{"id":26110,"date":"2025-05-29T09:20:02","date_gmt":"2025-05-29T01:20:02","guid":{"rendered":"https:\/\/aif.amtbbs.org\/?p=26110"},"modified":"2025-05-29T09:20:02","modified_gmt":"2025-05-29T01:20:02","slug":"%e5%bf%ab%e9%80%9f%e7%90%86%e8%a7%a3%e7%83%ad%e9%97%a8-llm-%e5%a4%a7%e8%af%ad%e8%a8%80%e6%a8%a1%e5%9e%8b","status":"publish","type":"post","link":"https:\/\/aif.amtbbs.org\/index.php\/2025\/05\/29\/26110\/","title":{"rendered":"\u5feb\u901f\u7406\u89e3\u70ed\u95e8 LLM \u5927\u8bed\u8a00\u6a21\u578b"},"content":{"rendered":"
\n

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

<\/div>\n
\u672c\u6587\u5c3d\u91cf\u7528\u6700\u7b80\u5355\u7684\u65b9\u5f0f, \u5e2e\u8bfb\u8005\u7406\u89e3 LLM, Transformer, Prompt, Function calling, MCP, Agent, A2A \u7b49\u8fd9\u4e9b\u57fa\u672c\u6982\u5ff5\u3002<\/div>\n
\n
\n

 <\/p>\n

\u672c\u6587\u5c3d\u91cf\u7528\u6700\u7b80\u5355\u7684\u65b9\u5f0f, \u5e2e\u8bfb\u8005\u7406\u89e3 LLM\uff0cTransformer, Prompt, Function calling, MCP, Agent, A2A \u7b49\u8fd9\u4e9b\u57fa\u672c\u6982\u5ff5\u3002 \u8868\u8ff0\u65f6\u4e0d\u8ffd\u6c42\u7edd\u5bf9\u51c6\u786e\uff0c\u5c3d\u91cf\u901a\u4fd7\u6613\u61c2\uff0c\u90e8\u5206\u5185\u5bb9\u6709\u4e2a\u4eba\u7406\u89e3\u7684\u6210\u4efd\uff0c\u5185\u5bb9\u96be\u514d\u758f\u6f0f, \u6b22\u8fce\u6307\u6b63\u3002<\/p>\n

\u6ce8\u610f\uff1a\u672c\u6587\u9700\u8981\u4f60\u6709\u57fa\u672c\u7684\u4ee3\u7801\u9605\u8bfb\u80fd\u529b\uff0c\u5f53\u7136\u975e\u5f00\u53d1\u9605\u8bfb\u4e5f\u4e0d\u4f1a\u5f88\u56f0\u96be\u3002<\/p>\n

<\/p>\n

\u4e00\u3001LLM (\u5927\u8bed\u8a00\u6a21\u578b)<\/h3>\n

\u672c\u8d28\u5c31\u662f\u6587\u5b57\u63a5\u9f99<\/p>\n

<\/p>\n

\u628a\u95ee\u9898\u5f53\u6210\u8f93\u5165,\u00a0\u628a\u5927\u6a21\u578b\u5f53\u6210\u51fd\u6570,\u00a0\u628a\u56de\u7b54\u5f53\u6210\u8f93\u51fa\u3002<\/p>\n

\u5927\u6a21\u578b\u56de\u7b54\u95ee\u9898\u7684\u8fc7\u7a0b,\u00a0\u5c31\u662f\u4e00\u4e2a\u5faa\u73af\u6267\u884c\u51fd\u6570\u7684\u8fc7\u7a0b\u3002<\/p>\n

\u53e6\u5916\u6709\u5fc5\u8981\u4e86\u89e3\u4e00\u4e0b,\u00a0AI\u6280\u672f\u7206\u53d1\u4e8e2023\u5e74, ChatGPT\u7ecf\u8fc7\u4e86\u51e0\u6b21\u8fed\u4ee3\u624d\u5d2d\u9732\u5934\u89d2\u3002<\/p>\n

