{"id":53769,"date":"2025-02-16T16:39:24","date_gmt":"2025-02-16T08:39:24","guid":{"rendered":"https:\/\/fwq.ai\/blog\/53769\/"},"modified":"2025-02-16T16:39:24","modified_gmt":"2025-02-16T08:39:24","slug":"%e5%9f%ba%e4%ba%8e%e5%a4%a7%e6%a8%a1%e5%9e%8b%e7%9a%84%e5%90%88%e6%88%90%e6%95%b0%e6%8d%ae%e7%94%9f%e6%88%90","status":"publish","type":"post","link":"https:\/\/fwq.ai\/blog\/53769\/","title":{"rendered":"\u57fa\u4e8e\u5927\u6a21\u578b\u7684\u5408\u6210\u6570\u636e\u751f\u6210"},"content":{"rendered":"<p>\u6784\u5efa\u5927\u89c4\u6a21\u3001\u5168\u9762\u7684\u6570\u636e\u96c6\u6765\u6d4b\u8bd5 LLM \u8f93\u51fa\u53ef\u80fd\u662f\u4e00\u4e2a\u8d39\u529b\u3001\u6602\u8d35\u4e14\u5177\u6709\u6311\u6218\u6027\u7684\u8fc7\u7a0b\uff0c\u5c24\u5176\u662f\u4ece\u5934\u5f00\u59cb\u6784\u5efa\u65f6\u3002\u4f46\u662f\uff0c\u5982\u679c\u6211\u544a\u8bc9\u4f60\uff0c\u73b0\u5728\u53ea\u9700\u51e0\u5206\u949f\u5c31\u53ef\u4ee5\u751f\u6210\u901a\u5e38\u82b1\u8d39\u6570\u5468\u7cbe\u5fc3\u5236\u4f5c\u7684\u6570\u5343\u4e2a\u9ad8\u8d28\u91cf\u6d4b\u8bd5\u7528\u4f8b\uff0c\u4f60\u4f1a\u600e\u4e48\u60f3\uff1f<\/p>\n<p>\u5408\u6210\u6570\u636e\u751f\u6210\u5229\u7528 LLM \u521b\u5efa\u9ad8\u8d28\u91cf\u6570\u636e\uff0c\u800c\u65e0\u9700\u624b\u52a8\u6536\u96c6\u3001\u6e05\u7406\u548c\u6807\u6ce8\u5927\u91cf\u6570\u636e\u96c6\u3002\u501f\u52a9 GPT-4 \u7b49\u6a21\u578b\uff0c\u73b0\u5728\u53ef\u4ee5\u5728\u66f4\u77ed\u7684\u65f6\u95f4\u5185\u5408\u6210\u751f\u6210\u6bd4\u4eba\u5de5\u6807\u8bb0\u6570\u636e\u96c6\u66f4\u5168\u9762\u3001\u66f4\u591a\u6837\u5316\u7684\u6570\u636e\u96c6\uff0c\u8fd9\u4e9b\u6570\u636e\u96c6\u53ef\u7528\u4e8e\u501f\u52a9\u4e00\u4e9b LLM \u8bc4\u4f30\u6307\u6807\u5bf9 LLM\uff08\u7cfb\u7edf\uff09\u8fdb\u884c\u57fa\u51c6\u6d4b\u8bd5\u3002<\/p>\n<p>\u5728\u672c\u6587\u4e2d\uff0c\u6211\u5c06\u6559\u4f60\u6709\u5173\u5982\u4f55\u4f7f\u7528 LLM \u751f\u6210\u5408\u6210\u6570\u636e\u96c6\uff08\u4f8b\u5982\uff0c\u53ef\u7528\u4e8e\u8bc4\u4f30 RAG \u7ba1\u9053\uff09\u7684\u6240\u6709\u77e5\u8bc6\u3002\u6211\u4eec\u5c06\u63a2\u7d22\uff1a<\/p>\n<ul>\n<li>\u5408\u6210\u751f\u6210\u65b9\u6cd5\uff08\u84b8\u998f\u548c\u81ea\u6211\u6539\u8fdb\uff09<\/li>\n<li>\u4ec0\u4e48\u662f\u6570\u636e\u6f14\u5316\u3001\u5404\u79cd\u6f14\u5316\u6280\u672f\u53ca\u5176\u5728\u5408\u6210\u6570\u636e\u751f\u6210\u4e2d\u7684\u4f5c\u7528<\/li>\n<li>\u4f7f\u7528 LLM \u4ece\u5934\u5f00\u59cb\u200b\u200b\u521b\u5efa\u9ad8\u8d28\u91cf\u5408\u6210\u6570\u636e\u7684\u5206\u6b65\u6559\u7a0b\u3002<\/li>\n<li>\u5982\u4f55\u4f7f\u7528 DeepEval \u5728\u4e0d\u5230 5 \u884c\u4ee3\u7801\u4e2d\u751f\u6210\u5408\u6210\u6570\u636e\u96c6\u3002<\/li>\n<\/ul>\n<p>\u611f\u5174\u8da3\u5417\uff1f\u8ba9\u6211\u4eec\u6df1\u5165\u4e86\u89e3\u3002<\/p>\n<h2>1\u3001\u4ec0\u4e48\u662f\u4f7f\u7528 LLM \u7684\u5408\u6210\u6570\u636e\u751f\u6210\uff1f<\/h2>\n<p>\u4f7f\u7528 LLM \u8fdb\u884c\u5408\u6210\u6570\u636e\u751f\u6210\u6d89\u53ca\u4f7f\u7528 LLM \u521b\u5efa\u4eba\u5de5\u6570\u636e\uff0c\u8fd9\u4e9b\u4eba\u5de5\u6570\u636e\u901a\u5e38\u662f\u53ef\u7528\u4e8e\u8bad\u7ec3\u3001\u5fae\u8c03\u751a\u81f3\u8bc4\u4f30 LLM \u672c\u8eab\u7684\u6570\u636e\u96c6\u3002\u751f\u6210\u5408\u6210\u6570\u636e\u96c6\u4e0d\u4ec5\u6bd4\u641c\u7d22\u516c\u5171\u6570\u636e\u96c6\u66f4\u5feb\u3001\u6bd4\u4eba\u5de5\u6807\u6ce8\u66f4\u4fbf\u5b9c\uff0c\u800c\u4e14\u8fd8\u80fd\u63d0\u9ad8\u8d28\u91cf\u548c\u6570\u636e\u591a\u6837\u6027\uff0c\u8fd9\u5bf9\u4e8e\u7ea2\u961f LLM \u5e94\u7528\u7a0b\u5e8f\u4e5f\u662f\u5fc5\u4e0d\u53ef\u5c11\u7684\u3002<\/p>\n<p>\u8be5\u8fc7\u7a0b\u4ece\u521b\u5efa\u5408\u6210\u67e5\u8be2\u5f00\u59cb\uff0c\u8fd9\u4e9b\u67e5\u8be2\u662f\u4f7f\u7528\u77e5\u8bc6\u5e93\u4e2d\u7684\u4e0a\u4e0b\u6587\uff08\u901a\u5e38\u4ee5\u6587\u6863\u7684\u5f62\u5f0f\uff09\u4f5c\u4e3a\u57fa\u672c\u4e8b\u5b9e\u751f\u6210\u7684\u3002\u7136\u540e\uff0c\u751f\u6210\u7684\u67e5\u8be2\u4f1a\u7ecf\u8fc7\u591a\u6b21\u201c\u6f14\u5316\u201d\uff0c\u4f7f\u5176\u590d\u6742\u5316\u548c\u903c\u771f\u5316\uff0c\u5e76\u4e0e\u751f\u6210\u67e5\u8be2\u7684\u539f\u59cb\u4e0a\u4e0b\u6587\u76f8\u7ed3\u5408\uff0c\u6784\u6210\u6700\u7ec8\u7684\u5408\u6210\u6570\u636e\u96c6\u3002\u867d\u7136\u8fd9\u662f\u53ef\u9009\u7684\uff0c\u4f46\u4f60\u4e5f\u53ef\u4ee5\u9009\u62e9\u4e3a\u6bcf\u4e2a\u5408\u6210\u67e5\u8be2-\u4e0a\u4e0b\u6587\u5bf9\u751f\u6210\u4e00\u4e2a\u76ee\u6807\u6807\u7b7e\uff0c\u8be5\u6807\u7b7e\u5c06\u4f5c\u4e3a\u7ed9\u5b9a\u67e5\u8be2\u7684 LLM \u7cfb\u7edf\u7684\u9884\u671f\u8f93\u51fa\u3002<\/p>\n<p>  \u6570\u636e\u5408\u6210\u5668\u67b6\u6784 <\/p>\n<p>\u5728\u751f\u6210\u7528\u4e8e\u8bc4\u4f30\u7684\u5408\u6210\u6570\u636e\u96c6\u65f6\uff0c\u4e3b\u8981\u6709\u4e24\u79cd\u65b9\u6cd5\uff1a\u4f7f\u7528\u6a21\u578b\u8f93\u51fa\u8fdb\u884c\u81ea\u6211\u6539\u8fdb\uff0c\u6216\u4ece\u66f4\u9ad8\u7ea7\u7684\u6a21\u578b\u4e2d\u8fdb\u884c\u63d0\u70bc\u3002<\/p>\n<ul>\n<li>\u81ea\u6211\u6539\u8fdb\uff1a\u6d89\u53ca\u60a8\u7684\u6a21\u578b\u4ece\u5176\u81ea\u8eab\u8f93\u51fa\u8fed\u4ee3\u751f\u6210\u6570\u636e\uff0c\u800c\u65e0\u9700\u5916\u90e8\u4f9d\u8d56<\/li>\n<li>\u84b8\u998f\uff1a\u6d89\u53ca\u4f7f\u7528\u66f4\u5f3a\u5927\u7684\u6a21\u578b\u751f\u6210\u5408\u6210\u6570\u636e\u4ee5\u8bc4\u4f30\u8f83\u5f31\u7684\u6a21\u578b<\/li>\n<\/ul>\n<p>\u81ea\u6211\u6539\u8fdb\u65b9\u6cd5\uff08\u4f8b\u5982 Self-Instruct \u6216 