{"id":25551,"date":"2025-04-23T10:00:49","date_gmt":"2025-04-23T02:00:49","guid":{"rendered":"https:\/\/aif.amtbbs.org\/?p=25551"},"modified":"2025-04-23T10:00:49","modified_gmt":"2025-04-23T02:00:49","slug":"%e8%bf%9eclaude-3-5%e9%83%bd%e8%b4%a5%e4%b8%8b%e9%98%b5%e6%9d%a5%ef%bc%8c%e5%a4%a7%e8%af%ad%e8%a8%80%e6%a8%a1%e5%9e%8b%e8%83%bd%e5%90%a6%e5%ae%9a%e4%bd%8d%e8%bd%af%e4%bb%b6%e6%9c%8d%e5%8a%a1%e7%9a%84","status":"publish","type":"post","link":"https:\/\/aif.amtbbs.org\/index.php\/2025\/04\/23\/25551\/","title":{"rendered":"\u8fdeClaude 3.5\u90fd\u8d25\u4e0b\u9635\u6765\uff0c\u5927\u8bed\u8a00\u6a21\u578b\u80fd\u5426\u5b9a\u4f4d\u8f6f\u4ef6\u670d\u52a1\u7684\u6545\u969c\u6839\u56e0\uff1f"},"content":{"rendered":"
\n

\"\"<\/p>\n<\/div>\n

<\/div>\n
\u5fae\u8f6f DKI \u56e2\u961f\u3001\u9999\u6e2f\u4e2d\u6587\u5927\u5b66\uff08\u6df1\u5733\uff09\u8d3a\u54c1\u5609\u6559\u6388\u56e2\u961f\u4e0e\u6e05\u534e\u5927\u5b66\u88f4\u4e39\u6559\u6388\u5171\u540c\u63d0\u51fa\u4e86\u5f53\u524d\u9996\u4e2a\u516c\u5f00\u7684\u3001\u7528\u4e8e\u8bc4\u4f30 LLM \u6839\u56e0\u5206\u6790\u80fd\u529b\u7684\u57fa\u51c6\u8bc4\u4f30\u96c6 \u2014\u2014OpenRCA\u3002<\/div>\n
\n
\n

\u8bba\u6587\u7684\u7b2c\u4e00\u4f5c\u8005\u662f\u9999\u6e2f\u4e2d\u6587\u5927\u5b66\uff08\u6df1\u5733\uff09\u6570\u636e\u79d1\u5b66\u5b66\u9662\u4e09\u5e74\u7ea7\u535a\u58eb\u751f\u5f90\u4fca\u6770\u9f99\uff0c\u6307\u5bfc\u8001\u5e08\u4e3a\u9999\u6e2f\u4e2d\u6587\u5927\u5b66\uff08\u6df1\u5733\uff09\u6570\u636e\u79d1\u5b66\u5b66\u9662\u7684\u8d3a\u54c1\u5609\u6559\u6388\u548c\u5fae\u8f6f\u4e3b\u7ba1\u7814\u7a76\u5458\u4f55\u4e16\u6797\u535a\u58eb\u3002\u8d3a\u54c1\u5609\u8001\u5e08\u56e2\u961f\u7684\u7814\u7a76\u91cd\u70b9\u662f\u8f6f\u4ef6\u5de5\u7a0b\u3001LLM for DevOps\u3001\u5927\u6a21\u578b\u5b89\u5168\u3002<\/p>\n

\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u8fd1\u671f\u5728\u8f6f\u4ef6\u5de5\u7a0b\u9886\u57df\u53d6\u5f97\u4e86\u663e\u8457\u8fdb\u5c55\uff0c\u50ac\u751f\u4e86 MetaGPT\u3001SWE-agent\u3001OpenDevin\u3001Copilot \u548c Cursor \u7b49\u5927\u91cf\u7814\u7a76\u6210\u679c\u4e0e\u5b9e\u9645\u5e94\u7528\uff0c\u6df1\u523b\u5f71\u54cd\u7740\u8f6f\u4ef6\u5f00\u53d1\u7684\u65b9\u6cd5\u8bba\u548c\u5b9e\u8df5\u3002\u73b0\u6709\u7814\u7a76\u4e3b\u8981\u805a\u7126\u4e8e\u8f6f\u4ef6\u5f00\u53d1\u751f\u547d\u5468\u671f\uff08SDLC\uff09\u7684\u65e9\u671f\u9636\u6bb5\u7684\u4efb\u52a1\uff0c\u5982\u4ee3\u7801\u751f\u6210\uff08LiveCodeBench\uff09, \u7a0b\u5e8f\u4fee\u590d\uff08SWE-Bench\uff09\uff0c\u6d4b\u8bd5\u751f\u6210\uff08SWT-Bench\uff09\u7b49\u7b49\u3002\u7136\u800c\uff0c\u8fd9\u4e9b\u7814\u7a76\u5f80\u5f80\u5ffd\u89c6\u4e86\u8f6f\u4ef6\u90e8\u7f72\u540e\u7684\u8fd0\u7ef4\u9636\u6bb5\u3002\u5728\u5b9e\u9645\u751f\u4ea7\u73af\u5883\u4e2d\uff0c\u7ebf\u4e0a\u8f6f\u4ef6\u7684\u6545\u969c\u53ef\u80fd\u5bfc\u81f4\u670d\u52a1\u63d0\u4f9b\u5546\u906d\u53d7\u6570\u5341\u4ebf\u7f8e\u5143\u7684\u635f\u5931\uff0c\u8fd9\u51f8\u663e\u4e86\u5728\u6839\u56e0\u5206\u6790\uff08RCA\uff09\u9886\u57df\u5f00\u53d1\u66f4\u6709\u6548\u89e3\u51b3\u65b9\u6848\u7684\u8feb\u5207\u9700\u6c42\u3002<\/p>\n

