fd\u5e26\u6765\u6536\u76ca\uff0c\u4f46 Decode \u9636\u6bb5\u6bcf\u4e00\u6b65\u8ba1\u7b97\u90fd\u5f88\u5c0f\uff0c\u591a GPU \u4e4b\u95f4\u7684\u901a\u4fe1\u548c\u540c\u6b65\u53cd\u800c\u4f1a\u6210\u4e3a\u8d1f\u62c5\uff0c\u5bfc\u81f4\u6027\u80fd\u63d0\u5347\u4e0d\u660e\u663e\u751a\u81f3\u53d8\u6162\u3002\u56e0\u6b64\uff0c\u5728 Decode \u4e3a\u4e3b\u7684\u573a\u666f\u4e0b\uff0c\u4f7f\u7528\u5355 GPU \u6216\u8f83\u8f7b\u91cf\u7684\u6d41\u6c34\u5e76\u884c\u5f80\u5f80\u66f4\u5408\u9002\uff0c\u8fd9\u4e5f\u6253\u7834\u4e86GPU \u8d8a\u591a\u8d8a\u5feb\u7684\u76f4\u89c9\u3002<\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-f5e54335b2a238128cf22ef922c59bb5.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-f5e54335b2a238128cf22ef922c59bb5.png\"><\/a><\/p>\n<p>\u6700\u540e\uff0c\u8bba\u6587\u8fd8\u5206\u6790\u4e86\u65b0\u7684\u63a8\u7406\u8303\u5f0f\u3002\u5bf9\u4e8e MoE \u6a21\u578b\uff0c\u63a8\u7406\u901f\u5ea6\u4e3b\u8981\u53d6\u51b3\u4e8e\u6bcf\u6b21\u5b9e\u9645\u53c2\u4e0e\u8ba1\u7b97\u7684\u53c2\u6570\u89c4\u6a21\uff0c\u800c\u4e0d\u662f\u6a21\u578b\u7684\u603b\u53c2\u6570\u91cf\uff0c\u8fd9\u5e26\u6765\u4e86\u660e\u663e\u7684\u6027\u80fd\u4f18\u52bf\uff0c\u4f46\u540c\u65f6\uff0c\u5728 Decode \u9636\u6bb5\u4f1a\u989d\u5916\u5f15\u5165\u4e13\u5bb6\u9009\u62e9\u548c\u8c03\u5ea6\u7684\u5f00\u9500\uff0c\u4f7f\u6027\u80fd\u8868\u73b0\u66f4\u52a0\u590d\u6742\u3002<\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-0bb9eea38356cf12e1fe66ad5da67189.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-0bb9eea38356cf12e1fe66ad5da67189.png\"><\/a><\/p>\n<p>\u5bf9\u4e8e RAG \u5de5\u4f5c\u6d41\uff0c\u7814\u7a76\u56e2\u961f\u53d1\u73b0\u968f\u7740\u5916\u90e8\u77e5\u8bc6\u89c4\u6a21\u53d8\u5927\uff0c\u7cfb\u7edf\u74f6\u9888\u4f1a\u4ece GPU \u63a8\u7406\u8f6c\u79fb\u5230 CPU \u4fa7\u7684\u68c0\u7d22\u548c\u5185\u5b58\u8bbf\u95ee\u4e0a\uff0c\u5f62\u6210\u65b0\u7684\u6027\u80fd\u9650\u5236\u3002\u5c3d\u7ba1\u63a8\u7406\u6d41\u7a0b\u53d8\u5f97\u66f4\u52a0\u590d\u6742\uff0c\u4f46 Prefill \u548c Decode \u5728\u6027\u80fd\u4e0a\u7684\u6839\u672c\u5dee\u5f02\u4f9d\u7136\u5b58\u5728\uff0c\u4ecd\u7136\u662f\u7406\u89e3\u6574\u4f53\u884c\u4e3a\u7684\u5173\u952e\u3002<\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-8cf011931de743421f671e84bca4e752.