{"id":14903,"date":"2024-12-07T10:24:04","date_gmt":"2024-12-07T02:24:04","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=14903"},"modified":"2024-12-08T10:36:40","modified_gmt":"2024-12-08T02:36:40","slug":"aigongchengxueyuan21","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/en\/aigongchengxueyuan21\/","title":{"rendered":"AI\u5de5\u7a0b\u5b66\u9662\uff1a2.1\u4ece\u96f6\u5f00\u59cb\u5b9e\u73b0 RAG"},"content":{"rendered":"

\u6982\u8ff0<\/h2>\n

\u672c\u6307\u5357\u5c06\u5f15\u5bfc\u60a8\u4f7f\u7528\u7eaf Python \u521b\u5efa\u4e00\u4e2a\u7b80\u5355\u7684\u68c0\u7d22\u589e\u5f3a\u751f\u6210 (RAG<\/a>) \u7cfb\u7edf\u3002\u6211\u4eec\u5c06\u4f7f\u7528\u4e00\u4e2a\u5d4c\u5165\u6a21\u578b\u548c\u4e00\u4e2a\u5927\u8bed\u8a00\u6a21\u578b (LLM) \u6765\u68c0\u7d22\u76f8\u5173\u6587\u6863\u5e76\u57fa\u4e8e\u7528\u6237\u7684\u67e5\u8be2\u751f\u6210\u56de\u590d\u3002<\/p>\n

\"\"<\/p>\n

 <\/p>\n

https:\/\/github.com\/adithya-s-k\/AI-Engineering.academy\/tree\/main\/RAG\/00_RAG_from_Scratch<\/p>\n

 <\/p>\n

\u6d89\u53ca\u7684\u6b65\u9aa4<\/h2>\n

\u6574\u4e2a\u8fc7\u7a0b\u53ef\u4ee5\u5206\u4e3a\u4e24\u4e2a\u4e3b\u8981\u6b65\u9aa4\uff1a<\/p>\n

    \n
  1. \u77e5\u8bc6\u5e93\u521b\u5efa<\/strong><\/li>\n
  2. \u751f\u6210\u90e8\u5206<\/strong><\/li>\n<\/ol>\n

    \u77e5\u8bc6\u5e93\u521b\u5efa<\/h3>\n

    \u9996\u5148\uff0c\u60a8\u9700\u8981\u51c6\u5907\u4e00\u4e2a\u77e5\u8bc6\u5e93\uff08\u6587\u6863\u3001PDF\u3001\u7ef4\u57fa\u9875\u9762\uff09\u3002\u8fd9\u4e9b\u662f\u8bed\u8a00\u6a21\u578b (LLM) \u7684\u57fa\u7840\u6570\u636e\u3002\u5177\u4f53\u8fc7\u7a0b\u5305\u62ec\uff1a<\/p>\n

      \n
    • \u5206\u5757<\/strong>\uff1a\u5c06\u6587\u672c\u5206\u6210\u5c0f\u7684\u5b50\u6587\u6863\u5757\u4ee5\u7b80\u5316\u5904\u7406\u3002<\/li>\n
    • \u5d4c\u5165<\/strong>\uff1a\u4e3a\u6bcf\u4e2a\u5b50\u6587\u6863\u5757\u8ba1\u7b97\u6570\u503c\u5d4c\u5165\uff0c\u4ee5\u4fbf\u7406\u89e3\u67e5\u8be2\u7684\u8bed\u4e49\u76f8\u4f3c\u6027\u3002<\/li>\n
    • \u5b58\u50a8<\/strong>\uff1a\u4ee5\u80fd\u591f\u5feb\u901f\u68c0\u7d22\u7684\u65b9\u5f0f\u5b58\u50a8\u8fd9\u4e9b\u5d4c\u5165\u3002\u867d\u7136\u901a\u5e38\u4f1a\u4f7f\u7528\u5411\u91cf\u5b58\u50a8\/\u6570\u636e\u5e93\uff0c\u4f46\u672c\u6559\u7a0b\u8868\u660e\u8fd9\u5e76\u975e\u5fc5\u9700\u3002<\/li>\n<\/ul>\n

      \u751f\u6210\u90e8\u5206<\/h3>\n

      \u5f53\u7528\u6237\u67e5\u8be2\u8f93\u5165\u65f6\uff0c\u4e3a\u67e5\u8be2\u8ba1\u7b97\u5d4c\u5165\uff0c\u5e76\u4ece\u77e5\u8bc6\u5e93\u4e2d\u68c0\u7d22\u6700\u76f8\u5173\u7684\u5b50\u6587\u6863\u5757\u3002\u8fd9\u4e9b\u76f8\u5173\u5757\u4f1a\u88ab\u9644\u52a0\u5230\u7528\u6237\u67e5\u8be2\u540e\uff0c\u5f62\u6210\u4e00\u4e2a\u4e0a\u4e0b\u6587\u5e76\u8f93\u5165\u5230 LLM \u4e2d\u751f\u6210\u56de\u590d\u3002<\/p>\n