    \n
  • Transformer\u67b6\u6784<\/li>\n
  • \u53c2\u6570\u7206\u53d1\u589e\u957f<\/li>\n
  • \u4eba\u5de5\u5e72\u9884\u5956\u52b1\u6a21\u578b<\/li>\n<\/ul>\n

    \u601d\u8003\u9898: \u8bed\u8a00\u80fd\u4ee3\u8868\u667a\u80fd\u5417?<\/p>\n

    \u4e8c\u3001Transformer (\u81ea\u6ce8\u610f\u529b\u673a\u5236)<\/h3>\n

    \u81ea\u6ce8\u610f\u529b\u673a\u5236\u5c31\u662f\u52a8\u6001\u5173\u8054\u4e0a\u4e0b\u6587\u7684\u80fd\u529b. \u5982\u4f55\u5b9e\u73b0\u7684\u5462?<\/p>\n

    (1) \u6bcf\u4e2a\u5206\u8bcd\u5c31\u662f\u4e00\u4e2a token<\/p>\n

    (2) \u6bcf\u4e2atoken \u90fd\u6709\u4e00\u4e2a Q, K, V \u5411\u91cf (\u53c2\u6570)<\/p>\n

      \n
    • Q \u662f\u67e5\u8be2\u5411\u91cf<\/li>\n
    • K \u662f\u7ebf\u7d22\u5411\u91cf<\/li>\n
    • V \u662f\u7b54\u6848\u5411\u91cf<\/li>\n<\/ul>\n

      (3) \u63a8\u7406\u7684\u8fc7\u7a0b:<\/p>\n

        \n
      • \u5f53\u524dtoken \u7684Q \u4e0e \u524d\u9762\u6240\u6709\u7684 K \u8ba1\u7b97\u6743\u91cd<\/li>\n
      • \u6bcf\u4e2a token \u7684V\u52a0\u6743\u76f8\u52a0\u5f97\u5230\u4e00\u4e2a token\u9884\u6d4b\u503c<\/li>\n
      • \u9009\u62e9 N \u4e2a\u4e0e\u9884\u6d4b\u503c\u6700\u63a5\u8fd1\u7684 token, \u63b7\u9ab0\u5b50\u9009\u62e9<\/li>\n<\/ul>\n

        \u6700\u7b80\u5316\u793a\u4f8b\uff1a\u5c0f\u660e\u5403\u5b8c\u51b0\u6dc7\u6dcb\uff0c\u7ed3\u679c => \u809a\u5b50\u75bc<\/p>\n

        \u9996\u5148\u5206\u8bcd\u53ca\u6bcf\u4e2atoken\u7684 Q, K, V\u5411\u91cf\uff1a<\/p>\n\n\n\n\n\n\n<\/colgroup>\n\n\n\n\n\n\n
        token<\/td>\nQ\uff08\u67e5\u8be2\uff09<\/td>\nK\uff08\u952e\uff09<\/td>\nV\uff08\u503c\uff09<\/td>\n\u8bed\u4e49\u89e3\u91ca<\/td>\n<\/tr>\n
        \u5c0f\u660e<\/td>\n[0.2, 0.3]<\/td>\n[0.5, -0.1]<\/td>\n[0.1, 0.4]<\/td>\n\u4eba\u7269\u4e3b\u4f53<\/td>\n<\/tr>\n
        \u5403\u5b8c<\/td>\n[-0.4, 0.6]<\/td>\n[0.3, 0.8]<\/td>\n[-0.2, 0.5]<\/td>\n\u52a8\u4f5c\uff08\u5403\u5b8c\uff09<\/td>\n<\/tr>\n
        \u51b0\u6dc7\u6dcb<\/td>\n[0.7, -0.5]<\/td>\n[-0.6, 0.9]<\/td>\n[0.9, -0.3]<\/td>\n\u98df\u7269\uff08\u51b7\u996e\uff0c\u53ef\u80fd\u81f4\u8179\u6cfb\uff09<\/td>\n<\/tr>\n
        \u7ed3\u679c<\/td>\n[0.8, 0.2]<\/td>\n[0.2, -0.7]<\/td>\n[0.4, 0.1]<\/td>\n\u7ed3\u679c\uff08\u9700\u5173\u8054\u539f\u56e0\uff09<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