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-8cf011931de743421f671e84bca4e752.png\"><\/a><\/p>\n<p class=\"heading-2 ace-line old-record-id-K8tNdWEy0ovG3nxDejBcnUxLnK8\">\u9762\u5411\u7cfb\u7edf\u7406\u89e3\u7684\u5927\u6a21\u578b\u63a8\u7406\u5b9e\u9a8c\u6846\u67b6<\/p>\n<p>\u4e3a\u4e86\u8ba9\u7ed3\u8bba\u66f4\u5b8c\u6574\u3001\u4e5f\u66f4\u5bb9\u6613\u7406\u89e3\uff0c\u8fd9\u7bc7\u8bba\u6587\u5728\u5b9e\u9a8c\u8bbe\u8ba1\u4e0a\u91c7\u7528\u4e86\u4e00\u79cd\u7531\u6d45\u5165\u6df1\u7684\u5206\u6790\u601d\u8def\u3002\u7814\u7a76\u56e2\u961f\u5e76\u4e0d\u662f\u4e00\u5f00\u59cb\u5c31\u7814\u7a76\u5e95\u5c42\u786c\u4ef6\u7ec6\u8282\uff0c\u800c\u662f\u5148\u89c2\u5bdf\u6574\u4f53\u63a8\u7406\u6027\u80fd\u8868\u73b0\uff0c\u518d\u9010\u6b65\u6df1\u5165\u5230 GPU \u7684\u6267\u884c\u548c\u5b58\u50a8\u884c\u4e3a\uff0c\u6700\u540e\u628a\u5f97\u5230\u7684\u89c4\u5f8b\u653e\u56de\u5230\u771f\u5b9e\u7cfb\u7edf\u548c\u65b0\u578b\u63a8\u7406\u573a\u666f\u4e2d\u8fdb\u884c\u9a8c\u8bc1\u3002\u96f7\u5cf0\u7f51<\/p>\n<p>\u5728\u5b9e\u9a8c\u5e73\u53f0\u65b9\u9762\uff0c\u8bba\u6587\u540c\u65f6\u4f7f\u7528\u4e86\u6570\u636e\u4e2d\u5fc3\u7ea7\u7684 GPU\uff08A100\uff09\u548c\u8fb9\u7f18\u8bbe\u5907\u4e0a\u7684 GPU\uff08Jetson AGX Orin\uff09\u3002\u8fd9\u6837\u505a\u7684\u76ee\u7684\uff0c\u662f\u68c0\u9a8c\u524d\u9762\u603b\u7ed3\u51fa\u7684\u6027\u80fd\u89c4\u5f8b\u662f\u5426\u53ea\u5728\u9ad8\u7aef\u670d\u52a1\u5668\u4e0a\u6210\u7acb\uff0c\u8fd8\u662f\u5728\u7b97\u529b\u548c\u5185\u5b58\u6761\u4ef6\u66f4\u53d7\u9650\u7684\u8bbe\u5907\u4e0a\u540c\u6837\u9002\u7528\u3002<\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-d953e5eed36b2542f70d7f5db78ec817.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-d953e5eed36b2542f70d7f5db78ec817.png\"><\/a><\/p>\n<p>\u6a21\u578b\u9009\u62e9\u4e0a\uff0c\u7814\u7a76\u56e2\u961f\u8986\u76d6\u4e86\u591a\u79cd\u4e3b\u6d41\u7684 dense \u6a21\u578b\uff08\u4ece 7B \u5230 32B\uff09\uff0c\u540c\u65f6\u8fd8\u5f15\u5165\u4e86\u4e00\u4e2a\u5177\u6709\u4ee3\u8868\u6027\u7684 MoE \u6a21\u578b\uff0c\u4ee5\u89c2\u5bdf\u4e0d\u540c\u53c2\u6570\u89c4\u6a21\u548c\u4e0d\u540c\u67b6\u6784\u8bbe\u8ba1\u5bf9\u63a8\u7406\u6027\u80fd\u7684\u5f71\u54cd\u3002\u4e3a\u4e86\u4fdd\u8bc1\u5bf9\u6bd4\u516c\u5e73\uff0c\u6240\u6709\u5b9e\u9a8c\u90fd\u5728\u540c\u4e00\u5957\u63a8\u7406\u6846\u67b6\u548c\u76f8\u540c\u7cbe\u5ea6\u8bbe\u7f6e\u4e0b\u8fdb\u884c\uff0c\u5c3d\u91cf\u51cf\u5c11\u5b9e\u73b0\u7ec6\u8282\u5e26\u6765\u7684\u5e72\u6270\u3002<\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-a36a46869afb5041ae56930d3be4a698.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-a36a46869afb5041ae56930d3be4a698.png\"><\/a><\/p>\n<p>\u5728\u5de5\u4f5c\u8d1f\u8f7d\u8bbe\u8ba1\u4e0a\uff0c\u7814\u7a76\u56e2\u961f\u5e76\u6ca1\u6709\u7b80\u5355\u5730\u8dd1\u51e0\u7ec4\u56fa\u5b9a benchmark\uff0c\u800c\u662f\u6709\u9488\u5bf9\u6027\u5730\u8bbe\u8ba1\u4e86\u591a\u79cd\u8f93\u5165\u548c\u8f93\u51fa\u7ec4\u5408\u3002\u4f8b\u5982\uff0c\u6709\u7684\u4efb\u52a1\u8f93\u5165\u77ed\u3001\u8f93\u51fa\u957f\uff0c\u6709\u7684\u8f93\u5165\u957f\u3001\u8f93\u51fa\u77ed\uff0c\u8fd8\u6709\u771f\u5b9e\u5bf9\u8bdd\u6570\u636e\u548c\u53ef\u63a7\u7684\u5408\u6210\u6570\u636e\u3002\u8fd9\u6837\u7684\u8bbe\u8ba1\u662f\u4e3a\u4e86\u6709\u610f\u8bc6\u5730\u5236\u9020 