\u4e3a\u4e86\u63a2\u7d22 LLM \u5728\u8fd9\u4e00\u9886\u57df\u7684\u53ef\u884c\u6027\uff0c\u5b66\u672f\u754c\u548c\u5de5\u4e1a\u754c\u7684\u8bb8\u591a\u7814\u7a76\u8005\u5df2\u7ecf\u5f00\u59cb\u8fdb\u884c\u76f8\u5173\u7814\u7a76\u3002\u7136\u800c\uff0c\u53d7\u9650\u4e8e\u8fd0\u7ef4\u6570\u636e\u7684\u9690\u79c1\u6027\u4ee5\u53ca\u4f01\u4e1a\u7cfb\u7edf\u7684\u5dee\u5f02\u6027\uff0c\u76ee\u524d\u57fa\u4e8e LLM \u7684\u6839\u56e0\u5206\u6790\u7814\u7a76\u7f3a\u4e4f\u7edf\u4e00\u4e14\u6e05\u6670\u7684\u4efb\u52a1\u5efa\u6a21\uff0c\u4e5f\u6ca1\u6709\u516c\u5f00\u7684\u8bc4\u4f30\u6570\u636e\u548c\u901a\u7528\u7684\u8bc4\u4f30\u6307\u6807\u3002\u8fd9\u4f7f\u5f97\u516c\u5e73\u5730\u8bc4\u4f30 LLM \u5728\u6839\u56e0\u5206\u6790\u65b9\u9762\u7684\u80fd\u529b\u53d8\u5f97\u56f0\u96be\uff0c\u8fdb\u800c\u963b\u788d\u4e86\u8be5\u9886\u57df\u7684\u53d1\u5c55\u3002<\/p>\n

\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e00\u95ee\u9898\uff0c\u5fae\u8f6f DKI \u56e2\u961f\u3001\u9999\u6e2f\u4e2d\u6587\u5927\u5b66\uff08\u6df1\u5733\uff09\u8d3a\u54c1\u5609\u6559\u6388\u56e2\u961f\u4e0e\u6e05\u534e\u5927\u5b66\u88f4\u4e39\u6559\u6388\u5171\u540c\u63d0\u51fa\u4e86\u5f53\u524d\u9996\u4e2a\u516c\u5f00\u7684\u3001\u7528\u4e8e\u8bc4\u4f30 LLM \u6839\u56e0\u5206\u6790\u80fd\u529b\u7684\u57fa\u51c6\u8bc4\u4f30\u96c6 \u2014\u2014OpenRCA\u3002\u672c\u6587\u5df2\u88ab ICLR 2025 \u63a5\u6536\u3002<\/p>\n

OpenRCA \u4e3a\u57fa\u4e8e LLM \u7684 RCA \u4efb\u52a1\u5236\u5b9a\u4e86\u6e05\u6670\u7684\u4efb\u52a1\u5efa\u6a21\u548c\u5bf9\u5e94\u7684\u8bc4\u4f30\u65b9\u6cd5\uff0c\u5e76\u63d0\u4f9b\u4e86\u4e00\u7ec4\u516c\u5f00\u4e14\u7ecf\u8fc7\u4eba\u5de5\u5bf9\u9f50\u7684\u6545\u969c\u8bb0\u5f55\u548c\u5927\u91cf\u7684\u8fd0\u7ef4\u53ef\u89c2\u6d4b\u6570\u636e\u3002\u8fd9\u4e3a\u672a\u6765\u57fa\u4e8e LLM \u7684 RCA \u65b9\u6cd5\u7684\u63a2\u7d22\u5960\u5b9a\u4e86\u57fa\u7840\u3002<\/p>\n

\"\u56fe\u7247\"<\/p>\n

    \n
  • \u8bba\u6587\u6807\u9898\uff1aOpenRCA: Can Large Language Models Locate the Root Cause of Software Failures?<\/li>\n
  • \u8bba\u6587\u5730\u5740\uff1ahttps:\/\/openreview.net\/pdf?id=M4qNIzQYpd<\/li>\n
  • \u5f00\u6e90\u4ee3\u7801\uff1ahttps:\/\/github.com\/microsoft\/OpenRCA<\/li>\n
  • \u8bc4\u6d4b\u699c\u5355\uff1ahttps:\/\/microsoft.github.io\/OpenRCA\/<\/li>\n<\/ul>\n

