{"id":34772,"date":"2024-12-12T18:57:42","date_gmt":"2024-12-12T10:57:42","guid":{"rendered":"https:\/\/17aitech.com\/?p=34772"},"modified":"2024-12-12T18:57:42","modified_gmt":"2024-12-12T10:57:42","slug":"%e8%b4%be%e6%89%ac%e6%b8%85%e7%82%b9%e8%b5%9e%ef%bc%9a3k-star%e9%87%8f%e7%9a%84sglang%e4%b8%8a%e6%96%b0%ef%bc%8c%e5%8a%a0%e9%80%9fllama-405b%e6%8e%a8%e7%90%86%e7%a7%92%e6%9d%80vllm%e3%80%81tensorrt-ll","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=34772","title":{"rendered":"\u8d3e\u626c\u6e05\u70b9\u8d5e\uff1a3K star\u91cf\u7684SGLang\u4e0a\u65b0\uff0c\u52a0\u901fLlama 405B\u63a8\u7406\u79d2\u6740vLLM\u3001TensorRT-LLM"},"content":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2024-07-27-3\" target=\"_blank\">\u8d3e\u626c\u6e05\u70b9\u8d5e\uff1a3K star\u91cf\u7684SGLang\u4e0a\u65b0\uff0c\u52a0\u901fLlama 405B\u63a8\u7406\u79d2\u6740vLLM\u3001TensorRT-LLM<\/a><\/p>\n<blockquote data-author-name=\"\" data-content-utf8-length=\"1\" data-source-title=\"\" data-type=\"2\" data-url=\"\">\n<p>\u7528\u6765\u8fd0\u884c Llama 3 405B \u4f18\u52bf\u660e\u663e\u3002<\/p>\n<\/blockquote>\n<section><\/section>\n<section>\u6700\u8fd1\uff0cMeta \u5f00\u6e90\u4e86\u6700\u65b0\u7684 405B \u6a21\u578b\uff08Llama 3.1 405B\uff09\uff0c\u628a\u5f00\u6e90\u6a21\u578b\u7684\u6027\u80fd\u62c9\u5230\u4e86\u65b0\u9ad8\u5ea6\u3002\u7531\u4e8e\u6a21\u578b\u53c2\u6570\u91cf\u5f88\u5927\uff0c\u5f88\u591a\u5f00\u53d1\u8005\u90fd\u5173\u5fc3\u4e00\u4e2a\u95ee\u9898\uff1a\u600e\u4e48\u63d0\u9ad8\u6a21\u578b\u7684\u63a8\u7406\u901f\u5ea6\uff1f<\/section>\n<section><\/section>\n<section>\u65f6\u9694\u624d\u4e24\u5929\uff0cLMSYS Org \u56e2\u961f\u5c31\u51fa\u624b\u4e86\uff0c\u63a8\u51fa\u4e86\u5168\u65b0\u7684 SGLang Runtime v0.2\u3002\u8fd9\u662f\u4e00\u4e2a\u7528\u4e8e LLM \u548c VLM \u7684\u901a\u7528\u670d\u52a1\u5f15\u64ce\u3002\u5728\u8fd0\u884c Llama 3.1 405B \u65f6\uff0c\u5b83\u7684\u541e\u5410\u91cf\u548c\u5ef6\u8fdf\u8868\u73b0\u90fd\u4f18\u4e8e vLLM \u548c TensorRT-LLM\u3002<\/section>\n<section><\/section>\n<section>\u5728\u67d0\u4e9b\u60c5\u51b5\u4e0b\uff08\u8fd0\u884c Llama \u7cfb\u5217\u6a21\u578b\uff09\uff0c\u5b83\u7684\u541e\u5410\u91cf\u751a\u81f3\u80fd\u8fbe\u5230 TensorRT-LLM \u7684 2.1 \u500d\uff0cvLLm \u7684 3.8 \u500d\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-812f407c5c3403d72926d81477d4d6a7.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-812f407c5c3403d72926d81477d4d6a7.png\"><\/a><\/section>\n<section>LMSYS Org \u56e2\u961f\u662f\u4e00\u4e2a\u7531\u52a0\u5dde\u5927\u5b66\u4f2f\u514b\u5229\u5206\u6821\u3001\u52a0\u5dde\u5927\u5b66\u5723\u5730\u4e9a\u54e5\u5206\u6821\u4ee5\u53ca\u5361\u5185\u57fa\u6885\u9686\u5927\u5b66\u7684\u5b66\u751f\u4e0e\u6559\u804c\u5458\u5de5\u5171\u540c\u7ec4\u5efa\u7684\u516c\u5f00\u6027\u8d28\u7684\u7814\u7a76\u56e2\u4f53\u3002\u4ed6\u4eec\u5f00\u53d1\u7684\u5927\u6a21\u578b\u8bc4\u6d4b\u5e73\u53f0 \u2014\u2014Chatbot Arena \u5df2\u7ecf\u6210\u4e3a\u68c0\u9a8c\u5927\u6a21\u578b\u80fd\u529b\u7684\u91cd\u8981\u5e73\u53f0\uff0c\u4e5f\u88ab\u8ba4\u4e3a\u662f\u4e00\u79cd\u76f8\u5bf9\u516c\u5e73\u7684\u8bc4\u6d4b\u65b9\u5f0f\u3002<\/section>\n<section><\/section>\n<section><a data-itemshowtype=\"0\" data-linktype=\"2\" href=\"http:\/\/mp.weixin.qq.com\/s?