Prefill \u5360\u4e3b\u5bfc\u6216 Decode \u5360\u4e3b\u5bfc\u7684\u4e0d\u540c\u573a\u666f\uff0c\u4ece\u800c\u9a8c\u8bc1\u4e24\u9636\u6bb5\u5728\u4e0d\u540c\u6761\u4ef6\u4e0b\u662f\u5426\u59cb\u7ec8\u8868\u73b0\u51fa\u4e0d\u540c\u7684\u6027\u80fd\u7279\u5f81\u3002<\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-7eb2d128e471e382b019a49bd2f15317.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-7eb2d128e471e382b019a49bd2f15317.png\"><\/a><\/p>\n<p>\u5728\u6027\u80fd\u5206\u6790\u65b9\u6cd5\u4e0a\uff0c\u8bba\u6587\u91c7\u7528\u4e86\u5206\u5c42\u5256\u6790\u7684\u65b9\u5f0f\u3002\u9996\u5148\u5728\u6574\u4f53\u5c42\u9762\u4e0a\uff0c\u6d4b\u91cf\u7aef\u5230\u7aef\u7684\u5ef6\u8fdf\u3001\u541e\u5410\u91cf\u548c\u80fd\u8017\uff0c\u5efa\u7acb\u5bf9\u7cfb\u7edf\u884c\u4e3a\u7684\u76f4\u89c2\u8ba4\u8bc6\uff1b\u63a5\u7740\u5728\u9636\u6bb5\u548c\u7b97\u5b50\u5c42\u9762\uff0c\u5206\u6790 Prefill \u548c Decode \u5404\u81ea\u5360\u7528\u4e86\u591a\u5c11\u65f6\u95f4\uff0c\u4ee5\u53ca\u4e0d\u540c\u7b97\u5b50\u5728\u5176\u4e2d\u7684\u4f5c\u7528\uff1b\u6700\u540e\u6df1\u5165\u5230\u786c\u4ef6\u5c42\u9762\uff0c\u901a\u8fc7 Roofline \u6a21\u578b\u3001warp \u505c\u987f\u5206\u6790\uff0c\u4ee5\u53ca\u7f13\u5b58\u547d\u4e2d\u7387\u548c\u5185\u5b58\u5e26\u5bbd\u4f7f\u7528\u60c5\u51b5\uff0c\u6765\u5224\u65ad\u6027\u80fd\u7a76\u7adf\u662f\u53d7\u9650\u4e8e\u8ba1\u7b97\u8fd8\u662f\u53d7\u9650\u4e8e\u5185\u5b58\u3002<\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-fad8a03d15e3a0db373beb39f6b38a98.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-fad8a03d15e3a0db373beb39f6b38a98.png\"><\/a><\/p>\n<p>\u5728\u80fd\u8017\u5206\u6790\u4e2d\uff0c\u7814\u7a76\u4eba\u5458\u901a\u8fc7\u9ad8\u9891\u529f\u7387\u91c7\u6837\u5e76\u6263\u9664\u7a7a\u95f2\u529f\u8017\u7684\u65b9\u5f0f\uff0c\u5c3d\u91cf\u4fdd\u8bc1\u6d4b\u91cf\u7ed3\u679c\u7684\u51c6\u786e\u6027\u3002\u5728\u6d89\u53ca RAG \u7684\u5b9e\u9a8c\u4e2d\uff0c\u8fd8\u989d\u5916\u4f7f\u7528\u4e86 CPU \u4fa7\u7684\u6027\u80fd\u5206\u6790\u5de5\u5177\uff0c\u4e13\u95e8\u5206\u6790\u68c0\u7d22\u9636\u6bb5\u7684\u74f6\u9888\u6765\u6e90\u3002<\/p>\n<p>\u901a\u8fc7\u8fd9\u79cd\u4ece\u201c\u770b\u73b0\u8c61\u201d\u5230\u201c\u627e\u539f\u56e0\u201d\u7684\u9010\u6b65\u5206\u6790\u8fc7\u7a0b\uff0c\u8bba\u6587\u907f\u514d\u4e86\u53ea\u51ed\u7ecf\u9a8c\u5224\u65ad\u6216\u7b80\u5355\u76f8\u5173\u6027\u5206\u6790\u4e0b\u7ed3\u8bba\u7684\u95ee\u9898\uff0c\u4f7f\u5f97\u6bcf\u4e00\u4e2a\u5b8f\u89c2\u5c42\u9762\u7684\u6027\u80fd\u73b0\u8c61\uff0c\u90fd\u80fd\u5728\u5e95\u5c42\u786c\u4ef6\u6267\u884c\u673a\u5236\u4e0a\u627e\u5230\u6e05\u6670\u7684\u89e3\u91ca\u3002<\/p>\n<p