SPIN\uff09\u53d7\u5230\u6a21\u578b\u529f\u80fd\u7684\u9650\u5236\uff0c\u53ef\u80fd\u4f1a\u53d7\u5230\u504f\u5dee\u548c\u9519\u8bef\u653e\u5927\u7684\u5f71\u54cd\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0c\u84b8\u998f\u6280\u672f\u4ec5\u53d7\u53ef\u7528\u7684\u6700\u4f73\u6a21\u578b\u7684\u9650\u5236\uff0c\u53ef\u786e\u4fdd\u6700\u9ad8\u8d28\u91cf\u7684\u751f\u6210\u3002<\/p>\n<h2>2\u3001\u4ece\u77e5\u8bc6\u5e93\u751f\u6210\u6570\u636e<\/h2>\n<p>\u5408\u6210\u6570\u636e\u751f\u6210\u7684\u7b2c\u4e00\u6b65\u662f\u4ece\u4e0a\u4e0b\u6587\u5217\u8868\u4e2d\u521b\u5efa\u5408\u6210\u67e5\u8be2\uff0c\u8fd9\u4e9b\u67e5\u8be2\u76f4\u63a5\u6765\u81ea\u77e5\u8bc6\u5e93\u3002\u672c\u8d28\u4e0a\uff0c\u4e0a\u4e0b\u6587\u662f LLM \u5e94\u7528\u7a0b\u5e8f\u7684\u7406\u60f3\u68c0\u7d22\u4e0a\u4e0b\u6587\uff0c\u5c31\u50cf\u9884\u671f\u8f93\u51fa\u662f LLM \u5b9e\u9645\u8f93\u51fa\u7684\u57fa\u672c\u4e8b\u5b9e\u53c2\u8003\u4e00\u6837\u3002<\/p>\n<p>\u5bf9\u4e8e\u90a3\u4e9b\u60f3\u8981\u7acb\u5373\u5de5\u4f5c\u7684\u4eba\uff0c\u4ee5\u4e0b\u662f\u4f7f\u7528\u5f00\u6e90 LLM \u8bc4\u4f30\u6846\u67b6 \u751f\u6210\u9ad8\u8d28\u91cf\u5408\u6210\u6570\u636e\u7684\u65b9\u6cd5\uff1a<\/p>\n<pre><code>from deepeval.synthesizer import Synthesizer\n\nsynthesizer = Synthesizer()\nsynthesizer.generate_goldens_from_docs(\n    document_paths=['example.txt', 'example.docx', 'example.pdf'],\n)<\/code><\/pre>\n<h3>2.1 \u6784\u5efa\u4e0a\u4e0b\u6587<\/h3>\n<p>\u5728\u4e0a\u4e0b\u6587\u751f\u6210\u8fc7\u7a0b\u4e2d\uff0c\u4f7f\u7528\u6807\u8bb0\u62c6\u5206\u5668\u5c06\u77e5\u8bc6\u5e93\u4e2d\u7684\u6587\u6863\uff08\u6216\u591a\u4e2a\u6587\u6863\uff09\u5206\u6210\u5757\u3002\u7136\u540e\u9009\u62e9\u968f\u673a\u5757\uff0c\u5e76\u6839\u636e\u76f8\u4f3c\u6027\u68c0\u7d22\u5176\u4ed6\u5757\u5e76\u5c06\u5176\u4e0e\u6240\u9009\u5757\u5206\u7ec4\u3002<\/p>\n<p>  \u4f7f\u7528\u4f59\u5f26\u76f8\u4f3c\u5ea6\u751f\u6210\u4e0a\u4e0b\u6587 <\/p>\n<p>\u53ef\u4ee5\u4f7f\u7528\u591a\u79cd\u65b9\u6cd5\u8ba1\u7b97\u76f8\u4f3c\u5ea6\uff1a<\/p>\n<ul>\n<li>\u5229\u7528\u76f8\u4f3c\u5ea6\u6216\u8ddd\u79bb\u7b97\u6cd5\uff0c\u4f8b\u5982\u4f59\u5f26\u76f8\u4f3c\u5ea6<\/li>\n<li>\u4f7f\u7528\u77e5\u8bc6\u56fe\u8c31<\/li>\n<li>\u5229\u7528 LLM \u672c\u8eab\uff0c\u867d\u7136\u4e0d\u592a\u73b0\u5b9e\uff0c\u4f46\u51c6\u786e\u6027\u6700\u9ad8<\/li>\n<li>\u5e94\u7528\u805a\u7c7b\u6280\u672f\u6765\u8bc6\u522b\u6a21\u5f0f<\/li>\n<li>\u4f7f\u7528\u673a\u5668\u5b66\u4e60\u6a21\u578b\u6839\u636e\u7279\u5f81\u9884\u6d4b\u5206\u7ec4\u3002<\/li>\n<\/ul>\n<p>\u65e0\u8bba\u5982\u4f55\uff0c\u603b\u4f53\u76ee\u6807\u4fdd\u6301\u4e0d\u53d8\uff1a\u6709\u6548\u5730\u805a\u5408\u76f8\u4f3c\u7684\u4fe1\u606f\u5757\u3002<\/p>\n<p>\u4e3a\u4e86\u786e\u4fdd\u8fd9\u4e9b\u5206\u7ec4\u6709\u6548\u5e76\u7b26\u5408 LLM \u5e94\u7528\u7a0b\u5e8f\u7684\u7279\u5b9a\u9700\u6c42\uff0c\u6700\u4f73\u505a\u6cd5\u662f\u5728\u4e0a\u4e0b\u6587\u751f\u6210\u8fc7\u7a0b\u4e2d\u955c\u50cf\u5e94\u7528\u7a0b\u5e8f\u7684\u68c0\u7d22\u5668\u903b\u8f91\u3002\u8fd9\u5305\u62ec\u4ed4\u7ec6\u8003\u8651\u8bf8\u5982\u6807\u8bb0\u62c6\u5206\u65b9\u6cd5\u3001\u5757\u5927\u5c0f\u548c\u5757\u91cd\u53e0\u7b49\u65b9\u9762\u3002<\/p>\n<p>\u8fd9\u79cd\u5bf9\u9f50\u53ef\u786e\u4fdd\u5408\u6210\u6570\u636e\u7684\u884c\u4e3a\u4e0e\u5e94\u7528\u7a0b\u5e8f\u7684\u671f\u671b\u4e00\u81f4\uff0c\u4ece\u800c\u9632\u6b62\u7531\u4e8e\u68c0\u7d22\u5668\u590d\u6742\u6027\u7684\u5dee\u5f02\uff0c\u7ed3\u679c\u4f1a\u6709\u6240\u4e0d\u540c\u3002<\/p>\n<h3>2.2 \u4ece\u4e0a\u4e0b\u6587\u751f\u6210\u5408\u6210\u8f93\u5165<\/h3>\n<p>\u4e00\u65e6\u521b\u5efa\u4e86\u4e0a\u4e0b\u6587\uff0c\u5c31\u4f1a\u4ece\u4e2d\u751f\u6210\u5408\u6210\u8f93\u5165\u3002\u6b64\u65b9\u6cd5\u53cd\u8f6c\u4e86\u6807\u51c6\u68c0\u7d22\u64cd\u4f5c &#8211; \u4e0d\u662f\u6839\u636e\u8f93\u5165\u5b9a\u4f4d\u4e0a\u4e0b\u6587\uff0c\u800c\u662f\u57fa\u4e8e\u9884\u5b9a\u4e49\u7684\u4e0a\u4e0b\u6587\u521b\u5efa\u8f93\u5165\u3002\u8fd9\u53ef\u786e\u4fdd\u6bcf\u4e2a\u5408\u6210\u8f93\u5165\u90fd\u76f4\u63a5\u5bf9\u5e94\u4e8e\u4e00\u4e2a\u4e0a\u4e0b\u6587\uff0c\u4ece\u800c\u63d0\u9ad8\u76f8\u5173\u6027\u548c\u51c6\u786e\u6027\u3002<\/p>\n<p>\u6b64\u5916\uff0c\u8fd9\u4e9b\u4e0a\u4e0b\u6587\u53ef\u7528\u4e8e\u901a\u8fc7\u5c06\u5b83\u4eec\u4e0e\u5408\u6210\u751f\u6210\u7684\u8f93\u5165\u5bf9\u9f50\u6765\u9009\u62e9\u6027\u5730\u751f\u6210\u9884\u671f\u7684\u8f93\u51fa\u3002<\/p>\n<p>  \u67e5\u8be2\u751f\u6210\u7684\u975e\u5bf9\u79f0\u65b9\u6cd5 <\/p>\n<p>\u8fd9\u79cd\u975e\u5bf9\u79f0\u65b9\u6cd5\u53ef\u786e\u4fdd\u6240\u6709\u7ec4\u4ef6\uff08\u8f93\u5165\u3001\u8f93\u51fa\u548c\u4e0a\u4e0b\u6587\uff09\u90fd\u5b8c\u5168\u540c\u6b65\u3002<\/p>\n<h2>3\u3001\u8fc7\u6ee4\u5408\u6210\u6570\u636e<\/h2>\n<p>\u5728\u5f00\u59cb\u6539\u8fdb\u65b0\u751f\u6210\u7684\u6570\u636e\u96c6\u4e4b\u524d\uff0c\u5fc5\u987b\u8fdb\u884c\u5f7b\u5e95\u7684\u8d28\u91cf\u68c0\u67e5\uff0c\u4ee5\u907f\u514d\u6539\u8fdb\u672c\u8eab\u5b58\u5728\u7f3a\u9677\u7684\u8f93\u5165\u3002\u6b64\u6b65\u9aa4\u5bf9\u4e8e\u786e\u4fdd\u4e0d\u4f1a\u6d6a\u8d39\u5b9d\u8d35\u7684\u8d44\u6e90\u4ee5\u53ca\u6700\u7ec8\u6570\u636e\u96c6\u4ec5\u5305\u542b\u9ad8\u8d28\u91cf\u7684\u9ec4\u91d1\u81f3\u5173\u91cd\u8981\u3002<\/p>\n<p>\u8fc7\u6ee4\u53d1\u751f\u5728\u5408\u6210\u6570\u636e\u751f\u6210\u7684\u4e24\u4e2a\u5173\u952e\u9636\u6bb5\uff1a\u9996\u5148\u662f\u5728\u4e0a\u4e0b\u6587\u751f\u6210\u671f\u95f4\uff0c\u7136\u540e\u662f\u4ece\u8fd9\u4e9b\u4e0a\u4e0b\u6587\u751f\u6210\u5408\u6210\u8f93\u5165\u671f\u95f4\u3002<\/p>\n<h3>3.1 \u4e0a\u4e0b\u6587\u8fc7\u6ee4<\/h3>\n<p>\u5728\u4e0a\u4e0b\u6587\u751f\u6210\u671f\u95f4\uff0c\u4f60\u53ef\u80fd\u4f1a\u968f\u673a\u9009\u62e9\u4e00\u4e2a\u4f4e\u8d28\u91cf\u7684\u5757\u3002\u901a\u5e38\uff0c\u4f60\u7684\u77e5\u8bc6\u5e93\u53ef\u80fd\u5305\u542b\u590d\u6742\u7684\u7ed3\u6784\u6216\u591a\u4f59\u7684\u7a7a\u767d\uff0c\u5206\u89e3\u540e\u4f1a\u53d8\u5f97\u96be\u4ee5\u7406\u89e3\u3002\u8058\u8bf7 LLM \u4f5c\u4e3a\u8bc4\u5224\u5458\u662f\u8bc6\u522b\u548c\u6d88\u9664\u8fd9\u4e9b\u4f4e\u8d28\u91cf\u4e0a\u4e0b\u6587\u7684\u53ef\u9760\u65b9\u6cd5\u3002<\/p>\n<p>  \u4e0a\u4e0b\u6587\u8fc7\u6ee4\u793a\u4f8b <\/p>\n<p>\u4f60\u53ef\u4ee5\u81ea\u5b9a\u4e49\u8bc4\u4f30\u548c\u8fc7\u6ee4\u8fd9\u4e9b\u4e0a\u4e0b\u6587\u7684\u6807\u51c6\uff0c\u4f46\u4ee5\u4e0b\u662f\u4e00\u4e9b\u9700\u8981\u8003\u8651\u7684\u57fa\u672c\u51c6\u5219\uff1a<\/p>\n<ul>\n<li>\u6e05\u6670\u5ea6\uff1a\u8bc4\u4f30\u4fe1\u606f\u7684\u6e05\u6670\u5ea6\u548c\u53ef\u7406\u89e3\u6027\u3002<\/li>\n<li>\u6df1\u5ea6\uff1a\u8bc4\u4f30\u8be6\u7ec6\u5206\u6790\u7684\u6c34\u5e73\u548c\u539f\u521b\u89c1\u89e3\u7684\u5b58\u5728\u3002<\/li>\n<li>\u7ed3\u6784\uff1a\u5ba1\u67e5\u5185\u5bb9\u7684\u7ec4\u7ec7\u548c\u903b\u8f91\u8fdb\u5c55\u3002<\/li>\n<li>\u76f8\u5173\u6027\uff1a\u786e\u5b9a\u5185\u5bb9\u4e0e\u4e3b\u8981\u4e3b\u9898\u7684\u76f8\u5173\u6027\u3002<\/li>\n<li>\u7cbe\u5ea6\uff1a\u8861\u91cf\u51c6\u786e\u6027\u548c\u5bf9\u7ec6\u8282\u7684\u5173\u6ce8\u3002<\/li>\n<li>\u65b0\u9896\u6027\uff1a\u8bc4\u4f30\u5185\u5bb9\u7684\u72ec\u7279\u6027\u548c\u539f\u521b\u6027\u3002<\/li>\n<li>\u7b80\u6d01\u6027\uff1a\u8bc4\u4f30\u6c9f\u901a\u7684\u7b80\u6d01\u6027\u548c\u6548\u7387\u3002<\/li>\n<li>\u5f71\u54cd\uff1a\u5224\u65ad\u5185\u5bb9\u5bf9\u53d7\u4f17\u7684\u6f5c\u5728\u5f71\u54cd\u3002<\/li>\n<\/ul>\n<p>\u8fc7\u6ee4\u6389\u4f4e\u8d28\u91cf\u7684\u5757\u540e\uff0c\u4f60\u8fd8\u9700\u8981\u786e\u4fdd\u5269\u4f59\u7684\u5757\u8db3\u591f\u76f8\u4f3c\u3002\u8fd9\u6d89\u53ca\u4e8c\u6b21\u8fc7\u6ee4\u8fc7\u7a0b\uff0c\u5728\u8be5\u8fc7\u7a0b\u4e2d\uff0c\u4f60\u4f1a\u5254\u9664\u672a\u901a\u8fc7\u76f8\u4f3c\u5ea6\u9608\u503c\u7684\u5757\u3002<\/p>\n<p>\u8fd9\u79cd\u7ed3\u6784\u5316\u7684\u8fc7\u6ee4\u65b9\u6cd5\u53ef\u786e\u4fdd\u53ea\u6709\u9ad8\u8d28\u91cf\u3001\u76f8\u5173\u4e14\u6709\u7528\u7684\u4e0a\u4e0b\u6587\u624d\u80fd\u8fdb\u5165\u5408\u6210\u6570\u636e\u751f\u6210\u7684\u4e0b\u4e00\u9636\u6bb5\u3002<\/p>\n<h3>3.2 \u8f93\u5165\u8fc7\u6ee4<\/h3>\n<p>\u7b2c\u4e8c\u9636\u6bb5\u7684\u91cd\u70b9\u662f\u4ece\u8fd9\u4e9b\u4e0a\u4e0b\u6587\u751f\u6210\u7684\u5408\u6210\u8f93\u5165\u3002\u6b64\u6b65\u9aa4\u81f3\u5173\u91cd\u8981\uff0c\u56e0\u4e3a\u5373\u4f7f\u662f\u7cbe\u5fc3\u7b56\u5212\u7684\u4e0a\u4e0b\u6587\u6709\u65f6\u4e5f\u4f1a\u5bfc\u81f4\u751f\u6210\u53ef\u80fd\u4e0d\u7b26\u5408\u8981\u6c42\u6807\u51c6\u7684\u8f93\u5165\u3002<\/p>\n<p>  \u8f93\u5165\u8fc7\u6ee4\u793a\u4f8b <\/p>\n<p>\u4ee5\u4e0b\u662f\u4f60\u53ef\u80fd\u5e0c\u671b\u5224\u65ad\u5408\u6210\u8f93\u5165\u7684\u51e0\u4e2a\u6807\u51c6\uff1a<\/p>\n<ul>\n<li>\u81ea\u5305\u542b\uff1a\u786e\u4fdd\u8f93\u5165\u5b8c\u6574\u5e76\u4e14\u53ef\u4ee5\u5728\u6ca1\u6709\u5916\u90e8\u5f15\u7528\u7684\u60c5\u51b5\u4e0b\u72ec\u7acb\u8fd0\u884c\u3002<\/li>\n<li>\u6e05\u6670\u5ea6\uff1a\u68c0\u67e5\u8f93\u5165\u662f\u5426\u6e05\u695a\u5730\u4f20\u8fbe\u4e86\u5176\u9884\u671f\u7684\u4fe1\u606f\u6216\u95ee\u9898\uff0c\u4ee5\u907f\u514d\u8bef\u89e3\u3002<\/li>\n<li>\u4e00\u81f4\u6027\uff1a\u786e\u4fdd\u8f93\u5165\u5728\u4e3b\u9898\u548c\u4e8b\u5b9e\u4e0a\u4e0e\u63d0\u4f9b\u7684\u4e0a\u4e0b\u6587\u6216\u80cc\u666f\u4fe1\u606f\u76f8\u7b26\u3002<\/li>\n<li>\u76f8\u5173\u6027\uff1a\u9a8c\u8bc1\u8f93\u5165\u662f\u5426\u4e0e\u9884\u671f\u4efb\u52a1\u6216\u67e5\u8be2\u76f4\u63a5\u76f8\u5173\uff0c\u786e\u4fdd\u5176\u6709\u76ee\u7684\u4e14\u5207\u9898\u3002<\/li>\n<li>\u5b8c\u6574\u6027\uff1a\u786e\u8ba4\u8f93\u5165\u5305\u542b\u6709\u6548\u4ea4\u4e92\u6216\u67e5\u8be2\u89e3\u51b3\u6240\u9700\u7684\u6240\u6709\u5fc5\u8981\u7ec6\u8282\u3002<\/li>\n<\/ul>\n<p>\u4f7f\u7528\u8fd9\u4e9b\u6807\u51c6\u6709\u52a9\u4e8e\u786e\u4fdd\u5408\u6210\u8f93\u5165\u4e0d\u4ec5\u8d28\u91cf\u9ad8\uff0c\u800c\u4e14\u5b8c\u5168\u9002\u5408\u5176\u9884\u671f\u5e94\u7528\u3002<\/p>\n<h2>4\u3001\u5408\u6210\u6570\u636e\u6837\u5f0f<\/h2>\n<p>\u6700\u540e\uff0c\u4f60\u53ef\u80fd\u5e0c\u671b\u9488\u5bf9\u7279\u5b9a\u4e3b\u9898\u5b9a\u5236\u67e5\u8be2\uff0c\u5e76\u81ea\u5b9a\u4e49\u5176\u8f93\u5165\u548c\u8f93\u51fa\u683c\u5f0f\u4ee5\u9002\u5408\u60a8\u7684\u72ec\u7279\u7528\u4f8b\u3002<\/p>\n<p>\u4f8b\u5982\uff0c\u5982\u679c\u4f60\u7684\u5e94\u7528\u7a0b\u5e8f\u6d89\u53ca\u5c06\u6587\u672c\u8f6c\u6362\u4e3a SQL\uff0c\u5219\u8f93\u51fa\u5e94\u51c6\u786e\u53cd\u6620 SQL \u8bed\u53e5\u3002\u5728\u6d89\u53ca\u8bc4\u4f30 LLM \u7684\u573a\u666f\u4e2d\uff0c\u4f7f\u7528\u5e26\u6709\u201c\u5206\u6570\u201d\u548c\u201c\u539f\u56e0\u201d\u7b49\u952e\u7684 JSON \u683c\u5f0f\u53ef\u80fd\u66f4\u5408\u9002\u3002<br \/>\u60a8\u5e94\u8be5\u8ba1\u5212\u5728\u521d\u59cb\u751f\u6210\u671f\u95f4\u3001\u4efb\u4f55\u6f14\u53d8\u53d8\u5316\u671f\u95f4\u4ee5\u53ca\u751f\u6210\u6700\u7ec8\u8f93\u51fa\u540e\u5e94\u7528\u7279\u5b9a\u6837\u5f0f\u3002\u5728\u521d\u59cb\u751f\u6210\u4e4b\u540e\u91cd\u65b0\u5ba1\u89c6\u6837\u5f0f\u81f3\u5173\u91cd\u8981\uff0c\u56e0\u4e3a\u5408\u6210\u67e5\u8be2\u7684\u6f14\u53d8\u53ef\u80fd\u4f1a\u6539\u53d8\u521d\u59cb\u6837\u5f0f\u610f\u56fe\u3002<\/p>\n<p>\u7b2c\u4e00\u8f6e\u4e4b\u540e\u7684\u6837\u5f0f\u8c03\u6574\u7a0b\u5ea6\u5c06\u53d6\u51b3\u4e8e\u4f60\u5bf9\u6700\u7ec8\u4ea7\u54c1\u7684\u63a7\u5236\u7a0b\u5ea6\u4ee5\u53ca\u76f8\u5173\u7684\u6210\u672c\u5f71\u54cd\u3002<\/p>\n<h2>5\u3001\u6570\u636e\u9002\u8005\u751f\u5b58<\/h2>\n<p>\u8ba9\u6211\u4eec\u6f84\u6e05\u4ec0\u4e48\u662f\u6570\u636e\u6f14\u53d8\u4ee5\u53ca\u4e3a\u4ec0\u4e48\u5b83\u5bf9\u4f7f\u7528 LLM \u8fdb\u884c\u5408\u6210\u6570\u636e\u751f\u6210\u5982\u6b64\u91cd\u8981\u3002\u6570\u636e\u6f14\u53d8\u6700\u65e9\u5728 Microsoft \u7684 \u4e2d\u5f15\u5165\uff0c\u5b83\u6d89\u53ca\u8fed\u4ee3\u589e\u5f3a\u73b0\u6709\u67e5\u8be2\u96c6\uff0c\u4ee5\u901a\u8fc7\u5feb\u901f\u5de5\u7a0b\u751f\u6210\u66f4\u590d\u6742\u548c\u591a\u6837\u5316\u7684\u67e5\u8be2\u3002\u6b64\u6b65\u9aa4\u5bf9\u4e8e\u786e\u4fdd\u8d28\u91cf\u81f3\u5173\u91cd\u8981<\/p>\n<p>\u6570\u636e\u96c6\u7684\u771f\u5b9e\u6027\u3001\u5168\u9762\u6027\u3001\u590d\u6742\u6027\u548c\u591a\u6837\u6027\u3002\u8fd9\u5c31\u662f\u5408\u6210\u6570\u636e\u4f18\u4e8e\u516c\u5171\u6216\u4eba\u5de5\u6ce8\u91ca\u7684\u6570\u636e\u96c6\u7684\u539f\u56e0\u3002<\/p>\n<p>\u4e8b\u5b9e\u4e0a\uff0c\u539f\u4f5c\u8005\u4ec5\u4ece 175 \u4e2a\u4eba\u5de5\u521b\u5efa\u7684\u67e5\u8be2\u4e2d\u5c31\u6210\u529f\u751f\u6210\u4e86 250,000 \u6761\u6307\u4ee4\u3002\u6570\u636e\u6f14\u5316\u6709 3 \u79cd\u7c7b\u578b\uff1a<\/p>\n<ul>\n<li>\u6df1\u5ea6\u6f14\u5316\uff1a\u5c06\u7b80\u5355\u6307\u4ee4\u6269\u5c55\u4e3a\u66f4\u8be6\u7ec6\u548c\u66f4\u590d\u6742\u7684\u7248\u672c\u3002<\/li>\n<li>\u5e7f\u5ea6\u6f14\u5316\uff1a\u751f\u6210\u65b0\u7684\u3001\u591a\u6837\u5316\u7684\u6307\u4ee4\u4ee5\u4e30\u5bcc\u6570\u636e\u96c6\u3002<\/li>\n<li>\u6d88\u9664\u6f14\u5316\uff1a\u5220\u9664\u6548\u7387\u8f83\u4f4e\u6216\u5931\u8d25\u7684\u6307\u4ee4\u3002<\/li>\n<\/ul>\n<p>  \u201c1+1\u201d\u67e5\u8be2\u7684\u6df1\u5ea6\uff08\u84dd\u8272\uff09\u548c\u5e7f\u5ea6\uff08\u7ea2\u8272\uff09\u6f14\u5316 <\/p>\n<p>\u6709\u51e0\u79cd\u65b9\u6cd5\u53ef\u4ee5\u6267\u884c\u6df1\u5ea6\u6f14\u5316\uff0c\u4f8b\u5982\u4f7f\u8f93\u5165\u590d\u6742\u5316\u3001\u589e\u52a0\u63a8\u7406\u9700\u6c42\u6216\u6dfb\u52a0\u591a\u4e2a\u6b65\u9aa4\u6765\u5b8c\u6210\u4efb\u52a1\u3002\u6bcf\u79cd\u65b9\u6cd5\u90fd\u6709\u52a9\u4e8e\u63d0\u9ad8\u751f\u6210\u7684\u6570\u636e\u7684\u590d\u6742\u7a0b\u5ea6\u3002<\/p>\n<p>\u6df1\u5ea6\u6f14\u5316\u786e\u4fdd\u521b\u5efa\u7ec6\u81f4\u5165\u5fae\u7684\u9ad8\u8d28\u91cf\u67e5\u8be2\uff0c\u800c\u5e7f\u5ea6\u6f14\u5316\u5219\u589e\u5f3a\u4e86\u591a\u6837\u6027\u548c\u5168\u9762\u6027\u3002\u901a\u8fc7\u591a\u6b21\u6f14\u5316\u6bcf\u4e2a\u67e5\u8be2\u6216\u6307\u4ee4\uff0c\u6211\u4eec\u589e\u52a0\u4e86\u5176\u590d\u6742\u6027\uff0c\u4ece\u800c\u4ea7\u751f\u4e86\u4e30\u5bcc\u800c\u591a\u65b9\u9762\u7684\u6570\u636e\u96c6\u3002\u4e0d\u8fc7\uff0c\u6211\u8bf4\u5f97\u591f\u591a\u4e86\uff0c\u8ba9\u6211\u4eec\u5411\u60a8\u5c55\u793a\u5982\u4f55\u5c06\u4e00\u5207\u4ed8\u8bf8\u884c\u52a8\u3002<\/p>\n<p>\u4ee5\u8fd9\u4e2a\u67e5\u8be2\u4e3a\u4f8b\uff1a<\/p>\n<blockquote><p>\n  1+1 \u662f\u4ec0\u4e48\uff1f\n<\/p><\/blockquote>\n<p>\u6211\u4eec\u53ef\u4ee5\u5c06\u5176\u6df1\u5ea6\u6f14\u5316\u4e3a\u7c7b\u4f3c\u8fd9\u6837\u7684\u5185\u5bb9\uff1a<\/p>\n<blockquote><p>\n  \u5728\u4ec0\u4e48\u60c5\u51b5\u4e0b 1+1 \u4e0d\u7b49\u4e8e 2\uff1f\n<\/p><\/blockquote>\n<p>\u6211\u5e0c\u671b\u6211\u4eec\u90fd\u540c\u610f\u8fd9\u6bd4\u4e00\u822c\u7684 1+1 \u66f4\u590d\u6742\u3001\u66f4\u73b0\u5b9e\u3002\u5728\u4e0b\u4e00\u8282\u4e2d\uff0c\u6211\u4eec\u5c06\u5c55\u793a\u5982\u4f55\u5728\u751f\u6210\u5408\u6210\u6570\u636e\u96c6\u65f6\u5b9e\u9645\u4f7f\u7528\u8fd9\u4e9b\u6f14\u5316\u65b9\u6cd5\u3002<\/p>\n<h2>6\u3001\u5206\u6b65\u6307\u5357\uff1a\u4f7f\u7528 LLM \u751f\u6210\u5408\u6210\u6570\u636e<\/h2>\n<p>\u5728\u5f00\u59cb\u4e4b\u524d\uff0c\u8ba9\u6211\u4eec\u56de\u987e\u4e00\u4e0b\u6211\u4eec\u5c06\u8981\u6784\u5efa\u7684\u6570\u636e\u5408\u6210\u5668\u67b6\u6784\uff1a<\/p>\n<p>  \u6570\u636e\u5408\u6210\u5668\u67b6\u6784 <\/p>\n<p>\u4f60\u4f1a\u6ce8\u610f\u5230\u6709\u4e94\u4e2a\u4e3b\u8981\u6b65\u9aa4\uff1a<\/p>\n<ul>\n<li>\u6587\u6863\u5206\u5757<\/li>\n<li>\u4e0a\u4e0b\u6587\u751f\u6210<\/li>\n<li>\u67e5\u8be2\u751f\u6210<\/li>\n<li>\u6570\u636e\u6f14\u5316<\/li>\n<li>\u6807\u7b7e\/\u9884\u671f\u8f93\u51fa\u751f\u6210\uff08\u53ef\u9009\uff09<\/li>\n<\/ul>\n<p>\u5bf9\u4e8e\u90a3\u4e9b\u60f3\u8981\u7acb\u5373\u5f00\u59cb\u5de5\u4f5c\u7684\u4eba\uff0c\u6211\u5df2\u7ecf\u5728 DeepEval \u4e2d\uff0c\u4f60\u53ef\u4ee5\u7acb\u5373\u51c6\u5907\u597d\u751f\u6210\u5408\u6210\u6570\u636e\u96c6\uff0c\u5e76\u652f\u6301\u8fc7\u6ee4\u548c\u6837\u5f0f\uff08\u6211\u5c06\u5728\u6700\u540e\u4e00\u8282\u4e2d\u5411\u4f60\u5c55\u793a\uff09\u3002<\/p>\n<pre><code>from deepeval.synthesizer import Synthesizer\n\nsynthesizer = Synthesizer()\nsynthesizer.generate_goldens_from_docs(\n    document_paths=['example.txt', 'example.docx', 'example.pdf'],\n)<\/code><\/pre>\n<p>\u5982\u679c\u4f60\u6765\u8fd9\u91cc\u662f\u4e3a\u4e86\u4e86\u89e3\u5b83\u7684\u5de5\u4f5c\u539f\u7406\uff0c\u8bf7\u7ee7\u7eed\u9605\u8bfb\u3002<\/p>\n<h3>6.1 \u6587\u6863\u5206\u5757<\/h3>\n<p>\u7b2c\u4e00\u6b65\u662f\u5206\u5757\u4f60\u7684\u6587\u6863\u3002\u987e\u540d\u601d\u4e49\uff0c\u6587\u6863\u5206\u5757\u610f\u5473\u7740\u5c06\u5176\u5206\u6210\u66f4\u5c0f\u3001\u66f4\u6709\u610f\u4e49\u7684\u201c\u5757\u201d\u3002\u8fd9\u6837\uff0c\u4f60\u53ef\u4ee5\u5c06\u8f83\u5927\u7684\u6587\u6863\u5206\u89e3\u4e3a\u53ef\u7ba1\u7406\u7684\u5b50\u6587\u6863\uff0c\u540c\u65f6\u4fdd\u7559\u5176\u4e0a\u4e0b\u6587\u3002\u5206\u5757\u8fd8\u5141\u8bb8\u5728\u8d85\u51fa\u5d4c\u5165\u6a21\u578b\u7684\u6807\u8bb0\u9650\u5236\u7684\u6587\u6863\u4e2d\u751f\u6210\u5d4c\u5165\u3002<\/p>\n<p>\u6b64\u6b65\u9aa4\u81f3\u5173\u91cd\u8981\uff0c\u56e0\u4e3a\u5b83\u6709\u52a9\u4e8e\u8bc6\u522b\u8bed\u4e49\u76f8\u4f3c\u7684\u5757\u5e76\u6839\u636e\u5171\u4eab\u4e0a\u4e0b\u6587\u751f\u6210\u67e5\u8be2\u6216\u4efb\u52a1\u3002<\/p>\n<p>\u6709\u51e0\u79cd\u5206\u5757\u7b56\u7565\uff0c\u5982\u56fa\u5b9a\u5927\u5c0f\u5206\u5757\u548c\u4e0a\u4e0b\u6587\u611f\u77e5\u5206\u5757\u3002\u4f60\u8fd8\u53ef\u4ee5\u8c03\u6574\u8d85\u53c2\u6570\uff0c\u4f8b\u5982\u5b57\u7b26\u5927\u5c0f\u548c\u5757\u91cd\u53e0\u3002\u5728\u4e0b\u9762\u7684\u793a\u4f8b\u4e2d\uff0c\u6211\u4eec\u5c06\u4f7f\u7528\u57fa\u4e8e token \u7684\u5206\u5757\uff0c\u5b57\u7b26\u5927\u5c0f\u4e3a 1024\uff0c\u6ca1\u6709\u91cd\u53e0\u3002\u4ee5\u4e0b\u662f\u5206\u5757\u6587\u6863\u7684\u65b9\u6cd5\uff1a<\/p>\n<pre><code>pip install langchain langchain_openai<\/code><\/pre>\n<pre><code># Step 1. Chunk Documents\nfrom langchain_community.document_loaders import PyPDFLoader\nfrom langchain_text_splitters import TokenTextSplitter\n\ntext_splitter = TokenTextSplitter(chunk_size=\n1024\n, chunk_overlap=\n0\n)\nloader = PyPDFLoader(\"chatbot_information.pdf\")\nraw_chunks = loader.load_and_split(text_splitter)<\/code><\/pre>\n<p>\u83b7\u5f97\u5206\u5757\u540e\uff0c\u5c06\u6bcf\u4e2a\u5206\u5757\u8f6c\u6362\u4e3a\u5d4c\u5165\u3002\u8fd9\u4e9b\u5d4c\u5165\u6355\u83b7\u6bcf\u4e2a\u5206\u5757\u7684\u8bed\u4e49\u542b\u4e49\uff0c\u5e76\u4e0e\u5206\u5757\u7684\u5185\u5bb9\u7ec4\u5408\u4ee5\u5f62\u6210\u4e00\u4e2a Chunk \u5bf9\u8c61\u5217\u8868\u3002<\/p>\n<pre><code>from langchain_openai import OpenAIEmbeddings\n...\n\nembedding_model = OpenAIEmbeddings(api_key=\"...\")\ncontent = [rc.page_content for rc in raw_chunks]\nembeddings = embedding_model.embed_documents(content)<\/code><\/pre>\n<h3>6.2 \u4e0a\u4e0b\u6587\u751f\u6210<\/h3>\n<p>\u8981\u751f\u6210\u4e0a\u4e0b\u6587\uff0c\u9996\u5148\u968f\u673a\u9009\u62e9\u4e00\u5757\u6570\u636e\u4f5c\u4e3a\u67e5\u627e\u76f8\u5173\u4fe1\u606f\u7684\u7126\u70b9\u951a\u70b9\u3002<\/p>\n<pre><code># Step 2: Generate context by selecting chunks\nimport random\n...\n\nreference_index = random.randint(\n0\n, len(embeddings) - \n1\n)\nreference_embedding = embeddings[reference_index]\ncontexts = [content[reference_index]]<\/code><\/pre>\n<p>\u63a5\u4e0b\u6765\uff0c\u8bbe\u7f6e\u76f8\u4f3c\u5ea6\u9608\u503c\u5e76\u4f7f\u7528\u4f59\u5f26\u76f8\u4f3c\u5ea6\u6765\u8bc6\u522b\u76f8\u5173\u5757\u4ee5\u6784\u5efa\u4e0a\u4e0b\u6587\uff1a<\/p>\n<pre><code>...\n\nsimilarity_threshold = \n0.8\n\nsimilar_indices = []\nfor i, embedding in enumerate(embeddings):\n    product = np.dot(reference_embedding, embedding)\n    norm = np.linalg.norm(reference_embedding) * np.linalg.norm(embedding)\n    similarity = product \/ norm\n    if similarity &gt;= similarity_threshold:\n        similar_indices.append(i)\n\nfor i in similar_indices:\n    contexts.append(content[i])<\/code><\/pre>\n<p>\u6b64\u6b65\u9aa4\u81f3\u5173\u91cd\u8981\uff0c\u56e0\u4e3a\u5b83\u5141\u8bb8\u4f60\u901a\u8fc7\u56f4\u7ed5\u540c\u4e00\u4e3b\u9898\u591a\u6837\u5316\u4fe1\u606f\u6e90\u6765\u589e\u5f3a\u67e5\u8be2\u7684\u7a33\u5065\u6027\u3002\u901a\u8fc7\u5305\u542b\u5177\u6709\u76f8\u4f3c\u4e3b\u9898\u7684\u591a\u4e2a\u6570\u636e\u5757\uff0c\u4f60\u8fd8\u53ef\u4ee5\u4e3a\u6a21\u578b\u63d0\u4f9b\u6709\u5173\u8be5\u4e3b\u9898\u7684\u66f4\u4e30\u5bcc\u3001\u66f4\u7ec6\u81f4\u5165\u5fae\u7684\u4fe1\u606f\u3002<\/p>\n<p>\u8fd9\u53ef\u786e\u4fdd\u4f60\u7684\u67e5\u8be2\u5168\u9762\u6db5\u76d6\u8be5\u4e3b\u9898\uff0c\u4ece\u800c\u4ea7\u751f\u66f4\u5168\u9762\u548c\u51c6\u786e\u7684\u54cd\u5e94\u3002<\/p>\n<h3>6.3 \u67e5\u8be2\u751f\u6210<\/h3>\n<p>\u73b0\u5728\u662f LLM \u7684\u6709\u8da3\u90e8\u5206\u3002\u4f7f\u7528 GPT \u6a21\u578b\u4e3a\u4f7f\u7528\u7ed3\u6784\u5316\u63d0\u793a\u521b\u5efa\u7684\u4e0a\u4e0b\u6587\u751f\u6210\u4e00\u7cfb\u5217\u4efb\u52a1\u6216\u67e5\u8be2\u3002<\/p>\n<p>\u63d0\u4f9b\u4e00\u4e2a\u63d0\u793a\uff0c\u8981\u6c42\u6a21\u578b\u5145\u5f53\u6587\u6848\u64b0\u5199\u8005\uff0c\u751f\u6210\u5305\u542b\u8f93\u5165\u952e\uff08\u5373\u67e5\u8be2\uff09\u7684 JSON \u5bf9\u8c61\u3002\u6bcf\u4e2a\u8f93\u5165\u90fd\u5e94\u8be5\u662f\u4f7f\u7528\u63d0\u4f9b\u7684\u4e0a\u4e0b\u6587\u53ef\u4ee5\u56de\u7b54\u7684\u95ee\u9898\u6216\u8bed\u53e5\u3002<\/p>\n<pre><code># Step 3. Generate a series of queries for similar chunks\nfrom langchain_openai import ChatOpenAI\n...\n\nprompt = \nf\n\"\"\"I want you act as a copywriter. Based on the given context, \nwhich is list of strings, please generate a list of JSON objects \nwith a `input` key. The `input` can either be a question or a \nstatement that can be addressed by the given context.\n\ncontexts:\n{contexts}\"\"\"\n\nquery = ChatOpenAI(openai_api_key=\"...