       <\/p>\n

      1. \u73af\u5883\u8bbe\u7f6e<\/h2>\n

      \u5728\u5f00\u59cb\u4e4b\u524d\uff0c\u9700\u8981\u5b89\u88c5\u4e00\u4e9b\u5305\u3002<\/p>\n

        \n
      • sentence-transformers<\/code><\/strong>\uff1a\u7528\u4e8e\u4e3a\u6587\u6863\u548c\u67e5\u8be2\u751f\u6210\u5d4c\u5165\u3002<\/li>\n
      • numpy<\/code><\/strong>\uff1a\u7528\u4e8e\u76f8\u4f3c\u6027\u6bd4\u8f83\u3002<\/li>\n
      • scipy<\/code><\/strong>\uff1a\u7528\u4e8e\u9ad8\u7ea7\u76f8\u4f3c\u6027\u8ba1\u7b97\u3002<\/li>\n
      • wikipedia-api<\/code><\/strong>\uff1a\u7528\u4e8e\u5c06\u7ef4\u57fa\u767e\u79d1\u9875\u9762\u52a0\u8f7d\u4e3a\u77e5\u8bc6\u5e93\u3002<\/li>\n
      • textwrap<\/code><\/strong>\uff1a\u7528\u4e8e\u683c\u5f0f\u5316\u8f93\u51fa\u6587\u672c\u3002<\/li>\n<\/ul>\n
        !pip install -q sentence-transformers\r\n!pip install -q wikipedia-api\r\n!pip install -q numpy\r\n!pip install -q scipy\r\n<\/code><\/pre>\n
         <\/p>\n
        2. \u52a0\u8f7d\u5d4c\u5165\u6a21\u578b<\/h2>\n
        \u8ba9\u6211\u4eec\u52a0\u8f7d\u4e00\u4e2a\u5d4c\u5165\u6a21\u578b\u3002\u672c\u6559\u7a0b\u4e2d\u4f7f\u7528\u7684\u662f\u00a0gte-base-en-v1.5<\/code>\u3002<\/p>\n
        from sentence_transformers import SentenceTransformer\r\nmodel = SentenceTransformer(\"Alibaba-NLP\/gte-base-en-v1.5\", trust_remote_code=True)\r\n<\/code><\/pre>\n
        \u5173\u4e8e\u6a21\u578b<\/h3>\n
        gte-base-en-v1.5<\/code>\u00a0\u6a21\u578b\u662f\u7531\u963f\u91cc\u5df4\u5df4 NLP \u56e2\u961f\u63d0\u4f9b\u7684\u5f00\u6e90\u82f1\u8bed\u6a21\u578b\u3002\u5b83\u662f GTE\uff08\u901a\u7528\u6587\u672c\u5d4c\u5165\uff09\u7cfb\u5217\u7684\u4e00\u90e8\u5206\uff0c\u4e13\u4e3a\u751f\u6210\u9ad8\u8d28\u91cf\u5d4c\u5165\u800c\u8bbe\u8ba1\uff0c\u9002\u7528\u4e8e\u5404\u79cd\u81ea\u7136\u8bed\u8a00\u5904\u7406\u4efb\u52a1\u3002\u8be5\u6a21\u578b\u9488\u5bf9\u6355\u6349\u82f1\u8bed\u6587\u672c\u7684\u8bed\u4e49\u610f\u4e49\u8fdb\u884c\u4e86\u4f18\u5316\uff0c\u53ef\u7528\u4e8e\u53e5\u5b50\u76f8\u4f3c\u6027\u3001\u8bed\u4e49\u641c\u7d22\u548c\u805a\u7c7b\u7b49\u4efb\u52a1\u3002trust_remote_code=True<\/code>\u00a0\u53c2\u6570\u5141\u8bb8\u4f7f\u7528\u4e0e\u6a21\u578b\u76f8\u5173\u7684\u81ea\u5b9a\u4e49\u4ee3\u7801\uff0c\u786e\u4fdd\u5176\u6309\u9884\u671f\u8fd0\u884c\u3002<\/p>\n
         <\/p>\n
        3. \u4ece\u7ef4\u57fa\u767e\u79d1\u83b7\u53d6\u6587\u672c\u5185\u5bb9\u5e76\u51c6\u5907<\/h2>\n
          \n
        • \u9996\u5148\u52a0\u8f7d\u4e00\u4e2a\u7ef4\u57fa\u767e\u79d1\u6587\u7ae0\u4f5c\u4e3a\u77e5\u8bc6\u5e93\u3002\u6587\u672c\u5c06\u88ab\u5206\u5272\u4e3a\u6613\u4e8e\u7ba1\u7406\u7684\u5c0f\u5757\uff08\u5b50\u6587\u6863\uff09\uff0c\u901a\u5e38\u6309\u6bb5\u843d\u5206\u5272\u3002\n
          from wikipediaapi import Wikipedia\r\nwiki = Wikipedia('RAGBot\/0.0', 'en')\r\ndoc = wiki.page('Hayao_Miyazaki').text\r\nparagraphs = doc.split('\\n\\n')  # \u5206\u5757\r\n<\/code><\/pre>\n<\/li>\n
        • \u867d\u7136\u6709\u8bb8\u591a\u5206\u5757\u7b56\u7565\u53ef\u7528\uff0c\u4f46\u5176\u4e2d\u5f88\u591a\u672a\u5fc5\u9002\u7528\u3002\u6700\u597d\u68c0\u67e5\u60a8\u7684\u77e5\u8bc6\u5e93 (KB)\uff0c\u786e\u5b9a\u6700\u9002\u5408\u7684\u7b56\u7565\u3002\u5728\u672c\u4f8b\u4e2d\uff0c\u6211\u4eec\u6309\u6bb5\u843d\u5206\u5757\u3002<\/li>\n