class=\"heading-2 ace-line old-record-id-MhqMdeUeBotjNwxGMqtc5rFqnXg\">\u5f53\u95ee\u9898\u88ab\u770b\u6e05\uff0c\u4f18\u5316\u624d\u6709\u65b9\u5411<\/p>\n<p>\u8fd9\u7bc7\u8bba\u6587\u7684\u610f\u4e49\u4e0d\u5728\u4e8e\u63d0\u51fa\u67d0\u4e00\u79cd\u65b0\u7684\u4f18\u5316\u6280\u5de7\uff0c\u800c\u5728\u4e8e\u5efa\u7acb\u4e86\u4e00\u5957\u7edf\u4e00\u3001\u53ef\u89e3\u91ca\u7684\u5927\u6a21\u578b\u63a8\u7406\u6027\u80fd\u8ba4\u77e5\u6846\u67b6\u3002\u5b83\u9996\u6b21\u5c06 Prefill \u4e0e Decode \u7684\u9636\u6bb5\u5dee\u5f02\u63d0\u5347\u4e3a\u7cfb\u7edf\u7ea7\u57fa\u672c\u89c4\u5f8b\uff0c\u5e76\u8bc1\u660e\u8fd9\u4e00\u89c4\u5f8b\u5728\u4e0d\u540c\u6a21\u578b\u89c4\u6a21\u3001\u786c\u4ef6\u5e73\u53f0\u548c\u65b0\u5174\u63a8\u7406\u8303\u5f0f\u4e0b\u5747\u6210\u7acb\u3002<\/p>\n<p>\u4ece\u5de5\u7a0b\u89d2\u5ea6\u770b\uff0c\u8bba\u6587\u7ea0\u6b63\u4e86\u591a\u4e2a\u957f\u671f\u5b58\u5728\u7684\u76f4\u89c9\u6027\u8bef\u89e3\uff0c\u4f8b\u5982Attention \u6c38\u8fdc\u662f\u74f6\u9888\u3001\u591a GPU \u4e00\u5b9a\u66f4\u5feb\u3001Prefill \u662f\u4e3b\u8981\u80fd\u8017\u6765\u6e90\u7b49\uff0c\u5e76\u7ed9\u51fa\u4e86\u660e\u786e\u7684\u53cd\u4f8b\u548c\u673a\u5236\u89e3\u91ca\u3002\u8fd9\u4e9b\u7ed3\u8bba\u5bf9\u5b9e\u9645\u63a8\u7406\u670d\u52a1\u7684\u90e8\u7f72\u7b56\u7565\u3001\u8d44\u6e90\u914d\u7f6e\u548c\u6210\u672c\u63a7\u5236\u5177\u6709\u76f4\u63a5\u6307\u5bfc\u610f\u4e49\u3002<\/p>\n<p>\u4ece\u7cfb\u7edf\u7814\u7a76\u89d2\u5ea6\u770b\uff0c\u8fd9\u7bc7\u5185\u5bb9\u4e3a\u540e\u7eed\u5de5\u4f5c\u63d0\u4f9b\u4e86\u6e05\u6670\u7684\u95ee\u9898\u5206\u89e3\u65b9\u5f0f\uff1a\u4f18\u5316 Prefill \u548c\u4f18\u5316 Decode \u4e0d\u5e94\u6df7\u4e3a\u4e00\u8c08\uff0c\u800c\u5e94\u9488\u5bf9\u5404\u81ea\u7684\u6839\u672c\u74f6\u9888\u5206\u522b\u8bbe\u8ba1\u673a\u5236\u3002\u8fd9\u4e00\u601d\u60f3\u5bf9\u8c03\u5ea6\u5668\u8bbe\u8ba1\u3001\u5e76\u884c\u7b56\u7565\u9009\u62e9\u4ee5\u53ca\u65b0\u786c\u4ef6\u7279\u6027\u5229\u7528\u90fd\u5177\u6709\u542f\u53d1\u4f5c\u7528\u3002<\/p>\n<p>\u4ece\u672a\u6765\u53d1\u5c55\u770b\uff0c\u8bba\u6587\u6307\u51fa\u4e86 MoE \u548c RAG \u7b49\u65b0\u8303\u5f0f\u5982\u4f55\u91cd\u5851\u74f6\u9888\u4f4d\u7f6e\uff0c\u63d0\u793a\u7814\u7a76\u8005\u5728\u6a21\u578b\u548c\u7cfb\u7edf\u534f\u540c\u8bbe\u8ba1\u65f6\u9700\u8981\u5173\u6ce8\u8def\u7531\u5f00\u9500\u3001\u5185\u5b58\u5c40\u90e8\u6027\u548c CPU\u2013GPU \u534f\u540c\uff0c\u800c\u4e0d\u4ec5\u4ec5\u662f\u7b97\u529b\u89c4\u6a21\u3002<\/p>\n<p>\u603b\u4f53\u800c\u8a00\uff0c\u8fd9\u662f\u4e00\u7bc7\u4ee5\u5b9e\u9a8c\u4e3a\u57fa\u7840\u3001\u4ee5\u89e3\u91ca\u4e3a\u6838\u5fc3\u3001\u4ee5\u7cfb\u7edf\u8ba4\u77e5\u4e3a\u76ee\u6807\u7684\u8bba\u6587\uff0c\u5176\u4ef7\u503c\u5728\u4e8e\u56de\u7b54\u4e86\u201c\u4e3a\u4ec0\u4e48 LLM \u63a8\u7406\u4f1a\u8fd9\u6837\u8868\u73b0\u201d\uff0c\u800c\u4e0d\u4ec5\u662f\u201c\u5982\u4f55\u8ba9\u5b83\u66f4\u5feb\u4e00\u70b9\u201d\u3002<\/p>\n<h2><span class=\"ez-toc-section\" id=\"%E5%B7%A5%E4%BD%9C%E8%83%8C%E5%90%8E%E7%9A%84%E7%A0%94%E7%A9%B6%E8%80%85\"><\/span>\u5de5\u4f5c\u80cc\u540e\u7684\u7814\u7a76\u8005<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>\u672c\u6587\u901a\u8baf\u4f5c\u8005\u4e3a \u4e2d\u56fd\u79d1\u5b66\u9662\u8ba1\u7b97\u6280\u672f\u7814\u7a76\u6240\u7684\u4e25\u660e\u7389\u6559\u6388\u3002\u4ed6\u4e3b\u8981\u4ece\u4e8b\u8ba1\u7b97\u673a\u4f53\u7cfb\u7ed3\u6784\u76f8\u5173\u7814\u7a76\uff0c\u7814\u7a76\u65b9\u5411\u6db5\u76d6\u56fe\u673a\u5668\u5b66\u4e60\u3001\u8bbe\u8ba1\u7a7a\u95f4\u63a2\u7d22\u4ee5\u53ca\u590d\u6742\u8ba1\u7b97\u7cfb\u7edf\u7684\u6027\u80fd\u5206\u6790\u7b49\u95ee\u9898\u3002<\/p>\n<p style=\"text-align: center\"><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-8e30ca50437da150af12a158b48c2853.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-8e30ca50437da150af12a158b48c2853.png\"><\/a><\/p>\n<p>\u53c2\u8003\u94fe\u63a5\uff1ahttps:\/\/mingyuyan-ict.github.io\/MingyuYan-ICT\/<\/p>\n<p>\u5728\u5b66\u672f\u7814\u7a76\u65b9\u9762\uff0c\u4e25\u660e\u7389\u6559\u6388\u5df2\u5728\u591a\u4e2a\u56fd\u9645\u9876\u7ea7\u4f1a\u8bae\u548c\u671f\u520a\u4e0a\u53d1\u8868\u8fd1 20 \u7bc7\u8bba\u6587\uff0c\u76f8\u5173\u6210\u679c\u53d1\u8868\u4e8e MICRO\u3001HPCA\u3001DAC\u3001ICCAD\u3001IJCAI\u3001IEEE TC\u3001IEEE TPDS \u7b49\u91cd\u8981\u5b66\u672f\u5e73\u53f0\uff0c\u8986\u76d6\u4f53\u7cfb\u7ed3\u6784\u3001\u7cfb\u7edf\u4e0e\u5e94\u7528\u4ea4\u53c9\u7b49\u591a\u4e2a\u7814\u7a76\u9886\u57df\u3002<\/p>\n<p>\u9664\u79d1\u7814\u5de5\u4f5c\u5916\uff0c\u4e25\u660e\u7389\u6559\u6388\u4e5f\u662f IEEE \u548c\u4e2d\u56fd\u8ba1\u7b97\u673a\u5b66\u4f1a\uff08CCF\uff09\u7684\u9ad8\u7ea7\u4f1a\u5458\uff0c\u5e76\u591a\u6b21\u62c5\u4efb HPCA\u3001ISCA\u3001MICRO\u3001IJCAI\u3001ISPASS \u7b49\u56fd\u9645\u4f1a\u8bae\u7684\u6280\u672f\u7a0b\u5e8f\u59d4\u5458\u4f1a\u59d4\u5458\u6216\u5ba1\u7a3f\u4eba\uff0c\u957f\u671f\u53c2\u4e0e\u76f8\u5173\u9886\u57df\u7684\u5b66\u672f\u8bc4\u5ba1\u4e0e\u793e\u533a\u5efa\u8bbe\u3002<\/p>\n<p>\u5728\u5b66\u672f\u57f9\u517b\u4e0e\u79d1\u7814\u7ecf\u5386\u65b9\u9762\uff0c\u4ed6\u4e8e\u4e2d\u56fd\u79d1\u5b66\u9662\u5927\u5b66\u83b7\u5f97\u535a\u58eb\u5b66\u4f4d\uff0c\u5e76\u66fe\u8d74\u7f