    \u7814\u7a76\u8005\u53d1\u73b0\uff0c\u5f53\u524d\u4e3b\u6d41\u7684 LLM \u5728\u76f4\u63a5\u89e3\u51b3 OpenRCA \u95ee\u9898\u65f6\u9762\u4e34\u663e\u8457\u6311\u6218\u3002\u4f8b\u5982\uff0cClaude 3.5 \u5728\u63d0\u4f9b\u4e86 oracle KPI \u7684\u60c5\u51b5\u4e0b\uff0c\u4ec5\u89e3\u51b3\u4e86 5.37% \u7684 OpenRCA \u4efb\u52a1\u3002\u5f53\u4f7f\u7528\u968f\u673a\u5747\u5300\u62bd\u6837\u7b56\u7565\u63d0\u53d6\u53ef\u80fd\u76f8\u5173\u7684\u6570\u636e\u65f6\uff0c\u8fd9\u4e00\u7ed3\u679c\u8fdb\u4e00\u6b65\u4e0b\u964d\u5230 3.88%\u3002\u4e3a\u4e86\u4e3a\u89e3\u51b3 OpenRCA \u4efb\u52a1\u6307\u660e\u53ef\u80fd\u7684\u65b9\u5411\uff0c\u7814\u7a76\u8005\u8fdb\u4e00\u6b65\u5f00\u53d1\u4e86 RCA-agent\uff0c\u4ee5\u4f5c\u4e3a\u4e00\u4e2a\u66f4\u6709\u6548\u7684\u57fa\u7ebf\u65b9\u6cd5\u3002\u5728\u4f7f\u7528 RCA-agent \u540e\uff0cClaude 3.5 \u7684\u51c6\u786e\u7387\u63d0\u5347\u81f3 11.34%\uff0c\u4f46\u4f9d\u7136\u79bb\u89e3\u51b3\u597d OpenRCA \u95ee\u9898\u6709\u7740\u8f83\u5927\u7684\u5dee\u8ddd\u3002<\/p>\n

    \u8bc4\u4f30\u57fa\u51c6<\/h4>\n

    \u4efb\u52a1\u5efa\u6a21<\/strong><\/p>\n

    OpenRCA \u5c06\u57fa\u4e8e LLM \u7684\u6839\u56e0\u5206\u6790\u4efb\u52a1\u5b9a\u4e49\u4e3a\u76ee\u6807\u9a71\u52a8\u7684\u5f62\u5f0f\uff1a\u6a21\u578b\u6216\u667a\u80fd\u4f53\u5c06\u63a5\u6536\u7531\u81ea\u7136\u8bed\u8a00\u7ec4\u6210\u7684\u67e5\u8be2\u6307\u4ee4\uff0c\u6267\u884c\u4e0d\u540c\u76ee\u6807\u7684\u6839\u56e0\u5206\u6790\u4efb\u52a1\u3002\u6839\u636e\u67e5\u8be2\u6307\u4ee4\uff0c\u6a21\u578b\u6216\u667a\u80fd\u4f53\u9700\u8981\u901a\u8fc7\u68c0\u7d22\u548c\u5206\u6790\u5f53\u524d\u7cfb\u7edf\u4e2d\u4fdd\u5b58\u7684\u8fd0\u7ef4\u53ef\u89c2\u6d4b\u6570\u636e\uff08\u5305\u62ec\u6307\u6807\u3001\u65e5\u5fd7\u3001\u8c03\u7528\u94fe\uff09\uff0c\u4ece\u4e2d\u63a8\u7406\u5e76\u8bc6\u522b\u51fa\u4e09\u79cd\u6839\u56e0\u7684\u7ec4\u6210\u5143\u7d20\uff08\u6545\u969c\u65f6\u95f4\u3001\u6545\u969c\u7ec4\u4ef6\u548c\u6545\u969c\u539f\u56e0\uff09\u4e2d\u7684\u4e00\u4e2a\u6216\u591a\u4e2a\u5143\u7d20\uff0c\u5e76\u6700\u7ec8\u4ee5 JSON \u7ed3\u6784\u8f93\u51fa\u3002\u5f53\u8f93\u51fa\u4e2d\u7684\u6240\u6709\u5143\u7d20\u4e0e\u6807\u7b7e\u4e00\u81f4\u65f6\uff0c\u8be5\u6837\u672c\u88ab\u89c6\u4e3a\u6b63\u4f8b\uff1b\u5426\u5219\uff0c\u89c6\u4e3a\u53cd\u4f8b\u3002OpenRCA \u901a\u8fc7\u8ba1\u7b97\u6837\u672c\u7684\u5e73\u5747\u9884\u6d4b\u6b63\u786e\u7387\u6765\u8bc4\u4f30\u65b9\u6cd5\u7684\u80fd\u529b\u3002<\/p>\n