        • \u5982\u679c\u60f3\u67e5\u770b\u8fd9\u4e9b\u5757\u7684\u6837\u5b50\uff0c\u53ef\u4ee5\u5bfc\u5165\u00a0textwrap<\/code>\u00a0\u5e93\uff0c\u5e76\u9010\u6bb5\u6253\u5370\u51fa\u6765\u3002\n
          import textwrap\r\nfor i, p in enumerate(paragraphs):\r\nwrapped_text = textwrap.fill(p, width=100)\r\nprint(\"-----------------------------------------------------------------\")\r\nprint(wrapped_text)\r\nprint(\"-----------------------------------------------------------------\")\r\n<\/code><\/pre>\n<\/li>\n
        • \u5982\u679c\u6587\u6863\u4e2d\u5305\u542b\u56fe\u7247\u548c\u8868\u683c\uff0c\u5efa\u8bae\u5355\u72ec\u63d0\u53d6\u5e76\u4f7f\u7528\u89c6\u89c9\u6a21\u578b\u5d4c\u5165\u3002<\/li>\n<\/ul>\n
           <\/p>\n
          4. \u5d4c\u5165\u6587\u6863<\/h2>\n
            \n
          • \u63a5\u4e0b\u6765\uff0c\u901a\u8fc7\u8c03\u7528\u6a21\u578b\u7684\u00a0encode<\/code>\u00a0\u65b9\u6cd5\uff0c\u5c06\u6587\u672c\u6570\u636e\uff08\u4f8b\u5982\u00a0paragraphs<\/code>\uff09\u7f16\u7801\u4e3a\u5d4c\u5165\u3002\n
            docs_embed = model.encode(paragraphs, normalize_embeddings=True)\r\n<\/code><\/pre>\n<\/li>\n
          • \u8fd9\u4e9b\u5d4c\u5165\u662f\u6587\u672c\u7684\u5bc6\u96c6\u5411\u91cf\u8868\u793a\uff0c\u6355\u6349\u4e86\u8bed\u4e49\u610f\u4e49\uff0c\u4f7f\u6a21\u578b\u80fd\u591f\u4ee5\u6570\u5b66\u5f62\u5f0f\u7406\u89e3\u548c\u5904\u7406\u6587\u672c\u3002<\/li>\n
          • \u6211\u4eec\u5728\u6b64\u5bf9\u5d4c\u5165\u8fdb\u884c\u4e86\u5f52\u4e00\u5316\u3002\n
              \n
            • \u4ec0\u4e48\u662f\u5f52\u4e00\u5316\uff1f<\/strong>\u00a0\u5f52\u4e00\u5316\u662f\u4e00\u4e2a\u8c03\u6574\u5d4c\u5165\u503c\u4ee5\u4f7f\u5176\u5177\u6709\u5355\u4f4d\u8303\u6570\u7684\u8fc7\u7a0b\uff08\u5373\u5411\u91cf\u957f\u5ea6\u4e3a 1\uff09\u3002<\/li>\n
            • \u4e3a\u4ec0\u4e48\u5f52\u4e00\u5316\uff1f<\/strong>\u00a0\u5f52\u4e00\u5316\u7684\u5d4c\u5165\u786e\u4fdd\u5411\u91cf\u4e4b\u95f4\u7684\u8ddd\u79bb\u4e3b\u8981\u53cd\u6620\u65b9\u5411\u4e0a\u7684\u5dee\u5f02\uff0c\u800c\u4e0d\u662f\u5927\u5c0f\u4e0a\u7684\u5dee\u5f02\u3002\u8fd9\u53ef\u4ee5\u63d0\u9ad8\u6a21\u578b\u5728\u76f8\u4f3c\u6027\u641c\u7d22\u4efb\u52a1\u4e2d\u7684\u8868\u73b0\uff0c\u5728\u8fd9\u4e9b\u4efb\u52a1\u4e2d\u9700\u8981\u6bd4\u8f83\u6587\u672c\u4e4b\u95f4\u7684\u201c\u63a5\u8fd1\u7a0b\u5ea6\u201d\u6216\u201c\u76f8\u4f3c\u6027\u201d\u3002<\/li>\n<\/ul>\n<\/li>\n
            • \u7ed3\u679c\u00a0docs_embed<\/code>\u00a0\u662f\u4e00\u4e2a\u6587\u672c\u6570\u636e\u7684\u5411\u91cf\u8868\u793a\u96c6\u5408\uff0c\u5176\u4e2d\u6bcf\u4e2a\u5411\u91cf\u5bf9\u5e94\u00a0paragraphs<\/code>\u00a0\u5217\u8868\u4e2d\u7684\u4e00\u4e2a\u6bb5\u843d\u3002<\/li>\n
            • \u4f7f\u7528\u00a0shape<\/code>\u00a0\u547d\u4ee4\u53ef\u4ee5\u67e5\u770b\u5757\u7684\u6570\u91cf\u548c\u6bcf\u4e2a\u5d4c\u5165\u5411\u91cf\u7684\u7ef4\u5ea6\uff08\u5d4c\u5165\u5411\u91cf\u7684\u5927\u5c0f\u53d6\u51b3\u4e8e\u5d4c\u5165\u6a21\u578b\u7684\u7c7b\u578b\uff09\u3002\n