\").invoke(prompt)<\/code><\/pre>\n<p>\u6b64\u6b65\u9aa4\u6784\u6210\u67e5\u8be2\u7684\u57fa\u7840\uff0c\u67e5\u8be2\u5c06\u4e0d\u65ad\u53d1\u5c55\u5e76\u5305\u542b\u5728\u6700\u7ec8\u6570\u636e\u96c6\u4e2d\u3002<\/p>\n<h3>6.4 \u67e5\u8be2\u6f14\u5316<\/h3>\n<p>\u6700\u540e\uff0c\u6211\u4eec\u5c06\u4f7f\u7528\u591a\u4e2a\u6f14\u5316\u6a21\u677f\u6f14\u5316\u6b65\u9aa4 3 \u4e2d\u7684\u67e5\u8be2\u3002\u4f60\u53ef\u4ee5\u6839\u636e\u9700\u8981\u5b9a\u4e49\u4efb\u610f\u6570\u91cf\u7684\u6a21\u677f\uff0c\u4f46\u6211\u4eec\u5c06\u91cd\u70b9\u5173\u6ce8\u4e09\u4e2a\uff1a\u591a\u4e0a\u4e0b\u6587\u7406\u89e3\u3001\u591a\u6b65\u9aa4\u63a8\u7406\u548c\u5047\u8bbe\u573a\u666f\u3002<\/p>\n<pre><code># Evolution prompt templates as strings\nmulti_context_template = \nf\n\"\"\"\nI want you to rewrite the given `input` so that it requires readers to use information from all elements in `Context`.\n\n1. `Input` should require information from all `Context` elements. \n2. `Rewritten Input` must be concise and fully answerable from `Context`. \n3. Do not use phrases like 'based on the provided context.'\n4. `Rewritten Input` should not exceed 15 words.\n\nContext: {context}\nInput: {original_input}\nRewritten Input:\n\"\"\"\n\nreasoning_template = \nf\n\"\"\"\nI want you to rewrite the given `input` so that it explicitly requests multi-step reasoning.\n\n1. `Rewritten Input` should require multiple logical connections or inferences.\n2. `Rewritten Input` should be concise and understandable.\n3. Do not use phrases like 'based on the provided context.'\n4. `Rewritten Input` must be fully answerable from `Context`.\n5. `Rewritten Input` should not exceed 15 words.\n\nContext: {context}\nInput: {original_input}\nRewritten Input:\n\"\"\"\n\nhypothetical_scenario_template = \nf\n\"\"\"\nI want you to rewrite the given `input` to incorporate a hypothetical or speculative scenario.\n\n1. `Rewritten Input` should encourage applying knowledge from `Context` to deduce outcomes.\n2. `Rewritten Input` should be concise and understandable.\n3. Do not use phrases like 'based on the provided context.'\n4. `Rewritten Input` must be fully answerable from `Context`.\n5. `Rewritten Input` should not exceed 15 words.\n\nContext: {context}\nInput: {original_input}\nRewritten Input:\n\"\"\"<\/code><\/pre>\n<p>\u4f60\u53ef\u4ee5\u770b\u5230\u6bcf\u4e2a\u6a21\u677f\u90fd\u5bf9\u8f93\u51fa\u65bd\u52a0\u4e86\u7279\u5b9a\u7684\u7ea6\u675f\u3002\u60a8\u53ef\u4ee5\u6839\u636e\u5e0c\u671b\u8bc4\u4f30\u67e5\u8be2\u5728\u6700\u7ec8\u6570\u636e\u96c6\u4e2d\u663e\u793a\u7684\u65b9\u5f0f\u968f\u610f\u8c03\u6574\u5b83\u4eec\u3002\u6211\u4eec\u5c06\u4f7f\u7528\u8fd9\u4e9b\u6a21\u677f\u591a\u6b21\u6f14\u5316\u539f\u59cb\u67e5\u8be2\uff0c\u6bcf\u6b21\u968f\u673a\u9009\u62e9\u6a21\u677f\u3002<\/p>\n<pre><code># Step 4. Evolve Queries\n...\n\nexample_generated_query = \"How do chatbots use natural language understanding?\"\ncontext = contexts \noriginal_input = example_generated_query \nevolution_templates = [multi_context_template, reasoning_template, hypothetical_scenario_template]\n\n# Number of evolution steps to apply\nnum_evolution_steps = \n3\n\n\n# Function to perform random evolution steps\ndef evolve_query(original_input, context, steps):\n    current_input = original_input\n    for _ in range(steps):\n        # Choose a random (or using custom logic) template from the list\n        chosen_template = random.choice(evolution_templates)\n        # Replace the placeholders with the current context and input\n        evolved_prompt = chosen_template.replace(\"{context}\", str(context)).replace(\"{original_input}\", current_input)\n        # Update the current input with the \"Rewritten Input\" section\n        current_input = ChatOpenAI(openai_api_key=\"...\").invoke(evolved_prompt)\n    return current_input\n\n# Evolve the input by randomly selecting the evolution type\nevolved_query = evolve_query(original_input, context, num_evolution_steps)<\/code><\/pre>\n<p>\u8fd9\u5c31\u662f\u6211\u4eec\u6700\u7ec8\u7684\u8fdb\u5316\u67e5\u8be2\uff01\u91cd\u590d\u6b64\u8fc7\u7a0b\u4ee5\u751f\u6210\u66f4\u591a\u67e5\u8be2\u5e76\u8fdb\u4e00\u6b65\u4f18\u5316\u60a8\u7684\u6570\u636e\u96c6\u3002\u51fa\u4e8e\u8bc4\u4f30\u76ee\u7684\uff0c\u4f60\u9700\u8981\u6b63\u786e\u683c\u5f0f\u5316\u8fd9\u4e9b\u8f93\u5165<\/p>\n<p>\u5c06\u67e5\u8be2\u548c\u4e0a\u4e0b\u6587\u6574\u5408\u5230\u5408\u9002\u7684\u6d4b\u8bd5\u6846\u67b6\u4e2d\u3002<\/p>\n<h3>6.5 \u9884\u671f\u8f93\u51fa\u751f\u6210<\/h3>\n<p>\u867d\u7136\u6b64\u6b65\u9aa4\u662f\u53ef\u9009\u7684\uff0c\u4f46\u6211\u5f3a\u70c8\u5efa\u8bae\u4e3a\u6bcf\u4e2a\u6f14\u5316\u67e5\u8be2\u751f\u6210\u9884\u671f\u8f93\u51fa\u3002\u8fd9\u662f\u56e0\u4e3a\u5bf9\u4e8e\u4eba\u7c7b\u8bc4\u4f30\u8005\u6765\u8bf4\uff0c\u7ea0\u6b63\u548c\u6ce8\u91ca\u9884\u671f\u8f93\u51fa\u6bd4\u4ece\u5934\u5f00\u59cb\u521b\u5efa\u5b83\u4eec\u66f4\u5bb9\u6613\u3002<\/p>\n<pre><code># Step 5. Generate Expected Output\n...\n\n# Define prompt template\nexpected_output_template = \nf\n\"\"\"\nI want you to generate an answer for the given `input`. This answer has to be factually aligned to the provided context.\n\nContext: {context}\nInput: {evolved_query}\nAnswer:\n\"\"\"\n\n# Fill in the values\nprompt = expected_output_template.replace(\"{context}\", str(context)).replace(\"{evolved_query}\", evolved_query)\n\n# Generate expected output\nexpected_output = ChatOpenAI(openai_api_key=\"...