__biz=MzA3MzI4MjgzMw==&amp;mid=2650904838&amp;idx=4&amp;sn=81a9f09f54f1b89b98fdc3d6be11ce51&amp;chksm=84e45d78b393d46e38fed5d7e1952c9f465488b6953b63206a077f9d886c1104028b06c987c4&amp;scene=21#wechat_redirect\" target=\"_blank\">SGLang<\/a> \u662f\u8be5\u56e2\u961f\u5f00\u53d1\u7684\u4e00\u4e2a\u7528\u4e8e\u5927\u578b\u8bed\u8a00\u6a21\u578b\u548c\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u7684\u5feb\u901f\u670d\u52a1\u6846\u67b6\uff0c\u4e8e\u4eca\u5e74 1 \u6708\u4efd\u6b63\u5f0f\u63a8\u51fa\uff0c\u5728 GitHub \u4e0a\u5df2\u7ecf\u6536\u83b7\u4e86\u8d85\u8fc7 3k \u7684 star \u91cf\u3002<\/section>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-975e0cb67302b5347c0282e2dbf5481f.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-975e0cb67302b5347c0282e2dbf5481f.png\"><\/a><\/p>\n<section>\u8fd9\u6b21\u7684\u66f4\u65b0\u6548\u679c\u60ca\u8273\uff0c\u77e5\u540d AI \u7814\u7a76\u8005\u3001Lepton AI \u8054\u5408\u521b\u59cb\u4eba\u517c CEO \u8d3e\u626c\u6e05\u8bc4\u4ef7\u8bf4\u300c\u6211\u4e00\u76f4\u88ab\u6211\u7684\u535a\u58eb\u6bcd\u6821\u52a0\u5dde\u5927\u5b66\u4f2f\u514b\u5229\u5206\u6821\u60ca\u8273\uff0c\u56e0\u4e3a\u5b83\u4e0d\u65ad\u4ea4\u4ed8\u6700\u5148\u8fdb\u7684\u4eba\u5de5\u667a\u80fd\u548c\u7cfb\u7edf\u534f\u540c\u8bbe\u8ba1\u6210\u679c\u3002\u53bb\u5e74\u6211\u4eec\u770b\u5230\u4e86 SGLang \u7684\u4f7f\u7528\uff0c\u73b0\u5728\u5b83\u53d8\u5f97\u66f4\u597d\u4e86\u3002\u8feb\u4e0d\u53ca\u5f85\u5730\u60f3\u5728\u4ea7\u54c1\u4e2d\u90e8\u7f72\u5e76\u5c1d\u8bd5\u65b0\u7684 SGLang\uff01\u300d<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-26b34221aaea32469933ef80184e3e13.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-26b34221aaea32469933ef80184e3e13.png\"><\/a><\/section>\n<section>\u4e3a\u4ec0\u4e48 LMSYS Org \u8981\u5f00\u53d1\u5e76\u8fed\u4ee3 SGLang \u5462\uff1f\u4ed6\u4eec\u5728\u535a\u5ba2\u4e2d\u63d0\u5230\uff0c\u300c\u6211\u4eec\u5df2\u7ecf\u8fd0\u884c Chatbot Arena \u5e73\u53f0\u4e00\u5e74\u591a\uff0c\u4e3a\u6570\u767e\u4e07\u7528\u6237\u63d0\u4f9b\u670d\u52a1\u3002\u6211\u4eec\u6df1\u77e5\u9ad8\u6548\u670d\u52a1\u5bf9\u4eba\u5de5\u667a\u80fd\u4ea7\u54c1\u548c\u7814\u7a76\u7684\u91cd\u8981\u6027\u3002\u901a\u8fc7\u8fd0\u8425\u7ecf\u9a8c\u548c\u6df1\u5165\u7814\u7a76\uff0c\u6211\u4eec\u4e0d\u65ad\u589e\u5f3a\u5e95\u5c42\u670d\u52a1\u7cfb\u7edf\uff0c\u4ece\u9ad8\u7ea7\u591a\u6a21\u578b\u670d\u52a1\u6846\u67b6\u00a0FastChat \u5230\u9ad8\u6548\u670d\u52a1\u5f15\u64ce SGLang Runtime (SRT)\u3002\u300d<\/section>\n<section><\/section>\n<section>\u300c\u8fd9\u7bc7\u6587\u7ae0\u7684\u91cd\u70b9\u662f SGLang Runtime\uff0c\u5b83\u662f\u4e00\u4e2a\u7528\u4e8e LLM \u548c VLM \u7684\u901a\u7528\u670d\u52a1\u5f15\u64ce\u3002\u867d\u7136 TensorRT-LLM\u3001vLLM\u3001MLC-LLM \u548c Hugging Face TGI \u7b49\u73b0\u6709\u9009\u9879\u5404\u6709\u4f18\u70b9\uff0c\u4f46\u6211\u4eec\u53d1\u73b0\u5b83\u4eec\u6709\u65f6\u96be\u4ee5\u4f7f\u7528\u3001\u96be\u4ee5\u5b9a\u5236\u6216\u6027\u80fd\u4e0d\u4f73\u3002\u8fd9\u4fc3\u4f7f\u6211\u4eec\u5f00\u53d1\u4e86 SGLang v0.2\uff0c\u65e8\u5728\u521b\u5efa\u4e00\u4e2a\u4e0d\u4ec5\u7528\u6237\u53cb\u597d\u3001\u6613\u4e8e\u4fee\u6539\uff0c\u800c\u4e14\u6027\u80fd\u4e00\u6d41\u7684\u670d\u52a1\u5f15\u64ce\u3002\u300d<\/section>\n<section><\/section>\n<section>\u4e0e TensorRT-LLM \u548c vLLM \u76f8\u6bd4\uff0cSGLang Runtime \u5728\u5904\u7406\u4ece Llama-8B \u5230 Llama-405B \u7684\u6a21\u578b\u65f6\uff0c\u4ee5\u53ca\u5728 A100 \u548c H100 GPU \u4e0a\u4f7f\u7528 FP8 \u548c FP16 \u65f6\uff0c\u5728\u5728\u7ebf\u548c\u79bb\u7ebf\u573a\u666f\u4e0b\u90fd\u80fd\u6301\u7eed\u63d0\u4f9b\u5353\u8d8a\u6216\u6709\u7ade\u4e89\u529b\u7684\u6027\u80fd\u3002SGLang \u7684\u6027\u80fd\u59cb\u7ec8\u4f18\u4e8e vLLM\uff0c\u5728 Llama-70B \u4e0a\u7684\u541e\u5410\u91cf\u6700\u9ad8\u662f\u524d\u8005\u7684 3.8 \u500d\u3002\u5b83\u8fd8\u7ecf\u5e38\u4e0e TensorRT-LLM \u4e0d\u76f8\u4e0a\u4e0b\uff0c\u751a\u81f3\u8d85\u8fc7 