              docs_embed.shape\r\n<\/code><\/pre>\n<\/li>\n
            • \u60a8\u8fd8\u53ef\u4ee5\u67e5\u770b\u5b9e\u9645\u5d4c\u5165\u7684\u6837\u5b50\uff0c\u8fd9\u662f\u4e00\u7ec4\u5f52\u4e00\u5316\u7684\u6570\u503c\u3002\n
              docs_embed[0]\r\n<\/code><\/pre>\n<\/li>\n<\/ul>\n
               <\/p>\n
              5. \u5d4c\u5165\u67e5\u8be2<\/h2>\n
              \u4ee5\u4e0e\u5d4c\u5165\u6587\u6863\u7c7b\u4f3c\u7684\u65b9\u5f0f\u5d4c\u5165\u793a\u4f8b\u7528\u6237\u67e5\u8be2\u3002<\/p>\n
              query = \"What was Studio Ghibli's first film?\"\r\nquery_embed = model.encode(query, normalize_embeddings=True)\r\n<\/code><\/pre>\n
              \u60a8\u53ef\u4ee5\u68c0\u67e5\u00a0query_embed<\/code>\u00a0\u7684\u5f62\u72b6\u4ee5\u786e\u8ba4\u5d4c\u5165\u67e5\u8be2\u7684\u7ef4\u5ea6\u3002<\/p>\n
              query_embed.shape\r\n<\/code><\/pre>\n
               <\/p>\n
              6. \u627e\u5230\u4e0e\u67e5\u8be2\u6700\u63a5\u8fd1\u7684\u6bb5\u843d<\/h2>\n
              \u68c0\u7d22\u6700\u76f8\u5173\u7684\u5185\u5bb9\u5757\u6700\u7b80\u5355\u7684\u65b9\u6cd5\u4e4b\u4e00\u662f\u8ba1\u7b97\u6587\u6863\u5d4c\u5165\u548c\u67e5\u8be2\u5d4c\u5165\u7684\u70b9\u79ef\u3002<\/p>\n
              a. \u8ba1\u7b97\u70b9\u79ef<\/h3>\n
              \u70b9\u79ef\u662f\u4e00\u79cd\u6570\u5b66\u8fd0\u7b97\uff0c\u5c06\u4e24\u4e2a\u5411\u91cf\uff08\u6216\u77e9\u9635\uff09\u7684\u5bf9\u5e94\u5143\u7d20\u76f8\u4e58\u5e76\u6c42\u548c\u3002\u5b83\u901a\u5e38\u7528\u4e8e\u8861\u91cf\u4e24\u4e2a\u5411\u91cf\u4e4b\u95f4\u7684\u76f8\u4f3c\u6027\u3002<\/p>\n
              \uff08\u6ce8\u610f\uff0c\u8ba1\u7b97\u70b9\u79ef\u65f6\u53d6\u4e86\u00a0query_embed<\/code>\u00a0\u5411\u91cf\u7684\u8f6c\u7f6e\uff09\u3002<\/p>\n
              import numpy as np\r\nsimilarities = np.dot(docs_embed, query_embed.T)\r\n<\/code><\/pre>\n
              b. \u7406\u89e3\u70b9\u79ef\u53ca\u5176\u5f62\u72b6<\/h3>\n
              NumPy \u6570\u7ec4\u7684\u00a0.shape<\/code>\u00a0\u5c5e\u6027\u8fd4\u56de\u4e00\u4e2a\u8868\u793a\u6570\u7ec4\u7ef4\u5ea6\u7684\u5143\u7ec4\u3002<\/p>\n
              similarities.shape\r\n<\/code><\/pre>\n
              \u5728\u6b64\u4ee3\u7801\u4e2d\u7684\u9884\u671f\u5f62\u72b6\u5982\u4e0b\uff1a<\/p>\n
                \n
              • \u5982\u679c\u00a0docs_embed<\/code>\u00a0\u7684\u5f62\u72b6\u4e3a (n_docs, n_dim)\uff1a\n
                  \n
                • n_docs \u662f\u6587\u6863\u6570\u91cf\u3002<\/li>\n
                • n_dim \u662f\u6bcf\u4e2a\u6587\u6863\u5d4c\u5165\u7684\u7ef4\u5ea6\u3002<\/li>\n<\/ul>\n<\/li>\n
                • query_embed.T<\/code>\u00a0\u7684\u5f62\u72b6\u5c06\u4e3a (n_dim, 1)\uff0c\u56e0\u4e3a\u6211\u4eec\u662f\u9488\u5bf9\u5355\u4e2a\u67e5\u8be2\u8fdb\u884c\u6bd4\u8f83\u3002<\/li>\n
                • \u70b9\u79ef\u540e\u7684\u00a0similarities<\/code>\u00a0\u6570\u7ec4\u7684\u5f62\u72b6\u5c06\u4e3a (n_docs,)\uff0c\u8868\u793a\u8fd9\u662f\u4e00\u4e2a 1 \u7ef4\u6570\u7ec4\uff08\u5411\u91cf\uff09\uff0c\u5305\u542b n_docs \u4e2a\u5143\u7d20\u3002\u6bcf\u4e2a\u5143\u7d20\u4ee3\u8868\u67e5\u8be2\u4e0e\u67d0\u4e2a\u6587\u6863\u4e4b\u95f4\u7684\u76f8\u4f3c\u5ea6\u5206\u6570\u3002<\/li>\n