8e\u56fd\u52a0\u5dde\u5927\u5b66\u5723\u5854\u82ad\u82ad\u62c9\u5206\u6821\u8fdb\u884c\u8054\u5408\u57f9\u517b\u3002\u5176\u535a\u58eb\u8bba\u6587\u66fe\u83b7\u5f97\u4e2d\u56fd\u8ba1\u7b97\u673a\u5b66\u4f1a\u4f18\u79c0\u535a\u58eb\u8bba\u6587\u5956\u3002\u6b64\u5916\uff0c\u4ed6\u8fd8\u5165\u9009\u5317\u4eac\u5e02\u79d1\u6280\u65b0\u661f\u8ba1\u5212\u3001\u4e2d\u56fd\u79d1\u5b66\u9662\u9752\u5e74\u521b\u65b0\u4fc3\u8fdb\u4f1a\uff0c\u5e76\u4e3b\u6301\u6216\u53c2\u4e0e\u4e2d\u56fd\u79d1\u5b66\u9662\u9752\u5e74\u56e2\u961f\u9879\u76ee\u7b49\u79d1\u7814\u8ba1\u5212\u3002<\/p>\n<p>\u603b\u4f53\u800c\u8a00\uff0c\u4e25\u660e\u7389\u6559\u6388\u957f\u671f\u81f4\u529b\u4e8e\u4ece\u7cfb\u7edf\u89c6\u89d2\u7406\u89e3\u590d\u6742\u8ba1\u7b97\u8d1f\u8f7d\u5728\u786c\u4ef6\u5e73\u53f0\u4e0a\u7684\u6267\u884c\u884c\u4e3a\uff0c\u5f3a\u8c03\u901a\u8fc7\u7cfb\u7edf\u6027\u5b9e\u9a8c\u548c\u786c\u4ef6\u884c\u4e3a\u5206\u6790\u63ed\u793a\u6027\u80fd\u74f6\u9888\u7684\u5f62\u6210\u673a\u7406\uff0c\u5176\u7814\u7a76\u6210\u679c\u517c\u5177\u7406\u8bba\u6df1\u5ea6\u4e0e\u5de5\u7a0b\u5b9e\u8df5\u4ef7\u503c\u3002<\/p>\n<p>\u96f7\u5cf0\u7f51\u539f\u521b\u6587\u7ae0\uff0c\u672a\u7ecf\u6388\u6743\u7981\u6b62\u8f6c\u8f7d\u3002\u8be6\u60c5\u89c1<a href=\"https:\/\/r.xiumi.us\/board\/v5\/3qhbI\/392977584\" rel=\"nofollow\" target=\"_blank\">\u8f6c\u8f7d\u987b\u77e5<\/a>\u3002<\/p>\n<p>\n                        <a href=\"https:\/\/vip.yanxishe.com\/?from=leiphonearticle\" target=\"_blank\"><br \/>\n                            <a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-caffd5db911581c593515fca738917f2.jpeg\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-caffd5db911581c593515fca738917f2.jpeg\"><\/a><br \/>\n                        <\/a>\n                    <\/p>\n<p>                                            \u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.leiphone.com\/category\/academic\/6IlI4YnUgmE2ru6N.html\" target=\"_blank\">\u8ba1\u7b97\u6240\u4e25\u660e\u7389\u56e2\u961f\u65b0\u4f5c\uff1a Attention \u5e76\u975e\u6c38\u8fdc\u662f\u74f6\u9888\uff0c\u591a GPU \u5e76\u4e0d\u4e00\u5b9a\u66f4\u5feb<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u968f\u7740\u5927\u8bed\u8a00\u6a21\u578b\u9010\u6e10\u8d70\u5411\u771f\u5b9e\u5e94\u7528\uff0c\u63a8\u7406\u9636\u6bb5 [&hellip;]<\/p>\n","protected":false},"author":3,"featured_media":0,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[27],"tags":[71,68,55],"class_list":["post-40714","post","type-post","status-publish","format-standard","hentry","category-news","tag-rag","tag-68","tag-55"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v26.4 