TensorRT-LLM\uff0c\u5728 Llama-405B \u4e0a\u7684\u541e\u5410\u91cf\u6700\u9ad8\u662f\u524d\u8005\u7684 2.1 \u500d\u3002\u66f4\u91cd\u8981\u7684\u662f\uff0cSGLang \u662f\u5b8c\u5168\u5f00\u6e90\u7684\uff0c\u7531\u7eaf Python \u7f16\u5199\uff0c\u6838\u5fc3\u8c03\u5ea6\u5668\u53ea\u7528\u4e86\u4e0d\u5230 4K \u884c\u4ee3\u7801\u5c31\u5b9e\u73b0\u4e86\u3002<\/section>\n<section><\/section>\n<section>SGLang \u662f\u4e00\u4e2a\u5f00\u6e90\u9879\u76ee\uff0c\u91c7\u7528 Apache 2.0 \u8bb8\u53ef\u6388\u6743\u3002\u5b83\u5df2\u88ab LMSYS Chatbot Arena \u7528\u4e8e\u652f\u6301\u90e8\u5206\u6a21\u578b\u3001Databricks\u3001\u51e0\u5bb6\u521d\u521b\u516c\u53f8\u548c\u7814\u7a76\u673a\u6784\uff0c\u4ea7\u751f\u4e86\u6570\u4e07\u4ebf token\uff0c\u5b9e\u73b0\u4e86\u66f4\u5feb\u7684\u8fed\u4ee3\u3002<\/section>\n<section><\/section>\n<section>\u4ee5\u4e0b\u662f\u51e0\u4e2a\u6846\u67b6\u7684\u5bf9\u6bd4\u5b9e\u9a8c\u8bbe\u7f6e\u548c\u7ed3\u679c\u3002<\/section>\n<section><\/section>\n<section><strong>\u57fa\u51c6\u8bbe\u7f6e<\/strong><\/section>\n<section><\/section>\n<section>\u7814\u7a76\u8005\u5bf9\u79bb\u7ebf\u548c\u5728\u7ebf\u7528\u4f8b\u8fdb\u884c\u57fa\u51c6\u6d4b\u8bd5\uff1a<\/section>\n<section><\/section>\n<section>\u79bb\u7ebf\uff1a\u4ed6\u4eec\u4e00\u6b21\u53d1\u9001 2K \u5230 3K \u4e2a\u8bf7\u6c42\uff0c\u6d4b\u91cf\u8f93\u51fa\u541e\u5410\u91cf\uff08token \/ \u79d2\uff09\uff0c\u5373\u8f93\u51fa token \u6570\u9664\u4ee5\u603b\u6301\u7eed\u65f6\u95f4\u3002\u4ed6\u4eec\u6d4b\u8bd5\u7684\u5408\u6210\u6570\u636e\u96c6\u6765\u81ea\u00a0ShareGPT \u6570\u636e\u96c6\u3002\u4f8b\u5982\uff0cI-512-O-1024 \u8868\u793a\u5e73\u5747\u8f93\u5165 512 \u4e2a token\u3001\u5e73\u5747\u8f93\u51fa 1024 \u4e2a token \u7684\u6570\u636e\u96c6\u3002\u4e94\u4e2a\u6d4b\u8bd5\u6570\u636e\u96c6\u5206\u522b\u4e3a\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u6570\u636e\u96c6 1\uff1aI-243-O-770\uff1b<\/section>\n<\/li>\n<li>\n<section>\u6570\u636e\u96c6 2\uff1aI-295-O-770\uff1b<\/section>\n<\/li>\n<li>\n<section>\u6570\u636e\u96c6 3\uff1aI-243-O-386\uff1b<\/section>\n<\/li>\n<li>\n<section>\u6570\u636e\u96c6 4\uff1aI-295-O-386\uff1b<\/section>\n<\/li>\n<li>\n<section>\u6570\u636e\u96c6 5\uff1aI-221-O-201\u3002<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u5728\u7ebf\uff1a\u4ed6\u4eec\u4ee5\u6bcf\u79d2 1 \u5230 16 \u4e2a\u8bf7\u6c42 (RPS) \u7684\u901f\u7387\u53d1\u9001\u8bf7\u6c42\uff0c\u6d4b\u91cf\u7aef\u5230\u7aef\u5ef6\u8fdf\u7684\u4e2d\u4f4d\u6570\u3002\u4ed6\u4eec\u4f7f\u7528\u5408\u6210\u6570\u636e\u96c6 I-292-O-579\u3002<\/section>\n<section><\/section>\n<section>\u4ed6\u4eec\u4f7f\u7528 vLLM 0.5.2\uff08\u5e26\u9ed8\u8ba4\u53c2\u6570\uff09\u548c TensorRT-LLM\uff08\u5e26\u63a8\u8350\u53c2\u6570\u548c\u8c03\u6574\u540e\u7684\u6279\u5927\u5c0f\uff09\u3002\u6240\u6709\u5f15\u64ce\u90fd\u5173\u95ed\u4e86\u524d\u7f00\u7f13\u5b58\u3002\u76ee\u7684\u662f\u5728\u6ca1\u6709\u4efb\u4f55\u9644\u52a0\u529f\u80fd\uff08\u5982\u63a8\u6d4b\u89e3\u7801\u6216\u7f13\u5b58\uff09\u7684\u60c5\u51b5\u4e0b\uff0c\u5bf9\u57fa\u672c\u6027\u80fd\u8fdb\u884c\u57fa\u51c6\u6d4b\u8bd5\u3002\u4ed6\u4eec\u4f7f\u7528\u4e0e OpenAI \u517c\u5bb9\u7684 API \u5bf9 SGLang \u548c vLLM \u8fdb\u884c\u57fa\u51c6\u6d4b\u8bd5\uff0c\u5e76\u4f7f\u7528 Triton \u63a5\u53e3\u5bf9 TensorRT-LLM \u8fdb\u884c\u57fa\u51c6\u6d4b\u8bd5\u3002<\/section>\n<section><\/section>\n<section><strong>Llama-8B \u5728\u4e00\u4e2a A100 \u4e0a\u8fd0\u884c\uff08bf16\uff09<\/strong><\/section>\n<section><\/section>\n<section>\u7814\u7a76\u8005\u4ece\u5c0f\u578b\u6a21\u578b Llama-8B \u5f00\u59cb\u6d4b\u8bd5\u3002\u4e0b\u56fe\u663e\u793a\u4e86\u6bcf\u4e2a\u5f15\u64ce\u5728\u4e94\u4e2a\u4e0d\u540c\u6570\u636e\u96c6\u7684\u79bb\u7ebf\u8bbe\u7f6e\u4e0b\u6240\u80fd\u8fbe\u5230\u7684\u6700\u5927\u8f93\u51fa\u541e\u5410\u91cf\u3002TensorRT-LLM \u548c SGLang \u90fd\u80fd\u8fbe\u5230\u6bcf\u79d2\u7ea6 4000 \u4e2a token \u7684\u541e\u5410\u91cf\uff0c\u800c vLLM \u5219\u7a0d\u900a\u4e00\u7b79\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-0d421b20d64bdfecfb4aa998c4c32eca.