                • \u4e3a\u4ec0\u4e48\u68c0\u67e5\u5f62\u72b6\uff1f<\/strong>\u00a0\u786e\u4fdd\u5f62\u72b6\u4e3a\u9884\u671f\u7684 (n_docs,) \u53ef\u4ee5\u786e\u8ba4\u70b9\u79ef\u6267\u884c\u6b63\u786e\uff0c\u5e76\u4e14\u6bcf\u4e2a\u6587\u6863\u7684\u76f8\u4f3c\u5ea6\u5206\u6570\u5df2\u88ab\u6b63\u786e\u8ba1\u7b97\u3002<\/li>\n<\/ul>\n
                  \u60a8\u53ef\u4ee5\u6253\u5370\u00a0similarities<\/code>\u00a0\u6570\u7ec4\u4ee5\u68c0\u67e5\u76f8\u4f3c\u5ea6\u5206\u6570\uff0c\u5176\u4e2d\u6bcf\u4e2a\u503c\u5bf9\u5e94\u4e00\u4e2a\u70b9\u79ef\u7ed3\u679c\uff1a<\/p>\n
                  print(similarities)\r\n<\/code><\/pre>\n
                  c. \u70b9\u79ef\u7684\u89e3\u91ca<\/h3>\n
                  \u4e24\u4e2a\u5411\u91cf\uff08\u5d4c\u5165\uff09\u4e4b\u95f4\u7684\u70b9\u79ef\u8861\u91cf\u5176\u76f8\u4f3c\u6027\uff1a\u8f83\u9ad8\u7684\u503c\u8868\u793a\u67e5\u8be2\u4e0e\u6587\u6863\u4e4b\u95f4\u7684\u76f8\u4f3c\u6027\u8f83\u9ad8\u3002\u5982\u679c\u5d4c\u5165\u5df2\u5f52\u4e00\u5316\uff0c\u8fd9\u4e9b\u503c\u76f4\u63a5\u4e0e\u5411\u91cf\u4e4b\u95f4\u7684\u4f59\u5f26\u76f8\u4f3c\u5ea6\u6210\u6b63\u6bd4\u3002\u5982\u679c\u672a\u5f52\u4e00\u5316\uff0c\u5b83\u4eec\u4ecd\u7136\u8868\u793a\u76f8\u4f3c\u6027\uff0c\u4f46\u4e5f\u53cd\u6620\u4e86\u5d4c\u5165\u7684\u5927\u5c0f\u3002<\/p>\n
                  d. \u627e\u51fa\u6700\u76f8\u4f3c\u7684 3 \u4e2a\u6587\u6863<\/h3>\n
                  \u8981\u6839\u636e\u76f8\u4f3c\u5ea6\u5206\u6570\u627e\u51fa\u6700\u76f8\u4f3c\u7684 3 \u4e2a\u6587\u6863\uff0c\u53ef\u4ee5\u4f7f\u7528\u4ee5\u4e0b\u4ee3\u7801\uff1a<\/p>\n
                  top_3_idx = np.argsort(similarities, axis=0)[-3:][::-1].tolist()\r\n<\/code><\/pre>\n
                    \n
                  • np.argsort(similarities, axis=0):<\/strong>\u00a0\u6b64\u51fd\u6570\u6309\u5347\u5e8f\u5bf9\u00a0similarities<\/code>\u00a0\u6570\u7ec4\u7684\u7d22\u5f15\u8fdb\u884c\u6392\u5e8f\u3002\u4f8b\u5982\uff0c\u5982\u679c\u00a0similarities = [0.1, 0.7, 0.4]<\/code>\uff0cnp.argsort<\/code>\u00a0\u5c06\u8fd4\u56de\u00a0[0, 2, 1]<\/code>\uff0c\u5176\u4e2d 0 \u662f\u6700\u5c0f\u503c\u7684\u7d22\u5f15\uff0c1 \u662f\u6700\u5927\u503c\u7684\u7d22\u5f15\u3002<\/li>\n
                  • [-3:]:<\/strong>\u00a0\u6b64\u5207\u7247\u64cd\u4f5c\u9009\u62e9\u76f8\u4f3c\u5ea6\u5206\u6570\u6700\u9ad8\u7684 3 \u4e2a\u7d22\u5f15\uff08\u6392\u5e8f\u540e\u7684\u6700\u540e 3 \u4e2a\u5143\u7d20\uff09\u3002<\/li>\n
                  • [::-1]:<\/strong>\u00a0\u6b64\u64cd\u4f5c\u53cd\u8f6c\u987a\u5e8f\uff0c\u56e0\u6b64\u7d22\u5f15\u6309\u76f8\u4f3c\u5ea6\u7684\u964d\u5e8f\u6392\u5217\u3002<\/li>\n
                  • tolist():<\/strong>\u00a0\u5c06\u7d22\u5f15\u6570\u7ec4\u8f6c\u6362\u4e3a Python \u5217\u8868\u3002\u7ed3\u679c\uff1atop_3_idx<\/code>\u00a0\u5305\u542b\u6700\u76f8\u4f3c\u7684 3 \u4e2a\u6587\u6863\u7684\u7d22\u5f15\uff0c\u6309\u76f8\u4f3c\u5ea6\u964d\u5e8f\u6392\u5217\u3002<\/li>\n<\/ul>\n