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>\u8ba1\u7b97\u6240\u4e25\u660e\u7389\u56e2\u961f\u65b0\u4f5c\uff1a Attention \u5e76\u975e\u6c38\u8fdc\u662f\u74f6\u9888\uff0c\u591a GPU \u5e76\u4e0d\u4e00\u5b9a\u66f4\u5feb - \u4e00\u8d77AI\u6280\u672f<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/17aitech.com\/?p=40714\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/17aitech.com\/?p=40714\",\"url\":\"https:\/\/17aitech.com\/?p=40714\",\"name\":\"\u8ba1\u7b97\u6240\u4e25\u660e\u7389\u56e2\u961f\u65b0\u4f5c\uff1a Attention \u5e76\u975e\u6c38\u8fdc\u662f\u74f6\u9888\uff0c\u591a GPU \u5e76\u4e0d\u4e00\u5b9a\u66f4\u5feb - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\/\/17aitech.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/17aitech.com\/?p=40714#primaryimage\"},\"image\":{\"@id\":\"https:\/\/17aitech.com\/?p=40714#primaryimage\"},\"thumbnailUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-de193fbc1814016a93e693cf91e20ef6.png\",\"datePublished\":\"2025-12-22T14:06:09+00:00\",\"author\":{\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\/\/17aitech.com\/?p=40714#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/17aitech.com\/?p=40714\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/?p=40714#primaryimage\",\"url\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-de193fbc1814016a93e693cf91e20ef6.png\",\"contentUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-de193fbc1814016a93e693cf91e20ef6.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/17aitech.com\/?p=40714#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/17aitech.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u8ba1\u7b97\u6240\u4e25\u660e\u7389\u56e2\u961f\u65b0\u4f5c\uff1a Attention \u5e76\u975e\u6c38\u8fdc\u662f\u74f6\u9888\uff0c\u591a GPU \u5e76\u4e0d\u4e00\u5b9a\u66f4\u5feb\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/17aitech.com\/#website\",\"url\":\"https:\/\/17aitech.com\/\",\"name\":\"\u4e00\u8d77AI\u6280\u672f\",\"description\":\"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca\",\"alternateName\":\"\u4e00\u8d77AI\u6280\u672f\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/17aitech.com\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\",\"name\":\"AI\u5c0f\u52a9\u624b\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/image\/\",\"url\":\"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png\",\"contentUrl\":\"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png\",\"caption\":\"AI\u5c0f\u52a9\u624b\"},\"description\":\"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e\",\"url\":\"https:\/\/17aitech.com\/?page_id=33738&user=3\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u8ba1\u7b97\u6240\u4e25\u660e\u7389\u56e2\u961f\u65b0\u4f5c\uff1a