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-0d421b20d64bdfecfb4aa998c4c32eca.png\"><\/a><\/section>\n<section>\u4e0b\u9762\u7684\u5728\u7ebf\u57fa\u51c6\u56fe\u663e\u793a\u4e86\u4e0e\u79bb\u7ebf\u60c5\u51b5\u7c7b\u4f3c\u7684\u8d8b\u52bf\u3002TensorRT-LLM \u548c SGLang \u7684\u6027\u80fd\u76f8\u5f53\uff0c\u53ef\u4ee5\u4fdd\u6301 RPS &gt; 10\uff0c\u800c vLLM \u7684\u5ef6\u8fdf\u5728\u8bf7\u6c42\u7387\u8f83\u9ad8\u65f6\u663e\u8457\u589e\u52a0\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-0c92dd044356c119f4f1281311461c2d.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-0c92dd044356c119f4f1281311461c2d.png\"><\/a><\/section>\n<section><strong>Llama-70B \u5728 8 \u4e2a A100 \u4e0a\u8fd0\u884c\uff08bf16\uff09<\/strong><\/section>\n<section><\/section>\n<section>\u81f3\u4e8e\u5728 8 \u4e2a GPU \u4e0a\u8fdb\u884c\u5f20\u91cf\u5e76\u884c\u7684\u8f83\u5927\u578b Llama-70B \u6a21\u578b\uff0c\u8d8b\u52bf\u4e0e 8B \u76f8\u4f3c\u3002\u5728\u4e0b\u9762\u7684\u79bb\u7ebf\u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0cTensorRT-LLM \u548c SGLang \u90fd\u80fd\u8fbe\u5230\u5f88\u9ad8\u7684\u541e\u5410\u91cf\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-428631329bfe9b13c1273c3d16955ff2.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-428631329bfe9b13c1273c3d16955ff2.png\"><\/a><\/section>\n<section>\u5728\u4e0b\u56fe\u7684\u5728\u7ebf\u7ed3\u679c\u4e2d\uff0cTensorRT-LLM \u51ed\u501f\u9ad8\u6548\u7684\u5185\u6838\u5b9e\u73b0\u548c\u8fd0\u884c\u65f6\u95f4\uff0c\u663e\u793a\u51fa\u8f83\u4f4e\u7684\u5ef6\u8fdf\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-3d39e38ad2092edc2be17b20f3533c95.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-3d39e38ad2092edc2be17b20f3533c95.png\"><\/a><\/section>\n<section><strong>Llama-70B \u5728 8 \u4e2a H100 \u4e0a\u8fd0\u884c\uff08fp8\uff09<\/strong><\/section>\n<section><\/section>\n<section>\u73b0\u5728\u6765\u6d4b\u8bd5 FP8 \u6027\u80fd\u3002vLLM \u548c SGLang \u90fd\u4f7f\u7528\u4e86 CUTLASS \u7684 FP8 \u5185\u6838\u3002\u5728\u79bb\u7ebf\u8bbe\u7f6e\u4e2d\uff0cSGLang \u7684\u6279\u5904\u7406\u8c03\u5ea6\u5668\u975e\u5e38\u9ad8\u6548\uff0c\u53ef\u4ee5\u968f\u7740\u6279\u5904\u7406\u89c4\u6a21\u7684\u589e\u5927\u800c\u7ee7\u7eed\u6269\u5c55\u541e\u5410\u91cf\uff0c\u5728\u8fd9\u79cd\u60c5\u51b5\u4e0b\u5b9e\u73b0\u4e86\u6700\u9ad8\u541e\u5410\u91cf\u3002\u5176\u4ed6\u7cfb\u7edf\u5219\u7531\u4e8e OOM\u3001\u7f3a\u5c11\u5927\u91cf\u624b\u52a8\u8c03\u6574\u6216\u5b58\u5728\u5176\u4ed6\u5f00\u9500\u800c\u65e0\u6cd5\u6269\u5c55\u541e\u5410\u91cf\u6216\u6279\u5927\u5c0f\u3002\u5728\u7ebf\u60c5\u51b5\u4e0b\u4e5f\u662f\u5982\u6b64\uff0cSGLang \u548c TensorRT \u7684\u4e2d\u4f4d\u5ef6\u8fdf\u76f8\u4f3c\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-17d58ee2b284a485533ebb70d05a8938.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-17d58ee2b284a485533ebb70d05a8938.png\"><\/a><\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-4ccb3e55fddfab08fa9ab68d40871793.