                    e. \u63d0\u53d6\u6700\u76f8\u4f3c\u7684\u6587\u6863<\/h3>\n
                    most_similar_documents = [paragraphs[idx] for idx in top_3_idx]\r\n<\/code><\/pre>\n
                      \n
                    • \u5217\u8868\u63a8\u5bfc\u5f0f\uff1a<\/strong>\u00a0\u6b64\u884c\u521b\u5efa\u4e00\u4e2a\u540d\u4e3a\u00a0most_similar_documents<\/code>\u00a0\u7684\u5217\u8868\uff0c\u5176\u4e2d\u5305\u542b\u00a0paragraphs<\/code>\u00a0\u5217\u8868\u4e2d\u5bf9\u5e94\u4e8e\u00a0top_3_idx<\/code>\u00a0\u7d22\u5f15\u7684\u5b9e\u9645\u6bb5\u843d\u3002<\/li>\n
                    • paragraphs[idx]:<\/strong>\u00a0\u5bf9\u4e8e\u00a0top_3_idx<\/code>\u00a0\u4e2d\u7684\u6bcf\u4e2a\u7d22\u5f15\uff0c\u6b64\u64cd\u4f5c\u68c0\u7d22\u76f8\u5e94\u7684\u6bb5\u843d\u3002<\/li>\n<\/ul>\n
                      f. \u683c\u5f0f\u5316\u5e76\u663e\u793a\u6700\u76f8\u4f3c\u7684\u6587\u6863<\/h3>\n
                      CONTEXT<\/code>\u00a0\u53d8\u91cf\u6700\u521d\u88ab\u521d\u59cb\u5316\u4e3a\u7a7a\u5b57\u7b26\u4e32\uff0c\u968f\u540e\u5c06\u5728\u4e00\u4e2a\u679a\u4e3e\u5faa\u73af\u4e2d\u9644\u52a0\u6700\u76f8\u4f3c\u6587\u6863\u7684\u6362\u884c\u6587\u672c\u3002<\/p>\n
                      CONTEXT = \"\"\r\nfor i, p in enumerate(most_similar_documents):\r\nwrapped_text = textwrap.fill(p, width=100)\r\nprint(\"-----------------------------------------------------------------\")\r\nprint(wrapped_text)\r\nprint(\"-----------------------------------------------------------------\")\r\nCONTEXT += wrapped_text + \"\\n\\n\"\r\n<\/code><\/pre>\n
                       <\/p>\n
                      7. \u751f\u6210\u4e00\u4e2a\u56de\u590d<\/h2>\n
                      \u73b0\u5728\u6211\u4eec\u6709\u4e86\u4e00\u4e2a\u67e5\u8be2\u548c\u76f8\u5173\u5185\u5bb9\u5757\uff0c\u5b83\u4eec\u5c06\u4e00\u8d77\u4f20\u9012\u7ed9\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u3002<\/p>\n
                      a. \u58f0\u660e\u67e5\u8be2<\/h3>\n
                      query = \"What was Studio Ghibli's first film?\"\r\n<\/code><\/pre>\n
                      b. \u521b\u5efa\u4e00\u4e2a\u63d0\u793a<\/h3>\n
                      prompt = f\"\"\"\r\nuse the following CONTEXT to answer the QUESTION at the end.\r\nIf you don't know the answer, just say that you don't know, don't try to make up an answer.\r\nCONTEXT: {CONTEXT}\r\nQUESTION: {query}\r\n\"\"\"\r\n<\/code><\/pre>\n
                      c. \u8bbe\u7f6e OpenAI<\/h3>\n
                        \n
                      • \u5b89\u88c5 OpenAI \u4ee5\u8bbf\u95ee\u548c\u4f7f\u7528\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u3002\n
                        !pip install -q openai\r\n<\/code><\/pre>\n<\/li>\n
                      • \u542f\u7528\u5bf9 OpenAI API \u5bc6\u94a5\u7684\u8bbf\u95ee\uff08\u53ef\u4ee5\u5728 Google Colab \u7684 secrets \u4e2d\u8bbe\u7f6e\uff09\u3002\n
                        from google.colab import userdata\r\nuserdata.get('openai')\r\nimport openai\r\n<\/code><\/pre>\n<\/li>\n
                      • \u521b\u5efa\u4e00\u4e2a OpenAI \u5ba2\u6237\u7aef\u3002\n