\").invoke(prompt)<\/code><\/pre>\n<p>\u4f5c\u4e3a\u6700\u540e\u4e00\u6b65\uff0c\u5c06\u6f14\u5316\u67e5\u8be2\u3001\u4e0a\u4e0b\u6587\u548c\u9884\u671f\u8f93\u51fa\u5408\u5e76\u4e3a\u5408\u6210\u6570\u636e\u96c6\u4e2d\u7684\u6570\u636e\u884c\u3002<\/p>\n<pre><code>from pydantic import BaseModel\nfrom typing import Optional, List\n...\n\nclass SyntheticData(BaseModel):\n\tquery: str\n\texpected_output: Optional[str]\n\tcontext: List[str]\n\nsynthetic_data = SyntheticData(\n\tquery=evolved_query, \n\texpected_output=expected_output, \n\tcontext=context\n)\n\n# Simple implementation of synthetic dataset\nsynthetic_dataset = []\nsynthetic_dataset.append(synthetic_data)<\/code><\/pre>\n<p>\u73b0\u5728\u4f60\u9700\u8981\u505a\u7684\u5c31\u662f\u91cd\u590d\u6b65\u9aa4 1-5\uff0c\u76f4\u5230\u62e5\u6709\u4e00\u4e2a\u5408\u7406\u5927\u5c0f\u7684\u5408\u6210\u6570\u636e\u96c6\uff0c\u4f60\u4ee5\u540e\u53ef\u4ee5\u4f7f\u7528\u5b83\u6765\u8bc4\u4f30\u548c\u6d4b\u8bd5\u4f60\u7684 LLM\uff08\u7cfb\u7edf\uff09\uff01<\/p>\n<h2>7\u3001\u4f7f\u7528 DeepEval \u751f\u6210\u5408\u6210\u6570\u636e\u96c6<\/h2>\n<p>\u5728\u6700\u540e\u4e00\u8282\u4e2d\uff0c\u6211\u60f3\u5411\u60a8\u5c55\u793a\u4e00\u4e2a\u4e45\u7ecf\u8003\u9a8c\u7684\u6570\u636e\u5408\u6210\u5668\uff0c\u6211\u5728 \u4e2d\u5c06\u5176\u5f00\u6e90\u3002\u8fd9\u5305\u62ec\u4ece\u5408\u6210\u6570\u636e\u751f\u6210\u5230\u5c06\u5176\u683c\u5f0f\u5316\u4e3a\u53ef\u7528\u4e8e LLM \u8bc4\u4f30\u548c\u6d4b\u8bd5\u7684\u6d4b\u8bd5\u7528\u4f8b\uff0c\u4f60\u53ea\u9700 2 \u884c\u4ee3\u7801\u5373\u53ef\u4f7f\u7528\u3002\u6700\u597d\u7684\u90e8\u5206\u662f\uff0c\u4f60\u53ef\u4ee5\u5229\u7528\u4f60\u9009\u62e9\u7684\u4efb\u4f55 LLM\u3002\u4ee5\u4e0b\u662f\u4f7f\u7528 DeepEval \u751f\u6210\u5408\u6210\u6570\u636e\u96c6\u7684\u65b9\u6cd5\uff1a<\/p>\n<pre><code>pip install deepeval<\/code><\/pre>\n<pre><code>from deepeval.synthesizer import Synthesizer\n\nsynthesizer = Synthesizer()\nsynthesizer.generate_goldens_from_docs(\n    document_paths=['example.txt', 'example.docx', 'example.pdf'],\n)<\/code><\/pre>\n<p>\u4f60\u53ef\u4ee5\u5728 DeepEval \u7684\u6587\u6863\u4e2d\u9605\u8bfb\u6709\u5173\u5982\u4f55\u4f7f\u7528 DeepEval \u7684\u5408\u6210\u5668\u751f\u6210\u5408\u6210\u6570\u636e\u96c6\u7684\u66f4\u591a\u4fe1\u606f\uff0c\u4f46\u603b\u800c\u8a00\u4e4b\uff0cDeepEval \u4f1a\u63a5\u6536\u60a8\u7684\u6587\u6863\uff0c\u4e3a\u4f60\u5b8c\u6210\u6240\u6709\u5206\u5757\u548c\u4e0a\u4e0b\u6587\u751f\u6210\uff0c\u7136\u540e\u751f\u6210\u5408\u6210\u7684\u201c\u9ec4\u91d1\u201d\uff0c\u8fd9\u4e9b\u9ec4\u91d1\u57fa\u672c\u4e0a\u662f\u6700\u7ec8\u5f62\u6210\u5408\u6210\u6570\u636e\u96c6\u7684\u6570\u636e\u884c\u3002\u591f\u7b80\u5355\u5417\uff1f<\/p>\n<h2>8\u3001\u7ed3\u675f\u8bed<\/h2>\n<p>\u4f7f\u7528 LLM \u751f\u6210\u5408\u6210\u6570\u636e\u96c6\u975e\u5e38\u68d2\uff0c\u56e0\u4e3a\u5b83\u662f\u4e00\u79cd\u5feb\u901f\u4e14\u5ec9\u4ef7\u7684\u83b7\u53d6\u5927\u91cf\u6570\u636e\u7684\u65b9\u6cd5\u3002\u4f46\u662f\uff0c\u751f\u6210\u7684\u6570\u636e\u770b\u8d77\u6765\u975e\u5e38\u91cd\u590d\uff0c\u800c\u4e14\u5f80\u5f80\u4e0d\u80fd\u5f88\u597d\u5730\u4ee3\u8868\u5e95\u5c42\u6570\u636e\u5206\u5e03\uff0c\u56e0\u6b64\u4e0d\u88ab\u89c6\u4e3a\u6709\u7528\u3002\u5728\u672c\u6587\u4e2d\uff0c\u6211\u4eec\u8ba8\u8bba\u4e86\u5982\u4f55\u901a\u8fc7\u9996\u5148\u4ece\u6587\u6863\u4e2d\u9009\u62e9\u76f8\u5173\u4e0a\u4e0b\u6587\u6765\u89e3\u51b3\u6b64\u95ee\u9898\uff0c\u7136\u540e\u4f7f\u7528\u5b83\u6765\u751f\u6210\u53ef\u7528\u4e8e\u6d4b\u8bd5\u548c\u8bc4\u4f30 LLM \u7cfb\u7edf\u7684\u67e5\u8be2\u3002<\/p>\n<p>\u6211\u4eec\u8fd8\u63a2\u8ba8\u4e86\u6570\u636e\u6f14\u53d8\uff0c\u6211\u4eec\u7528\u5b83\u6765\u4f7f\u5408\u6210\u67e5\u8be2\u66f4\u52a0\u903c\u771f\u3002\u5982\u679c\u4f60\u60f3\u4ece\u5934\u5f00\u59cb\u6784\u5efa\u6570\u636e\u5408\u6210\u5668\uff0c\u672c\u6587\u5c06\u662f\u4e00\u4e2a\u5f88\u597d\u7684\u6559\u7a0b\u3002\u4f46\u662f\uff0c\u5982\u679c\u4f60\u6b63\u5728\u5bfb\u627e\u66f4\u5f3a\u5927\u4e14\u53ef\u7528\u4e8e\u751f\u4ea7\u7684\u4e1c\u897f\uff0c\u53ef\u4ee5\u4f7f\u7528 DeepEval\u3002\u5b83\u662f\u5f00\u6e90\u7684\uff0c\u975e\u5e38\u6613\u4e8e\u4f7f\u7528\uff08\u771f\u7684\uff09\uff0c\u5e76\u4e14\u62e5\u6709\u4e00\u6574\u5957\u8bc4\u4f30\u548c\u6d4b\u8bd5\u5957\u4ef6\uff0c\u4f60\u53ef\u4ee5\u4f7f\u7528\u751f\u6210\u7684\u5408\u6210\u6570\u636e\u96c6\u65e0\u7f1d\u5730\u6d4b\u8bd5\u548c\u8bc4\u4f30\u60a8\u7684 LLM \u7cfb\u7edf\u3002<\/p>\n<hr>\n<p>\n","protected":false},"excerpt":{"rendered":"<p>\u6784\u5efa\u5927\u89c4\u6a21\u3001\u5168\u9762\u7684\u6570\u636e\u96c6\u6765\u6d4b\u8bd5 LLM \u8f93\u51fa\u53ef\u80fd\u662f\u4e00\u4e2a\u8d39\u529b\u3001\u6602\u8d35\u4e14\u5177\u6709\u6311\u6218\u6027\u7684\u8fc7\u7a0b\uff0c\u5c24\u5176\u662f\u4ece\u5934\u5f00\u59cb\u6784\u5efa\u65f6\u3002\u4f46\u662f\uff0c\u5982\u679c\u6211\u544a\u8bc9\u4f60\uff0c\u73b0\u5728\u53ea\u9700\u51e0\u5206\u949f\u5c31\u53ef\u4ee5\u751f\u6210\u901a\u5e38\u82b1\u8d39\u6570\u5468\u7cbe\u5fc3\u5236\u4f5c\u7684\u6570\u5343\u4e2a\u9ad8\u8d28\u91cf\u6d4b\u8bd5\u7528\u4f8b\uff0c\u4f60\u4f1a\u600e\u4e48\u60f3\uff1f \u5408\u6210\u6570\u636e\u751f\u6210\u5229\u7528 LLM \u521b\u5efa\u9ad8\u8d28\u91cf\u6570\u636e\uff0c\u800c\u65e0\u9700\u624b\u52a8\u6536\u96c6\u3001\u6e05\u7406\u548c\u6807\u6ce8\u5927\u91cf\u6570\u636e\u96c6\u3002\u501f\u52a9 GPT-4 \u7b49\u6a21\u578b\uff0c\u73b0\u5728\u53ef\u4ee5\u5728\u66f4\u77ed\u7684\u65f6\u95f4\u5185\u5408\u6210\u751f\u6210\u6bd4\u4eba\u5de5\u6807\u8bb0\u6570\u636e\u96c6\u66f4\u5168\u9762\u3001\u66f4\u591a\u6837\u5316\u7684\u6570\u636e\u96c6\uff0c\u8fd9\u4e9b\u6570\u636e\u96c6\u53ef\u7528\u4e8e\u501f\u52a9\u4e00\u4e9b LLM \u8bc4\u4f30\u6307\u6807\u5bf9 LLM\uff08\u7cfb\u7edf\uff09\u8fdb\u884c\u57fa\u51c6\u6d4b\u8bd5\u3002 \u5728\u672c\u6587\u4e2d\uff0c\u6211\u5c06\u6559\u4f60\u6709\u5173\u5982\u4f55\u4f7f\u7528 LLM \u751f\u6210\u5408\u6210\u6570\u636e\u96c6\uff08\u4f8b\u5982\uff0c\u53ef\u7528\u4e8e\u8bc4\u4f30 RAG \u7ba1\u9053\uff09\u7684\u6240\u6709\u77e5\u8bc6\u3002\u6211\u4eec\u5c06\u63a2\u7d22\uff1a \u5408\u6210\u751f\u6210\u65b9\u6cd5\uff08\u84b8\u998f\u548c\u81ea\u6211\u6539\u8fdb\uff09 \u4ec0\u4e48\u662f\u6570\u636e\u6f14\u5316\u3001\u5404\u79cd\u6f14\u5316\u6280\u672f\u53ca\u5176\u5728\u5408\u6210\u6570\u636e\u751f\u6210\u4e2d\u7684\u4f5c\u7528 \u4f7f\u7528 LLM \u4ece\u5934\u5f00\u59cb\u200b\u200b\u521b\u5efa\u9ad8\u8d28\u91cf\u5408\u6210\u6570\u636e\u7684\u5206\u6b65\u6559\u7a0b\u3002 \u5982\u4f55\u4f7f\u7528 DeepEval \u5728\u4e0d\u5230 5 \u884c\u4ee3\u7801\u4e2d\u751f\u6210\u5408\u6210\u6570\u636e\u96c6\u3002 \u611f\u5174\u8da3\u5417\uff1f\u8ba9\u6211\u4eec\u6df1\u5165\u4e86\u89e3\u3002 1\u3001\u4ec0\u4e48\u662f\u4f7f\u7528 LLM \u7684\u5408\u6210\u6570\u636e\u751f\u6210\uff1f \u4f7f\u7528 LLM \u8fdb\u884c\u5408\u6210\u6570\u636e\u751f\u6210\u6d89\u53ca\u4f7f\u7528 LLM \u521b\u5efa\u4eba\u5de5\u6570\u636e\uff0c\u8fd9\u4e9b\u4eba\u5de5\u6570\u636e\u901a\u5e38\u662f\u53ef\u7528\u4e8e\u8bad\u7ec3\u3001\u5fae\u8c03\u751a\u81f3\u8bc4\u4f30 LLM \u672c\u8eab\u7684\u6570\u636e\u96c6\u3002\u751f\u6210\u5408\u6210\u6570\u636e\u96c6\u4e0d\u4ec5\u6bd4\u641c\u7d22\u516c\u5171\u6570\u636e\u96c6\u66f4\u5feb\u3001\u6bd4\u4eba\u5de5\u6807\u6ce8\u66f4\u4fbf\u5b9c\uff0c\u800c\u4e14\u8fd8\u80fd\u63d0\u9ad8\u8d28\u91cf\u548c\u6570\u636e\u591a\u6837\u6027\uff0c\u8fd9\u5bf9\u4e8e\u7ea2\u961f LLM \u5e94\u7528\u7a0b\u5e8f\u4e5f\u662f\u5fc5\u4e0d\u53ef\u5c11\u7684\u3002 \u8be5\u8fc7\u7a0b\u4ece\u521b\u5efa\u5408\u6210\u67e5\u8be2\u5f00\u59cb\uff0c\u8fd9\u4e9b\u67e5\u8be2\u662f\u4f7f\u7528\u77e5\u8bc6\u5e93\u4e2d\u7684\u4e0a\u4e0b\u6587\uff08\u901a\u5e38\u4ee5\u6587\u6863\u7684\u5f62\u5f0f\uff09\u4f5c\u4e3a\u57fa\u672c\u4e8b\u5b9e\u751f\u6210\u7684\u3002\u7136\u540e\uff0c\u751f\u6210\u7684\u67e5\u8be2\u4f1a\u7ecf\u8fc7\u591a\u6b21\u201c\u6f14\u5316\u201d\uff0c\u4f7f\u5176\u590d\u6742\u5316\u548c\u903c\u771f\u5316\uff0c\u5e76\u4e0e\u751f\u6210\u67e5\u8be2\u7684\u539f\u59cb\u4e0a\u4e0b\u6587\u76f8\u7ed3\u5408\uff0c\u6784\u6210\u6700\u7ec8\u7684\u5408\u6210\u6570\u636e\u96c6\u3002\u867d\u7136\u8fd9\u662f\u53ef\u9009\u7684\uff0c\u4f46\u4f60\u4e5f\u53ef\u4ee5\u9009\u62e9\u4e3a\u6bcf\u4e2a\u5408\u6210\u67e5\u8be2-\u4e0a\u4e0b\u6587\u5bf9\u751f\u6210\u4e00\u4e2a\u76ee\u6807\u6807\u7b7e\uff0c\u8be5\u6807\u7b7e\u5c06\u4f5c\u4e3a\u7ed9\u5b9a\u67e5\u8be2\u7684 LLM \u7cfb\u7edf\u7684\u9884\u671f\u8f93\u51fa\u3002 \u6570\u636e\u5408\u6210\u5668\u67b6\u6784 \u5728\u751f\u6210\u7528\u4e8e\u8bc4\u4f30\u7684\u5408\u6210\u6570\u636e\u96c6\u65f6\uff0c\u4e3b\u8981\u6709\u4e24\u79cd\u65b9\u6cd5\uff1a\u4f7f\u7528\u6a21\u578b\u8f93\u51fa\u8fdb\u884c\u81ea\u6211\u6539\u8fdb\uff0c\u6216\u4ece\u66f4\u9ad8\u7ea7\u7684\u6a21\u578b\u4e2d\u8fdb\u884c\u63d0\u70bc\u3002 \u81ea\u6211\u6539\u8fdb\uff1a\u6d89\u53ca\u60a8\u7684\u6a21\u578b\u4ece\u5176\u81ea\u8eab\u8f93\u51fa\u8fed\u4ee3\u751f\u6210\u6570\u636e\uff0c\u800c\u65e0\u9700\u5916\u90e8\u4f9d\u8d56 \u84b8\u998f\uff1a\u6d89\u53ca\u4f7f\u7528\u66f4\u5f3a\u5927\u7684\u6a21\u578b\u751f\u6210\u5408\u6210\u6570\u636e\u4ee5\u8bc4\u4f30\u8f83\u5f31\u7684\u6a21\u578b \u81ea\u6211\u6539\u8fdb\u65b9\u6cd5\uff08\u4f8b\u5982 Self-Instruct \u6216 SPIN\uff09\u53d7\u5230\u6a21\u578b\u529f\u80fd\u7684\u9650\u5236\uff0c\u53ef\u80fd\u4f1a\u53d7\u5230\u504f\u5dee\u548c\u9519\u8bef\u653e\u5927\u7684\u5f71\u54cd\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0c\u84b8\u998f\u6280\u672f\u4ec5\u53d7\u53ef\u7528\u7684\u6700\u4f73\u6a21\u578b\u7684\u9650\u5236\uff0c\u53ef\u786e\u4fdd\u6700\u9ad8\u8d28\u91cf\u7684\u751f\u6210\u3002 2\u3001\u4ece\u77e5\u8bc6\u5e93\u751f\u6210\u6570\u636e \u5408\u6210\u6570\u636e\u751f\u6210\u7684\u7b2c\u4e00\u6b65\u662f\u4ece\u4e0a\u4e0b\u6587\u5217\u8868\u4e2d\u521b\u5efa\u5408\u6210\u67e5\u8be2\uff0c\u8fd9\u4e9b\u67e5\u8be2\u76f4\u63a5\u6765\u81ea\u77e5\u8bc6\u5e93\u3002\u672c\u8d28\u4e0a\uff0c\u4e0a\u4e0b\u6587\u662f LLM \u5e94\u7528\u7a0b\u5e8f\u7684\u7406\u60f3\u68c0\u7d22\u4e0a\u4e0b\u6587\uff0c\u5c31\u50cf\u9884\u671f\u8f93\u51fa\u662f LLM [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[13],"tags":[],"class_list":["post-53769","post","type-post","status-publish","format-standard","hentry","category-ai"],"_links":{"self":[{"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/posts\/53769","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/comments?post=53769"}],"version-history":[{"count":0,"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/posts\/53769\/revisions"}],"wp:attachment":[{"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/media?parent=53769"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/categories?post=53769"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/tags?post=53769"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}