    \"\u56fe\u7247\"<\/p>\n

    \u8bc4\u4f30\u6570\u636e<\/strong><\/p>\n

    \u4e3a\u4e86\u786e\u4fdd\u6240\u4f7f\u7528\u7684\u8f6f\u4ef6\u6545\u969c\u8bb0\u5f55\u548c\u8fd0\u7ef4\u6570\u636e\u7684\u8d28\u91cf\uff0cOpenRCA \u5c06\u5f80\u5e74 AIOps \u6311\u6218\u8d5b\u7cfb\u5217\u4e2d\u63d0\u4f9b\u7684\u5927\u91cf\u6765\u81ea\u4f01\u4e1a\u7cfb\u7edf\u7684\u533f\u540d\u8fd0\u7ef4\u6570\u636e\u96c6\u5408\u4f5c\u4e3a\u6570\u636e\u6e90\u3002\u4e3a\u4fdd\u969c\u6570\u636e\u8d28\u91cf\u7684\u53ef\u9760\u6027\uff0c\u7814\u7a76\u8005\u8fdb\u884c\u4e86\u5305\u62ec\u56db\u4e2a\u6b65\u9aa4\u5728\u5185\u7684\u4eba\u5de5\u6570\u636e\u5904\u7406\u548c\u6807\u7b7e\u5bf9\u9f50\u3002\u5177\u4f53\u6765\u8bf4\uff0c\u7814\u7a76\u8005\u6392\u9664\u4e86\u65e0\u6cd5\u7528\u4e8e\u6839\u56e0\u5b9a\u4f4d\u7684\u7cfb\u7edf\u89c2\u6d4b\u6570\u636e\uff0c\u8fd9\u4e9b\u6570\u636e\u901a\u5e38\u4ec5\u80fd\u7528\u4e8e\u7c97\u7c92\u5ea6\u5f02\u5e38\u68c0\u6d4b\uff0c\u800c\u7f3a\u4e4f\u8be6\u7ec6\u7684\u6545\u969c\u8bb0\u5f55\uff0c\u4f8b\u5982\u6545\u969c\u539f\u56e0\u7b49\u4fe1\u606f\u3002\u63a5\u7740\uff0c\u7814\u7a76\u8005\u6574\u7406\u4e86\u8fd9\u4e9b\u7cfb\u7edf\u7684\u6570\u636e\uff0c\u5bf9\u4e0d\u540c\u7cfb\u7edf\u95f4\u7684\u6837\u672c\u6570\u91cf\u8fdb\u884c\u4e86\u5747\u8861\u5316\u5904\u7406\uff0c\u5e76\u6807\u51c6\u5316\u4e86\u4e0d\u540c\u7cfb\u7edf\u6570\u636e\u7684\u76ee\u5f55\u7ed3\u6784\uff0c\u4ee5\u4fbf\u4e8e\u6a21\u578b\u7684\u68c0\u7d22\u3002\u6700\u91cd\u8981\u7684\u662f\uff0c\u4e3a\u4fdd\u8bc1\u95ee\u9898\u7684\u53ef\u89e3\u6027\uff0c\u7814\u7a76\u8005\u624b\u52a8\u9a8c\u8bc1\u4e86\u5269\u4f59\u6570\u636e\u4e2d\u7684\u6545\u969c\u662f\u5426\u53ef\u4ee5\u901a\u8fc7\u76f8\u6570\u636e\u4eba\u5de5\u5b9a\u4f4d\u5230\u6545\u969c\u6839\u56e0\u3002\u7814\u7a76\u8005\u53bb\u9664\u4e86\u6ee1\u8db3\u4ee5\u4e0b\u4efb\u4e00\u6761\u4ef6\u7684\u6570\u636e\u8bb0\u5f55\uff1a\uff081\uff09\u65e0\u6cd5\u6570\u636e\u4e2d\u8bc6\u522b\u6839\u56e0\uff1b\uff082\uff09\u6545\u969c\u671f\u95f4\u6570\u636e\u7f3a\u5931\uff1b\uff083\uff09\u4ece\u6570\u636e\u63a8\u65ad\u51fa\u7684\u6839\u56e0\u4e0e\u6807\u7b7e\u4e0d\u4e00\u81f4\u3002\u6700\u7ec8\uff0c\u7814\u7a76\u8005\u6839\u636e\u4e0d\u540c\u7684\u6839\u56e0\u5b9a\u4f4d\u76ee\u6807\uff0c\u4f7f\u7528 LLM \u4e3a\u6bcf\u4e2a\u6545\u969c\u6848\u4f8b\u751f\u6210\u4e86\u76f8\u5e94\u7684\u67e5\u8be2\u6307\u4ee4\uff0c\u5e76\u5c06\u5176\u5bf9\u5e94\u7684\u6839\u56e0\u5143\u7d20\u4f5c\u4e3a\u6807\u7b7e\uff0c\u6784\u5efa\u4e86 335 \u4e2a\u6839\u56e0\u5b9a\u4f4d\u95ee\u9898\u3002<\/p>\n