        \u63a5\u7740\u5f00\u59cb\u63a8\u7406\uff1a<\/p>\n

        1. \u4f7f\u7528\u6700\u540e\u4e00\u4e2a token \u7684 Q\uff08\u201c\u7ed3\u679c\u201d\u7684 Q \u5411\u91cf\uff09<\/h3>\n

        Q_current = [0.8, 0.2]<\/p>\n

        2. \u8ba1\u7b97 Q_current \u4e0e\u6240\u6709 K \u7684\u70b9\u79ef\uff08\u76f8\u4f3c\u5ea6\uff09<\/h4>\n

        \u70b9\u79ef\u516c\u5f0f\uff1aQ\u00b7K = q\u2081*k\u2081 + q\u2082*k\u2082<\/p>\n\n\n\n\n\n<\/colgroup>\n\n\n\n\n\n\n
        Token<\/td>\nK\u5411\u91cf<\/td>\n\u70b9\u79ef\u8ba1\u7b97<\/td>\n\u7ed3\u679c<\/td>\n<\/tr>\n
        \u5c0f\u660e<\/td>\n[0.5, -0.1]<\/td>\n0.8 * 0.5 + 0.2*(-0.1) = 0.4 – 0.02<\/td>\n0.38<\/td>\n<\/tr>\n
        \u5403\u5b8c<\/td>\n[0.3, 0.8]<\/td>\n0.8 * 0.3 + 0.2 * 0.8 = 0.24 + 0.16<\/td>\n0.40<\/td>\n<\/tr>\n
        \u51b0\u6dc7\u6dcb<\/td>\n[-0.6, 0.9]<\/td>\n0.8*(-0.6) + 0.2 * 0.9 = -0.48 + 0.18<\/td>\n-0.30<\/td>\n<\/tr>\n
        \u7ed3\u679c<\/td>\n[0.2, -0.7]<\/td>\n0.8 * 0.2 + 0.2*(-0.7) = 0.16 – 0.14<\/td>\n0.02<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

        3. Softmax \u5f52\u4e00\u5316\u5f97\u5230\u6ce8\u610f\u529b\u6743\u91cd<\/h4>\n

        \u5c06\u70b9\u79ef\u7ed3\u679c\u8f93\u5165 Softmax \u51fd\u6570<\/p>\n

        <\/p>\n\n\n\n\n\n<\/colgroup>\n\n\n\n\n\n\n
        Token<\/td>\n\u70b9\u79ef<\/td>\n\u6307\u6570\u503c\uff08e^x\uff09<\/td>\n\u6743\u91cd<\/td>\n<\/tr>\n
        \u5c0f\u660e<\/td>\n0.38<\/td>\ne^0.38 \u2248 1.46<\/td>\n1.46 \/ 2.74 \u2248\u00a00.53<\/td>\n<\/tr>\n
        \u5403\u5b8c<\/td>\n0.4<\/td>\ne^0.40 \u2248 1.49<\/td>\n1.49 \/ 2.74 \u2248\u00a00.54<\/td>\n<\/tr>\n
        \u51b0\u6dc7\u6dcb<\/td>\n-0.3<\/td>\ne^-0.30 \u2248 0.74<\/td>\n0.74 \/ 2.74 \u2248\u00a00.27<\/td>\n<\/tr>\n
        \u7ed3\u679c<\/td>\n0.02<\/td>\ne^0.02 \u2248 1.02<\/td>\n1.02 \/ 2.74 \u2248\u00a00.37<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