Attention \u5e76\u975e\u6c38\u8fdc\u662f\u74f6\u9888\uff0c\u591a GPU \u5e76\u4e0d\u4e00\u5b9a\u66f4\u5feb - \u4e00\u8d77AI\u6280\u672f","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:\/\/17aitech.com\/?p=40714","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=40714","url":"https:\/\/17aitech.com\/?p=40714","name":"\u8ba1\u7b97\u6240\u4e25\u660e\u7389\u56e2\u961f\u65b0\u4f5c\uff1a Attention \u5e76\u975e\u6c38\u8fdc\u662f\u74f6\u9888\uff0c\u591a GPU \u5e76\u4e0d\u4e00\u5b9a\u66f4\u5feb - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=40714#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=40714#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-de193fbc1814016a93e693cf91e20ef6.png","datePublished":"2025-12-22T14:06:09+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=40714#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=40714"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=40714#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-de193fbc1814016a93e693cf91e20ef6.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/12\/frc-de193fbc1814016a93e693cf91e20ef6.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=40714#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"\u8ba1\u7b97\u6240\u4e25\u660e\u7389\u56e2\u961f\u65b0\u4f5c\uff1a Attention \u5e76\u975e\u6c38\u8fdc\u662f\u74f6\u9888\uff0c\u591a GPU \u5e76\u4e0d\u4e00\u5b9a\u66f4\u5feb"}]},{"@type":"WebSite","@id":"https:\/\/17aitech.com\/#website","url":"https:\/\/17aitech.com\/","name":"\u4e00\u8d77AI\u6280\u672f","description":"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca","alternateName":"\u4e00\u8d77AI\u6280\u672f","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/17aitech.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d","name":"AI\u5c0f\u52a9\u624b","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/#\/schema\/person\/image\/","url":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","contentUrl":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","caption":"AI\u5c0f\u52a9\u624b"},"description":"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e","url":"https:\/\/17aitech.com\/?page_id=33738&user=3"}]}},"_links":{"self":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/40714","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=40714"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/40714\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=40714"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=40714"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=40714"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}