    \"\u56fe\u7247\"<\/p>\n

    \u57fa\u7ebf\u65b9\u6cd5<\/h4>\n

    \u4e3a\u4e86\u8bc4\u4f30\u5f53\u524d LLM \u89e3\u51b3 OpenRCA \u95ee\u9898\u7684\u80fd\u529b\uff0c\u7814\u7a76\u8005\u6784\u5efa\u4e86\u4e09\u79cd\u57fa\u7ebf\u65b9\u6cd5\uff0c\u5176\u4e2d\u4e24\u4e2a\u662f\u57fa\u4e8e\u91c7\u6837\u7684 Prompting \u65b9\u6cd5\uff0c\u4e00\u4e2a\u662f\u57fa\u4e8e\u7b80\u5355\u7684 ReAct \u7684 Agentic \u65b9\u6cd5\u3002<\/p>\n

    Sampling-based Prompting<\/strong><\/p>\n

    \u8fd0\u7ef4\u53ef\u89c2\u6d4b\u6570\u636e\u89c4\u6a21\u5e9e\u5927\uff0c\u800c LLM \u7684\u4e0a\u4e0b\u6587\u7a97\u53e3\u6709\u9650\uff0c\u56e0\u6b64\u76f4\u63a5\u8f93\u5165\u5168\u90e8\u6570\u636e\u5e76\u4e0d\u73b0\u5b9e\u3002\u5728\u4f20\u7edf\u6839\u56e0\u5206\u6790\u4e2d\uff0c\u5e38\u89c1\u7684\u5904\u7406\u65b9\u5f0f\u662f\u91c7\u6837\u3002\u7814\u7a76\u8005\u5c06\u6240\u6709\u6570\u636e\uff08\u5305\u62ec\u8ffd\u8e2a\u3001\u65e5\u5fd7\u548c\u6307\u6807\uff09\u6309\u6bcf\u5206\u949f\u4e00\u4e2a\u503c\u8fdb\u884c\u4e0b\u91c7\u6837\uff0c\u5e76\u5bf9\u5177\u4f53\u7684\u6307\u6807\u7c7b\u578b\u8fdb\u4e00\u6b65\u62bd\u6837\u4ee5\u51cf\u5c11 KPI \u5e8f\u5217\u7684\u6570\u91cf\u3002\u7814\u7a76\u8005\u91c7\u7528\u4e86\u4e24\u79cd\u7b56\u7565\u6765\u6267\u884c\u8fd9\u79cd\u62bd\u6837\uff1a<\/p>\n

      \n
    • Balanced Sampling\uff1a\u91c7\u7528\u5206\u5c42\u62bd\u6837\u7b56\u7565\uff0c\u5373\u4ece\u6bcf\u7c7b KPI \u4e2d\u968f\u673a\u9009\u53d6\u4e00\u4e2a\uff0c\u5faa\u73af\u8fdb\u884c\uff0c\u76f4\u5230\u8fbe\u5230\u6a21\u578b\u7684 token \u4e0a\u9650\u3002\u8be5\u65b9\u6cd5\u7b80\u5355\u5b9e\u7528\uff0c\u786e\u4fdd KPI \u7c7b\u578b\u7684\u5206\u5e03\u5747\u8861\u3002\u4e3a\u4fdd\u8bc1\u53ef\u91cd\u590d\u6027\uff0c\u7814\u7a76\u8005\u5bf9\u6bcf\u79cd\u914d\u7f6e\u6d4b\u8bd5\u4e09\u6b21\uff0c\u5e76\u62a5\u544a\u4e2d\u4f4d\u6570\u7ed3\u679c\u3002<\/li>\n
    • Oracle Sampling\uff1a\u4e3a\u7814\u7a76\u62bd\u6837\u65b9\u6cd5\u7684\u6027\u80fd\u4e0a\u9650\uff0c\u7814\u7a76\u8005\u5f15\u5165\u4e86 oracle \u7b56\u7565\uff0c\u5373\u4f7f\u7528\u5728\u57fa\u51c6\u6784\u5efa\u4e2d\u5df2\u7ecf\u88ab\u5de5\u7a0b\u5e08\u9a8c\u8bc1\u80fd\u591f\u6709\u6548\u5b9a\u4f4d\u6839\u56e0\u7684 KPI \u96c6\u5408\u4f5c\u4e3a\u56fa\u5b9a\u7684\u8f93\u5165\u3002\u867d\u7136\u8fd9\u79cd\u65b9\u6cd5\u5728\u5728\u5b9e\u9645\u573a\u666f\u4e2d\u5e76\u4e0d\u73b0\u5b9e\uff0c\u4f46\u80fd\u4f53\u73b0\u91c7\u6837\u7684\u80fd\u529b\u4e0a\u9650\u3002<\/li>\n<\/ul>\n