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-4ccb3e55fddfab08fa9ab68d40871793.png\"><\/a><\/section>\n<section><strong>Llama-405B \u5728 8 \u4e2a H100 \u4e0a\u8fd0\u884c\uff08fp8\uff09<\/strong><\/section>\n<section><\/section>\n<section>\u6700\u540e\uff0c\u7814\u7a76\u8005\u5728\u6700\u5927\u7684 405B \u6a21\u578b\u4e0a\u5bf9\u5404\u79cd\u65b9\u6cd5\u7684\u6027\u80fd\u8fdb\u884c\u4e86\u57fa\u51c6\u6d4b\u8bd5\u3002\u7531\u4e8e\u6a21\u578b\u8f83\u5927\uff0c\u5927\u90e8\u5206\u65f6\u95f4\u90fd\u82b1\u5728\u4e86 GPU \u5185\u6838\u4e0a\u3002\u4e0d\u540c\u6846\u67b6\u4e4b\u95f4\u7684\u5dee\u8ddd\u7f29\u5c0f\u4e86\u3002TensorRT-LLM \u6027\u80fd\u4e0d\u4f73\u7684\u539f\u56e0\u53ef\u80fd\u662f 405B \u6a21\u578b\u521a\u521a\u95ee\u4e16\uff0c\u800c\u56fe\u4e2d\u4f7f\u7528\u7684\u7248\u672c\u5c1a\u672a\u96c6\u6210\u4e00\u4e9b\u6700\u65b0\u4f18\u5316\u3002\u5728\u5728\u7ebf\u548c\u79bb\u7ebf\u60c5\u51b5\u4e0b\uff0cSGLang \u7684\u6027\u80fd\u90fd\u662f\u6700\u597d\u7684\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-7d5e7ffbac907bede02f4691d8e010a4.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-7d5e7ffbac907bede02f4691d8e010a4.png\"><\/a><\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-edb8ce7e4f4f6cee214a6cde0b56e095.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-edb8ce7e4f4f6cee214a6cde0b56e095.png\"><\/a><\/section>\n<section><strong>SGLang \u6982\u89c8<\/strong><\/section>\n<section><\/section>\n<section>SGLang \u662f\u5927\u578b\u8bed\u8a00\u6a21\u578b\u548c\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u7684\u670d\u52a1\u6846\u67b6\u3002\u5b83\u57fa\u4e8e\u5e76\u589e\u5f3a\u4e86\u591a\u4e2a\u5f00\u6e90 LLM \u670d\u52a1\u5f15\u64ce\uff08\u5305\u62ec LightLLM\u3001vLLM \u548c Guidance\uff09\u7684\u8bb8\u591a\u4f18\u79c0\u8bbe\u8ba1\u3002\u5b83\u5229\u7528\u4e86\u6765\u81ea FlashInfer \u7684\u9ad8\u6027\u80fd\u6ce8\u610f\u529b CUDA \u5185\u6838\uff0c\u5e76\u96c6\u6210\u4e86\u53d7 gpt-fast \u542f\u53d1\u7684 torch.compile\u3002<\/section>\n<section><\/section>\n<section>\u6b64\u5916\uff0c\u7814\u7a76\u8005\u8fd8\u5f15\u5165\u4e86\u4e00\u4e9b\u521b\u65b0\u6280\u672f\uff0c\u5982\u7528\u4e8e\u81ea\u52a8 KV \u7f13\u5b58\u91cd\u7528\u7684 RadixAttention \u548c\u7528\u4e8e\u5feb\u901f\u7ea6\u675f\u89e3\u7801\u7684\u538b\u7f29\u72b6\u6001\u673a\u3002SGLang \u4ee5\u5176\u5b8c\u5168\u7528 Python \u5b9e\u73b0\u7684\u9ad8\u6548\u6279\u5904\u7406\u8c03\u5ea6\u5668\u800c\u95fb\u540d\u3002\u4e3a\u4e86\u8fdb\u884c\u516c\u5e73\u6bd4\u8f83\uff0c\u672c\u535a\u5ba2\u6d4b\u8bd5\u4e86\u8fd9\u4e9b\u670d\u52a1\u5f15\u64ce\u5728\u5173\u95ed\u7279\u5b9a\u573a\u666f\u6216\u5de5\u4f5c\u8d1f\u8f7d\u4f18\u5316\uff08\u5982\u524d\u7f00\u7f13\u5b58\u548c\u63a8\u6d4b\u89e3\u7801\uff09\u540e\u7684\u57fa\u672c\u6027\u80fd\u3002SGLang \u7684\u63d0\u901f\u662f\u901a\u8fc7\u9002\u5f53\u7684\u5de5\u7a0b\u8bbe\u8ba1\u5b9e\u73b0\u7684\u3002SGLang \u57fa\u4e8e Python \u7684\u9ad8\u6548\u6279\u5904\u7406\u8c03\u5ea6\u5668\u5177\u6709\u826f\u597d\u7684\u6269\u5c55\u6027\uff0c\u901a\u5e38\u53ef\u4e0e\u4f7f\u7528 C++ \u6784\u5efa\u7684\u95ed\u6e90\u5b9e\u73b0\u76f8\u5ab2\u7f8e\uff0c\u751a\u81f3\u66f4\u80dc\u4e00\u7b79\u3002<\/section>\n<section><\/section>\n<section>\u8868 1 \u6bd4\u8f83\u4e86 SGLang\u3001TensorRT-LLM \u548c vLLM \u7684\u5404\u4e2a\u65b9\u9762\u3002\u5728\u6027\u80fd\u65b9\u9762\uff0cSGLang \u548c TensorRT-LLM \u90fd\u975e\u5e38\u51fa\u8272\u3002\u5728\u53ef\u7528\u6027\u548c\u53ef\u5b9a\u5236\u6027\u65b9\u9762\uff0cSGLang \u7684\u8f7b\u91cf\u7ea7\u548c\u6a21\u5757\u5316\u5185\u6838\u4f7f\u5176\u6613\u4e8e\u5b9a\u5236\uff0c\u800c TensorRT-LLM \u590d\u6742\u7684 C++ \u6280\u672f\u6808\u548c\u8bbe\u7f6e\u8bf4\u660e\u4f7f\u5176\u66f4\u96be\u4f7f\u7528\u548c\u4fee\u6539\u3002SGLang \u7684\u6e90\u4ee3\u7801\u5b8c\u5168\u5f00\u6e90\uff0c\u800c TensorRT-LLM \u4ec5\u90e8\u5206\u5f00\u6e90\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0cvLLM \u7684 CPU \u8c03\u5ea6\u5f00\u9500\u8f83\u9ad8\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-4f9dd7977eebec78ccd51a7416610d56.