                        from openai import OpenAI\r\nclient = OpenAI(api_key=userdata.get('openai'))\r\n<\/code><\/pre>\n<\/li>\n<\/ul>\n
                        d. \u8c03\u7528 API \u751f\u6210\u56de\u590d<\/h3>\n
                        response = client.chat.completions.create(\r\nmodel=\"gpt-4o\",\r\nmessages=[\r\n{\"role\": \"user\", \"content\": prompt},\r\n]\r\n)\r\n<\/code><\/pre>\n
                          \n
                        • client.chat.completions.create:<\/strong>\u00a0\u6b64\u65b9\u6cd5\u8c03\u7528\u4e00\u4e2a\u57fa\u4e8e\u804a\u5929\u7684\u5927\u8bed\u8a00\u6a21\u578b\u521b\u5efa\u65b0\u7684\u56de\u590d\uff08\u751f\u6210\uff09\u3002<\/li>\n
                        • client:<\/strong>\u00a0\u8868\u793a\u8fde\u63a5\u5230\u670d\u52a1\uff08\u6b64\u5904\u4e3a OpenAI\uff09\u7684 API \u5ba2\u6237\u7aef\u5bf9\u8c61\u3002<\/li>\n
                        • chat.completions.create:<\/strong>\u00a0\u6307\u5b9a\u60a8\u6b63\u5728\u8bf7\u6c42\u521b\u5efa\u57fa\u4e8e\u804a\u5929\u7684\u751f\u6210\u3002<\/li>\n<\/ul>\n
                          \u5173\u4e8e\u4f20\u9012\u7ed9\u65b9\u6cd5\u7684\u53c2\u6570\u7684\u66f4\u591a\u4fe1\u606f<\/h4>\n
                            \n
                          • model=”gpt-4o”:<\/strong>\u00a0\u6307\u5b9a\u7528\u4e8e\u751f\u6210\u56de\u590d\u7684\u6a21\u578b\u3002”gpt-4o” \u662f GPT-4 \u6a21\u578b\u7684\u4e00\u4e2a\u7279\u5b9a\u53d8\u4f53\u3002\u4e0d\u540c\u7684\u6a21\u578b\u53ef\u80fd\u5177\u6709\u4e0d\u540c\u7684\u884c\u4e3a\u3001\u5fae\u8c03\u65b9\u5f0f\u6216\u80fd\u529b\uff0c\u56e0\u6b64\u6307\u5b9a\u6a21\u578b\u5bf9\u4e8e\u786e\u4fdd\u83b7\u5f97\u6240\u9700\u7684\u8f93\u51fa\u975e\u5e38\u91cd\u8981\u3002<\/li>\n
                          • messages:<\/strong>\u00a0\u6b64\u53c2\u6570\u662f\u4e00\u4e2a\u6d88\u606f\u5bf9\u8c61\u7684\u5217\u8868\uff0c\u7528\u4e8e\u8868\u793a\u5bf9\u8bdd\u5386\u53f2\u3002\u8fd9\u4f7f\u6a21\u578b\u80fd\u591f\u7406\u89e3\u804a\u5929\u7684\u4e0a\u4e0b\u6587\u3002\u5728\u672c\u4f8b\u4e2d\uff0c\u6211\u4eec\u5728\u5217\u8868\u4e2d\u4ec5\u63d0\u4f9b\u4e86\u4e00\u6761\u6d88\u606f\uff1a{\"role\": \"user\", \"content\": prompt}<\/code>\u3002<\/li>\n
                          • role:<\/strong>\u00a0“user” \u8868\u793a\u6d88\u606f\u53d1\u9001\u8005\u7684\u89d2\u8272\uff0c\u5373\u4e0e\u6a21\u578b\u4ea4\u4e92\u7684\u7528\u6237\u3002<\/li>\n
                          • content:<\/strong>\u00a0\u5305\u542b\u7528\u6237\u53d1\u9001\u7684\u6d88\u606f\u7684\u5b9e\u9645\u6587\u672c\u3002\u53d8\u91cf prompt \u4fdd\u5b58\u4e86\u6b64\u6587\u672c\uff0c\u6a21\u578b\u5c06\u4f7f\u7528\u8be5\u6587\u672c\u4f5c\u4e3a\u8f93\u5165\u6765\u751f\u6210\u56de\u590d\u3002<\/li>\n<\/ul>\n
                            e. \u5173\u4e8e\u63a5\u6536\u5230\u7684\u56de\u590d<\/h3>\n
                            \u5f53\u60a8\u5411\u7c7b\u4f3c OpenAI GPT \u6a21\u578b\u7684 API \u53d1\u51fa\u8bf7\u6c42\u4ee5\u751f\u6210\u804a\u5929\u56de\u590d\u65f6\uff0c\u54cd\u5e94\u901a\u5e38\u4ee5\u7ed3\u6784\u5316\u683c\u5f0f\u8fd4\u56de\uff0c\u901a\u5e38\u662f\u4e00\u4e2a\u5b57\u5178\u3002<\/p>\n
                            \u8fd9\u79cd\u7ed3\u6784\u901a\u5e38\u5305\u62ec\uff1a<\/p>\n