      RCA-agent<\/strong><\/p>\n

      \u5c3d\u7ba1\u91c7\u6837\u53ef\u7f13\u89e3\u957f\u4e0a\u4e0b\u6587\u7684\u95ee\u9898\uff0c\u4f46\u8fd0\u7ef4\u6570\u636e\u4e2d\u4ecd\u5305\u542b\u5927\u91cf\u975e\u81ea\u7136\u8bed\u8a00\uff08\u5982 GUID\u3001\u9519\u8bef\u7801\u7b49\uff09\uff0cLLM \u5904\u7406\u6b64\u7c7b\u4fe1\u606f\u7684\u80fd\u529b\u6709\u9650\u3002\u4e3a\u6b64\uff0c\u7814\u7a76\u8005\u8bbe\u8ba1\u4e86 RCA-agent\uff0c\u4e00\u4e2a\u57fa\u4e8e Python \u7684\u4ee3\u7801\u751f\u6210\u4e0e\u6267\u884c\u53cd\u9988\u7684\u8f7b\u91cf Agent \u6846\u67b6\uff0c\u5141\u8bb8\u6a21\u578b\u4f7f\u7528\u6570\u636e\u68c0\u7d22\u548c\u5206\u6790\u5de5\u5177\u4ee5\u63d0\u5347\u6a21\u578b\u5bf9\u590d\u6742\u6570\u636e\u7684\u7406\u89e3\u4e0e\u64cd\u4f5c\u80fd\u529b\u3002RCA-agent \u7531\u4e24\u90e8\u5206\u7ec4\u6210\uff1a<\/p>\n

        \n
      • Controller\uff1a\u8d1f\u8d23\u51b3\u7b56\u4e0e\u6d41\u7a0b\u63a7\u5236\uff0c\u5f15\u5bfc\u6a21\u578b\u5b8c\u6210\u5f02\u5e38\u68c0\u6d4b \u2192 \u6545\u969c\u8bc6\u522b \u2192 \u6839\u56e0\u5b9a\u4f4d\u7684\u5206\u6790\u6d41\u7a0b\u3002\u6bcf\u4e2a\u6b65\u9aa4\u90fd\u4f1a\u8981\u6c42 Executor \u5b8c\u6210\u5355\u4e00\u539f\u5b50\u5316\u7684\u4efb\u52a1\uff0c\u5e76\u6839\u636e\u8fd4\u56de\u7ed3\u679c\u6765\u51b3\u7b56\u4e0b\u4e00\u6b65\u3002<\/li>\n
      • Executor\uff1a\u6839\u636e\u63a7\u5236\u5668\u6307\u4ee4\u751f\u6210\u5e76\u6267\u884c Python \u4ee3\u7801\uff0c\u5e76\u8fd4\u56de\u7ed3\u679c\u3002\u5176\u5305\u542b LLM \u4ee3\u7801\u751f\u6210\u5668\u4e0e Python \u6267\u884c\u73af\u5883\u3002\u6240\u6709\u4ee3\u7801\u4e0e\u53d8\u91cf\u5747\u7f13\u5b58\u4e8e\u5185\u5b58\u4e2d\uff0c\u76f4\u81f3\u6574\u4e2a\u4efb\u52a1\u5b8c\u6210\u3002<\/li>\n<\/ul>\n

        \"\u56fe\u7247\"<\/p>\n

        \u4e3b\u8981\u5b9e\u9a8c<\/h4>\n

        \u4e3a\u4e86\u8bc4\u6d4b\u5f53\u524d\u5927\u6a21\u578b\u89e3\u51b3 OpenRCA \u95ee\u9898\u7684\u80fd\u529b\uff0c\u7814\u7a76\u8005\u6311\u9009\u4e86\u516d\u4e2a\u81f3\u5c11\u5177\u6709 128k token \u4e0a\u4e0b\u6587\u957f\u5ea6\u7684\u6a21\u578b\uff0c\u5982 Claude 3.5 sonnet, GPT-4o, Gemini 1.5 pro \u7b49\u3002\u7ed3\u679c\u663e\u793a\uff1a<\/p>\n

          \n
        • \u57fa\u4e8e\u667a\u80fd\u4f53\u7684\u65b9\u6cd5\u7684\u80fd\u529b\u4e0a\u9650\u6bd4\u63d0\u793a\u8bcd\u65b9\u6cd5\u66f4\u597d\u3002<\/li>\n
        • \u5f53\u524d\u6a21\u578b\u5728\u89e3\u51b3 OpenRCA \u95ee\u9898\u4e0a\u4ecd\u9762\u4e34\u6311\u6218\u3002<\/li>\n<\/ul>\n

          \"\u56fe\u7247\"<\/p>\n

          \u8868 1 \u57fa\u7ebf\u65b9\u6cd5\u7684\u51c6\u786e\u7387\u5bf9<\/p>\n

          \"\u56fe\u7247\"<\/p>\n

          \u56fe 2 \u6a21\u578b\u5728\u5404\u4e2a\u7cfb\u7edf\u4e0a\u7684\u51c6\u786e\u7387\u5206\u5e03<\/p>\n