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-4f9dd7977eebec78ccd51a7416610d56.png\"><\/a><\/section>\n<section>\u7814\u7a76\u8005\u8fd8\u8868\u793a\uff0c\u672a\u6765\u4ed6\u4eec\u8fd8\u5c06\u5f00\u53d1\u957f\u4e0a\u4e0b\u6587\u548c MoE \u4f18\u5316\u7b49\u65b0\u529f\u80fd\u3002<\/section>\n<section><\/section>\n<section><strong>\u4f7f\u7528\u65b9\u6cd5<\/strong><\/section>\n<section><\/section>\n<section>\u4f60\u53ef\u4ee5\u6309\u7167\u4ee5\u4e0b\u6b65\u9aa4\u8f7b\u677e\u670d\u52a1 Llama \u6a21\u578b\uff1a \u00a0<\/section>\n<section><\/section>\n<section>1\u3001\u4f7f\u7528 pip\u3001\u6e90\u4ee3\u7801\u6216 Docker \u5b89\u88c5 SGLang\uff1ahttps:\/\/github.com\/sgl-project\/sglang\/tree\/main?tab=readme-ov-file#install<\/section>\n<section><\/section>\n<section>2\u3001\u542f\u52a8\u670d\u52a1\u5668\uff1a<\/section>\n<section>\n<pre data-lang=\"properties\"><code># Llama 8B<\/code>\r\n<code>python -m sglang.launch_server --model-path meta-llama\/Meta-Llama-3.1-8B-Instruct<\/code><code>\r\n<\/code><code>\r\n<\/code><code># Llama 405B<\/code>\r\n<code>python -m sglang.launch_server --model-path meta-llama\/Meta-Llama-3.1-405B-Instruct-FP8 --tp 8<\/code><\/pre>\n<\/section>\n<section><\/section>\n<section>3\u3001\u4f7f\u7528 OpenAI \u517c\u5bb9\u7684 API \u53d1\u9001\u8bf7\u6c42\uff1a<\/section>\n<section>\n<pre data-lang=\"swift\"><code>curl http:\/\/localhost:30000\/v1\/completions <\/code>\r\n<code>  -H \"Content-Type: application\/json\" <\/code>\r\n<code>  -d '{<\/code>\r\n<code>    \"model\": \"default\",<\/code>\r\n<code>    \"prompt\": \"Say this is a test\",<\/code>\r\n<code>    \"max_tokens\": 7,<\/code>\r\n<code>    \"temperature\": 0<\/code>\r\n<code>\u00a0 }'<\/code><\/pre>\n<\/section>\n<section><\/section>\n<section>4\u3001\u8fd0\u884c\u57fa\u51c6\uff1a \u00a0<\/section>\n<section>\n<pre data-lang=\"css\"><code>python3 -m sglang.bench_serving --backend sglang --num-prompts 1000<\/code><\/pre>\n<\/section>\n<section><\/section>\n<section><strong>\u9644\u5f55\uff1a\u8be6\u7ec6\u7684\u57fa\u51c6\u8bbe\u7f6e<\/strong><\/section>\n<section><\/section>\n<section>\u91cd\u73b0\u57fa\u51c6\u7684\u8bf4\u660e\u4f4d\u4e8e sglang\/benchmark\/blog_v0_2\u3002<\/section>\n<section><\/section>\n<section>\u5bf9\u4e8e\u6240\u6709\u57fa\u51c6\u6d4b\u8bd5\uff0c\u7814\u7a76\u8005\u90fd\u8bbe\u7f6e\u4e86 ignore_eos \u6216 min_length\/end_id \u4ee5\u786e\u4fdd\u6bcf\u4e2a\u5f15\u64ce\u8f93\u51fa\u76f8\u540c\u6570\u91cf\u7684 token\u3002\u4ed6\u4eec\u66fe\u5c1d\u8bd5\u4f7f\u7528 vLLM 0.5.3.post1\uff0c\u4f46\u5b83\u5728\u9ad8\u8d1f\u8f7d\u60c5\u51b5\u4e0b\u7ecf\u5e38\u5d29\u6e83\uff0c\u4e0e\u90e8\u5206\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u7684 vLLM 0.5.2 \u76f8\u6bd4\uff0cvLLM 0.5.3.post1 \u6027\u80fd\u4f3c\u4e4e\u5dee\u4e0d\u591a\u751a\u81f3\u66f4\u5dee\u3002\u56e0\u6b64\uff0c\u4ed6\u4eec\u62a5\u544a\u7684\u662f vLLM 0.5.2 \u7684\u7ed3\u679c\u3002\u867d\u7136\u4ed6\u4eec\u77e5\u9053\u4e0d\u540c\u7684\u670d\u52a1\u5668\u914d\u7f6e\u4f1a\u5bf9\u670d\u52a1\u6027\u80fd\u4ea7\u751f\u91cd\u5927\u5f71\u54cd\uff0c\u4f46\u4ed6\u4eec\u4e3b\u8981\u4f7f\u7528\u6bcf\u4e2a\u5f15\u64ce\u7684\u9ed8\u8ba4\u53c2\u6570\u6765\u6a21\u62df\u666e\u901a\u7528\u6237\u7684\u60c5\u51b5\u3002<\/section>\n<section><\/section>\n<section>\u5bf9\u4e8e 