                              \n
                            • choices:<\/strong>\u00a0\u4e00\u4e2a\u5217\u8868\uff08\u6570\u7ec4\uff09\uff0c\u5176\u4e2d\u5305\u542b\u6a21\u578b\u751f\u6210\u7684\u591a\u4e2a\u53ef\u80fd\u7684\u56de\u590d\u3002\u6b64\u5217\u8868\u4e2d\u7684\u6bcf\u4e2a\u9879\u76ee\u4ee3\u8868\u4e00\u4e2a\u53ef\u80fd\u7684\u56de\u590d\u6216\u5b8c\u6210\u3002<\/li>\n
                            • message:<\/strong>\u00a0\u6bcf\u4e2a\u9009\u62e9\u4e2d\u7684\u4e00\u4e2a\u5bf9\u8c61\u6216\u5b57\u5178\uff0c\u5305\u542b\u6a21\u578b\u751f\u6210\u7684\u6d88\u606f\u7684\u5b9e\u9645\u5185\u5bb9\u3002<\/li>\n
                            • content:<\/strong>\u00a0\u6d88\u606f\u7684\u6587\u672c\u5185\u5bb9\uff0c\u5373\u6a21\u578b\u751f\u6210\u7684\u5b9e\u9645\u56de\u590d\u6216\u5b8c\u6210\u3002<\/li>\n<\/ul>\n
                              f. \u6253\u5370\u56de\u590d<\/h3>\n
                              print(response.choices[0].message.content)\r\n<\/code><\/pre>\n
                              \u6211\u4eec\u9009\u62e9\u00a0choices<\/code>\u00a0\u5217\u8868\u4e2d\u7684\u7b2c\u4e00\u4e2a\u9879\u76ee\uff0c\u7136\u540e\u8bbf\u95ee\u5176\u4e2d\u7684\u00a0message<\/code>\u00a0\u5bf9\u8c61\u3002\u6700\u540e\uff0c\u6211\u4eec\u8bbf\u95ee\u00a0message<\/code>\u00a0\u4e2d\u7684\u00a0content<\/code>\u00a0\u5b57\u6bb5\uff0c\u5b83\u5305\u542b\u6a21\u578b\u751f\u6210\u7684\u5b9e\u9645\u6587\u672c\u3002<\/p>\n
                               <\/p>\n
                              \u7ed3\u8bba<\/h2>\n
                              \u81f3\u6b64\uff0c\u6211\u4eec\u5b8c\u6210\u4e86\u4ece\u96f6\u5f00\u59cb\u6784\u5efa RAG \u7cfb\u7edf\u7684\u8bb2\u89e3\u3002\u5f3a\u70c8\u5efa\u8bae\u60a8\u9996\u5148\u4f7f\u7528\u7eaf Python \u6784\u5efa\u521d\u59cb RAG \u8bbe\u7f6e\uff0c\u4ee5\u66f4\u597d\u5730\u7406\u89e3\u8fd9\u4e9b\u7cfb\u7edf\u7684\u5de5\u4f5c\u539f\u7406\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
                              \u6982\u8ff0 \u672c\u6307\u5357\u5c06\u5f15\u5bfc\u60a8\u4f7f\u7528\u7eaf Python \u521b\u5efa\u4e00\u4e2a\u7b80\u5355\u7684\u68c0\u7d22\u589e\u5f3a\u751f\u6210 (RAG) \u7cfb\u7edf\u3002\u6211\u4eec\u5c06\u4f7f\u7528\u4e00\u4e2a\u5d4c\u5165\u6a21\u578b\u548c\u4e00\u4e2a\u5927\u8bed\u8a00\u6a21\u578b (LLM) \u6765\u68c0\u7d22\u76f8\u5173\u6587\u6863\u5e76\u57fa\u4e8e\u7528\u6237\u7684\u67e5\u8be2\u751f\u6210\u56de\u590d\u3002   https:\/\/github.com\/adi…<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[34],"tags":[],"class_list":["post-14903","post","type-post","status-publish","format-standard","hentry","category-knowledge"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts\/14903","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/comments?post=14903"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts\/14903\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/media?parent=14903"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/categories?post=14903"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/tags?post=14903"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}