          \u901a\u8fc7\u8fdb\u4e00\u6b65\u5206\u6790\uff0c\u7814\u7a76\u8005\u89c2\u5bdf\u5230\u5f53\u524d\u6a21\u578b\u503e\u5411\u4e8e\u4f7f\u7528\u66f4\u77ed\u7684\u4ea4\u4e92\uff086-10 \u6b65\uff09\u6765\u89e3\u51b3\u95ee\u9898\u3002\u7136\u800c\uff0c\u4ea4\u4e92\u6b21\u6570\u66f4\u591a\u7684\u60c5\u51b5\u4e0b\uff0c\u95ee\u9898\u7684\u6b63\u786e\u7387\u901a\u5e38\u66f4\u9ad8\u3002\u5176\u6b21\uff0c\u7814\u7a76\u8005\u53d1\u73b0\u6a21\u578b\u7684\u4ee3\u7801\u751f\u6210\u548c\u4ee3\u7801\u7ea0\u9519\u80fd\u529b\u4f1a\u5927\u5e45\u5f71\u54cd\u5176\u5728 RCA-agent \u4e0a\u7684\u8868\u73b0\u3002\u5728\u4ec5\u8003\u8651\u90a3\u4e9b\u6267\u884c\u8f68\u8ff9\u4e2d\u51fa\u73b0\u8fc7\u4ee3\u7801\u8fd0\u884c\u5931\u8d25\u60c5\u51b5\u7684\u4f8b\u5b50\u4e2d\uff0cClaude 3.5 sonnet \u7684\u6b63\u786e\u7387\u4ec5\u4e0b\u964d\u4e86 17.9% (11.34->9.31)\u3002\u800c Gemini 1.5 pro \u5219\u4e0b\u964d\u4e86 68.4% (2.69->0.85)\u3002\u8fd9\u4e9b\u53d1\u73b0\u53ef\u80fd\u7684\u542f\u53d1\u662f\uff0c\u5728\u4ee5\u57fa\u4e8e\u4ee3\u7801\u6267\u884c\u7684\u667a\u80fd\u4f53\u65b9\u6cd5\u89e3\u51b3 OpenRCA \u95ee\u9898\u65f6\uff0c\u9700\u5c3d\u53ef\u80fd\u4f7f\u7528\u4ee3\u7801\u80fd\u529b\u66f4\u5f3a\u7684\u6a21\u578b\u8fdb\u884c\u66f4\u957f\u94fe\u6761\u7684\u4ea4\u4e92\u548c\u601d\u8003\u3002<\/p>\n

          \"\u56fe\u7247\"<\/p>\n

          \u56fe 3\uff1a\u4ea4\u4e92\u94fe\u6761\u957f\u5ea6\u7684\u5206\u5e03\uff1b\u56fe 4\uff1a\u6b63\u786e\u7387\u968f\u4ea4\u4e92\u94fe\u6761\u957f\u5ea6\u7684\u5206\u5e03\uff1b\u8868 2\uff1a\u6a21\u578b\u4ee3\u7801\u6267\u884c\u6709\u9519\u65f6\u7684\u6b63\u786e\u7387<\/p>\n

          \u4f7f\u7528\u6307\u5357<\/h4>\n

          OpenRCA \u6570\u636e\u3001\u6587\u6863\u3001\u4ee5\u53ca\u76f8\u5173\u4ee3\u7801\u5df2\u5f00\u6e90\u5728\u4ed3\u5e93\u4e2d\uff1ahttps:\/\/github.com\/microsoft\/OpenRCA<\/p>\n

          \u8981\u4f7f\u7528 OpenRCA \u6570\u636e\uff0c\u9700\u8981\u9996\u5148\u5c06\u539f\u59cb\u6570\u636e\u4e0b\u8f7d\u5230\u672c\u5730\u3002\u6bcf\u4e2a\u5b50\u6570\u636e\u96c6\u4e0b\u90fd\u6709\u82e5\u5e72\u4e2a\u4ee5\u65e5\u671f\u547d\u540d\u7684\u8fd0\u7ef4\u6570\u636e\u76ee\u5f55\uff0c\u4ee5\u53ca\u4e00\u4efd\u539f\u59cb\u6570\u636e\u8bb0\u5f55 (record.csv) \u548c\u95ee\u9898\u6e05\u5355 (query.csv) \u4f7f\u7528\u8005\u9700\u8981\u8ba9\u4ed6\u4eec\u7684\u65b9\u6cd5\u80fd\u591f\u8bbf\u95ee\u5bf9\u5e94\u7684\u8fd0\u7ef4\u6570\u636e\u76ee\u5f55\uff0c\u6765\u89e3\u51b3\u95ee\u9898\u6e05\u5355\u4e0a\u7684\u95ee\u9898\u3002\u6700\u540e\uff0c\u4f7f\u7528\u8005\u53ef\u4ee5\u5229\u7528\u4ed3\u5e93\u4e2d\u7684\u8bc4\u4f30\u811a\u672c (evaluate.py) \u6765\u8bc4\u4f30\u5176\u65b9\u6cd5\u7684\u7ed3\u679c\u6b63\u786e\u6027\u3002<\/p>\n