8B \u548c 70B \u6a21\u578b\uff0c\u4ed6\u4eec\u4f7f\u7528 meta-llama\/Meta-Llama-3-8B-Instruct \u548c meta-llama\/Meta-Llama-3-70B-Instruct bf16 \u68c0\u67e5\u70b9\uff0c\u4ee5\u53ca neuralmagic\/Meta-Llama-3-70B-Instruct-FP8 fp8 \u68c0\u67e5\u70b9\u3002\u5bf9\u4e8e 405B \u6a21\u578b\uff0c\u4ed6\u4eec\u5728\u6240\u6709\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u90fd\u4f7f\u7528\u4e86\u865a\u62df\u6743\u91cd\u3002\u7531\u4e8e TensorRT-LLM \u6700\u65b0\u56fe\u50cf r24.06 \u4e0d\u652f\u6301\u5b98\u65b9 meta-llama\/Meta-Llama-3.1-405B-FP8 \u68c0\u67e5\u70b9\u4e2d\u7684 fbgemm_fp8 \u91cf\u5316\uff0c\u4ed6\u4eec\u5728\u6240\u6709\u6846\u67b6\u4e2d\u90fd\u4f7f\u7528\u4e86\u6bcf\u5c42 fp8 \u91cf\u5316\uff0c\u5e76\u5bf9\u9664 lm_head \u4ee5\u5916\u7684\u6240\u6709\u5c42\u8fdb\u884c\u4e86\u91cf\u5316\u3002\u4ed6\u4eec\u76f8\u4fe1\u8fd9\u6837\u53ef\u4ee5\u5bf9\u6240\u6709\u5f15\u64ce\u8fdb\u884c\u516c\u5e73\u7684\u6bd4\u8f83\u3002A100 \u548c H100 GPU \u4e3a 80GB SXM \u7248\u672c\u3002<\/section>\n<section><\/section>\n<section><em><sup>\u53c2\u8003\u94fe\u63a5\uff1ahttps:\/\/lmsys.org\/blog\/2024-07-25-sglang-llama3\/<\/sup><\/em><\/section>\n<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2024-07-27-3\" target=\"_blank\">\u8d3e\u626c\u6e05\u70b9\u8d5e\uff1a3K star\u91cf\u7684SGLang\u4e0a\u65b0\uff0c\u52a0\u901fLlama 405B\u63a8\u7406\u79d2\u6740vLLM\u3001TensorRT-LLM<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u8d3e\u626c\u6e05\u70b9\u8d5e\uff1a3K st [&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":[68],"class_list":["post-34772","post","type-post","status-publish","format-standard","hentry","category-news","tag-68"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v27.9 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>\u8d3e\u626c\u6e05\u70b9\u8d5e\uff1a3K star\u91cf\u7684SGLang\u4e0a\u65b0\uff0c\u52a0\u901fLlama 405B\u63a8\u7406\u79d2\u6740vLLM\u3001TensorRT-LLM - \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=34772\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=34772#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=34772\"},\"author\":{\"name\":\"AI\u5c0f\u52a9\u624b\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\"},\"headline\":\"\u8d3e\u626c\u6e05\u70b9\u8d5e\uff1a3K star\u91cf\u7684SGLang\u4e0a\u65b0\uff0c\u52a0\u901fLlama 405B\u63a8\u7406\u79d2\u6740vLLM\u3001TensorRT-LLM\",\"datePublished\":\"2024-12-12T10:57:42+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=34772\"},\"wordCount\":290,\"commentCount\":0,\"image\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=34772#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/07\\\/frc-812f407c5c3403d72926d81477d4d6a7.png\",\"keywords\":[\"\u5927\u6a21\u578b\"],\"articleSection\":[\"\u884c\u4e1a\u8d44\u8baf\"],\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/17aitech.com\\\/?p=34772#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=34772\",\"url\":\"https:\\\/\\\/17aitech.com\\\/?p=34772\",\"name\":\"\u8d3e\u626c\u6e05\u70b9\u8d5e\uff1a3K star\u91cf\u7684SGLang\u4e0a\u65b0\uff0c\u52a0\u901fLlama 405B\u63a8\u7406\u79d2\u6740vLLM\u3001TensorRT-LLM - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=34772#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=34772#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/07\\\/frc-812f407c5c3403d72926d81477d4d6a7.png\",\"datePublished\":\"2024-12-12T10:57:42+00:00\",\"author\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=34772#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/17aitech.com\\\/?p=34772\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=34772#primaryimage\",\"url\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/07\\\/frc-812f407c5c3403d72926d81477d4d6a7.png\",\"contentUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/07\\\/frc-812f407c5c3403d72926d81477d4d6a7.