          \u5982\u679c\u4f7f\u7528\u8005\u5e0c\u671b\u516c\u5f00\u4ed6\u4eec\u7684\u8bc4\u4f30\u7ed3\u679c\u5728 OpenRCA \u7684\u8bc4\u6d4b\u699c\u5355\u4e0a\uff08https:\/\/aka.ms\/openrca\uff09\uff0c\u53ef\u4ee5\u628a\u4ed6\u4eec\u65b9\u6cd5\u7684\u540d\u79f0\u3001\u539f\u59cb\u7ed3\u679c\u6587\u4ef6\u3001\u8dd1\u5206\u3001\u6267\u884c\u8f68\u8ff9\uff08\u5982\u679c\u6709\uff09\u3001\u4ed3\u5e93\u8fde\u63a5\uff08\u5982\u679c\u5f00\u6e90\uff09\u53d1\u9001\u5230 openrcanon@gmail.com\u3002\u6211\u4eec\u5c06\u5728\u786e\u8ba4\u7ed3\u679c\u53ef\u4fe1\u5ea6\u4e4b\u540e\u5c3d\u5feb\u5c06\u7ed3\u679c\u66f4\u65b0\u5728\u6392\u884c\u699c\u4e0a\u3002<\/p>\n

          \u7ed3\u8bed<\/h4>\n

          \u5927\u6a21\u578b\u5728\u8f6f\u4ef6\u5de5\u7a0b\u9886\u57df\u7684\u7814\u7a76\u4ecd\u7136\u662f\u4e00\u7247\u84dd\u6d77\u3002\u672c\u6587\u805a\u7126\u4e8e\u63d0\u4f9b\u4e00\u4e2a\u4efb\u52a1\u5b9a\u4e49\u6e05\u6670\u4e14\u6570\u636e\u5f00\u653e\u7684\u4ee3\u7406\u4efb\u52a1\u6570\u636e\u96c6\uff0c\u6765\u5141\u8bb8\u5404\u79cd\u4e0d\u540c\u7684\u5927\u6a21\u578b RCA \u65b9\u6cd5\u4f7f\u505a\u516c\u5e73\u5bf9\u6bd4\u3002\u672c\u6587\u7684\u8bc4\u6d4b\u4e5f\u4ec5\u5c40\u9650\u4e8e\u5927\u6a21\u578b\u672c\u8eab\u7684 RCA \u80fd\u529b\u4e0a\u3002\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\uff0c\u8fd8\u6709\u8bb8\u591a\u53ef\u4ee5\u8fdb\u4e00\u6b65\u5de5\u7a0b\u4f18\u5316\u7684\u70b9\uff0c\u5982\u914d\u7f6e\u5b9a\u5236\u5316\u5de5\u5177\u6765\u907f\u514d\u6a21\u578b\u5b8c\u5168\u81ea\u7531\u63a8\u7406\u548c\u7f16\u7801\u4ea7\u751f\u7684\u5e7b\u89c9\u95ee\u9898\u3002\u5e0c\u671b\u8fd9\u7bc7\u8bba\u6587\u80fd\u629b\u7816\u5f15\u7389\uff0c\u6fc0\u53d1\u66f4\u591a\u8f6f\u4ef6\u5de5\u7a0b\u4efb\u52a1\u4e0a\u7684\u5927\u6a21\u578b\u7814\u7a76\u7684\u4ea7\u751f\u3002<\/p>\n

          \u672c\u6587\u5df2\u88ab ICLR 2025 \u63a5\u6536\uff0c\u6d77\u62a5\u5c06\u4e8e 4 \u6708 25 \u65e5\u4e0b\u5348 3 \u70b9\u81f3 5 \u70b9\u534a\u5728 Hall 3 + Hall 2B #115 \u5c55\u51fa\u3002\u6b22\u8fce\u611f\u5174\u8da3\u7684\u540c\u884c\u4eec\u6765\u4e0e\u4f5c\u8005\u4ea4\u6d41\u8ba8\u8bba\u3002<\/p>\n

          \u6587\u7ae0\u6765\u81ea\uff1a51CTO<\/p>\n<\/div>\n<\/div>\n

          <\/div>\n

          <\/g><\/svg><\/i> \"Loading\"<\/p>\n

          <\/div>\n","protected":false},"excerpt":{"rendered":"

          \u5fae\u8f6f DKI \u56e2\u961f\u3001\u9999\u6e2f\u4e2d\u6587\u5927\u5b66\uff08\u6df1\u5733\uff09\u8d3a\u54c1\u5609\u6559\u6388\u56e2\u961f\u4e0e\u6e05\u534e\u5927\u5b66\u88f4\u4e39\u6559\u6388\u5171\u540c\u63d0\u51fa\u4e86\u5f53\u524d\u9996\u4e2a\u516c\u5f00\u7684\u3001\u7528\u4e8e\u8bc4\u4f30 L […]<\/p>\n

          <\/div>\n

          <\/g><\/svg><\/i> \"Loading\"<\/p>\n

          <\/div>\n","protected":false},"author":56,"featured_media":25553,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3,23,20,80],"tags":[124,986,820],"class_list":["post-25551","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","category-23","category-20","category-80","tag-ai","tag-llm","tag-820"],"_links":{"self":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/25551","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/users\/56"}],"replies":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/comments?post=25551"}],"version-history":[{"count":1,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/25551\/revisions"}],"predecessor-version":[{"id":25554,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/25551\/revisions\/25554"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media\/25553"}],"wp:attachment":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media?parent=25551"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/categories?post=25551"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/tags?post=25551"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}