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=34772#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\\\/\\\/17aitech.com\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u8d3e\u626c\u6e05\u70b9\u8d5e\uff1a3K star\u91cf\u7684SGLang\u4e0a\u65b0\uff0c\u52a0\u901fLlama 405B\u63a8\u7406\u79d2\u6740vLLM\u3001TensorRT-LLM\"}]},{\"@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\":\"\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/04\\\/robot_3.png\",\"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":"\u8d3e\u626c\u6e05\u70b9\u8d5e\uff1a3K star\u91cf\u7684SGLang\u4e0a\u65b0\uff0c\u52a0\u901fLlama 405B\u63a8\u7406\u79d2\u6740vLLM\u3001TensorRT-LLM - \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=34772","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/17aitech.com\/?p=34772#article","isPartOf":{"@id":"https:\/\/17aitech.com\/?p=34772"},"author":{"name":"AI\u5c0f\u52a9\u624b","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"headline":"\u8d3e\u626c\u6e05\u70b9\u8d5e\uff1a3K star\u91cf\u7684SGLang\u4e0a\u65b0\uff0c\u52a0\u901fLlama 405B\u63a8\u7406\u79d2\u6740vLLM\u3001TensorRT-LLM","datePublished":"2024-12-12T10:57:42+00:00","mainEntityOfPage":{"@id":"https:\/\/17aitech.com\/?p=34772"},"wordCount":290,"commentCount":0,"image":{"@id":"https:\/\/17aitech.com\/?p=34772#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-812f407c5c3403d72926d81477d4d6a7.png","keywords":["\u5927\u6a21\u578b"],"articleSection":["\u884c\u4e1a\u8d44\u8baf"],"inLanguage":"zh-Hans","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/17aitech.com\/?p=34772#respond"]}]},{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=34772","url":"https:\/\/17aitech.com\/?p=34772","name":"\u8d3e\u626c\u6e05\u70b9\u8d5e\uff1a3K star\u91cf\u7684SGLang\u4e0a\u65b0\uff0c\u52a0\u901fLlama 405B\u63a8\u7406\u79d2\u6740vLLM\u3001TensorRT-LLM - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=34772#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=34772#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-812f407c5c3403d72926d81477d4d6a7.png","datePublished":"2024-12-12T10:57:42+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=34772#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=34772"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=34772#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-812f407c5c3403d72926d81477d4d6a7.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/frc-812f407c5c3403d72926d81477d4d6a7.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=34772#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"\u8d3e\u626c\u6e05\u70b9\u8d5e\uff1a3K star\u91cf\u7684SGLang\u4e0a\u65b0\uff0c\u52a0\u901fLlama 405B\u63a8\u7406\u79d2\u6740vLLM\u3001TensorRT-LLM"}]},{"@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":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","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\/34772","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=34772"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/34772\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=34772"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=34772"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=34772"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}