        4. \u52a0\u6743\u6c42\u548c V \u5411\u91cf\u751f\u6210\u4e0a\u4e0b\u6587\u5411\u91cf<\/h4>\n

        \u5c06\u6743\u91cd\u4e0e\u5bf9\u5e94 V \u5411\u91cf\u76f8\u4e58\u540e\u76f8\u52a0\uff1a<\/p>\n\n\n\n\n\n<\/colgroup>\n\n\n\n\n\n\n
        Token<\/td>\n\u6743\u91cd<\/td>\nV\u5411\u91cf<\/td>\n\u52a0\u6743 V \u5411\u91cf<\/td>\n<\/tr>\n
        \u5c0f\u660e<\/td>\n0.53<\/td>\n[0.1, 0.4]<\/td>\n0.53*[0.1, 0.4] \u2248 [0.053, 0.212]<\/td>\n<\/tr>\n
        \u5403\u5b8c<\/td>\n0.54<\/td>\n[-0.2, 0.5]<\/td>\n0.54*[-0.2, 0.5] \u2248 [-0.108, 0.27]<\/td>\n<\/tr>\n
        \u51b0\u6dc7\u6dcb<\/td>\n0.27<\/td>\n[0.9, -0.3]<\/td>\n0.27*[0.9, -0.3] \u2248 [0.243, -0.081]<\/td>\n<\/tr>\n
        \u7ed3\u679c<\/td>\n0.37<\/td>\n[0.4, 0.1]<\/td>\n0.37*[0.4, 0.1] \u2248 [0.148, 0.037]<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

        \u6700\u7ec8\u4e0a\u4e0b\u6587\u5411\u91cf\uff1a<\/p>\n

        [0.053\u22120.108+0.243+0.148,0.212+0.27\u22120.081+0.037]=[0.336,0.438]<\/p>\n

        5. \u9884\u6d4b\u4e0b\u4e00\u4e2a token<\/h4>\n

        \u6a21\u578b\u5c06\u4e0a\u4e0b\u6587\u5411\u91cf [0.336, 0.438] \u4e0e\u5019\u9009 token \u7684\u5d4c\u5165\u5411\u91cf\u5bf9\u6bd4\uff1a<\/p>\n

        \u5d4c\u5165\u5411\u91cf\u4e0d\u4f5c\u8fc7\u591a\u89e3\u91ca, \u53ea\u8981\u77e5\u9053QKV\u4e09\u4e2a\u5411\u91cf\u53ef\u4ece\u5d4c\u5165\u5411\u91cf\u8ba1\u7b97\u5f97\u5230\u5373\u53ef<\/p>\n\n\n\n\n\n<\/colgroup>\n\n\n\n\n\n
        \u5019\u9009\u8bcd<\/td>\n\u5d4c\u5165\u5411\u91cf<\/td>\n\u76f8\u4f3c\u5ea6\uff08\u70b9\u79ef\uff09<\/td>\n\u6982\u7387<\/td>\n<\/tr>\n
        \u809a\u5b50\u75bc<\/td>\n[0.3, 0.5]<\/td>\n0.336 * 0.3 + 0.438 * 0.5 \u2248 0.101 + 0.219 =\u00a00.320<\/td>\n\u6700\u5927\u6982\u7387\uff08\u4f8b\u5982 65%\uff09<\/td>\n<\/tr>\n
        \u5934\u75bc<\/td>\n[0.2, 0.1]<\/td>\n0.336 * 0.2 + 0.438 * 0.1 \u2248 0.067 + 0.044 =\u00a00.111<\/td>\n\u6b21\u4e4b\uff08\u4f8b\u5982 20%\uff09<\/td>\n<\/tr>\n
        \u5f00\u5fc3<\/td>\n[-0.5, 0.8]<\/td>\n0.336*(-0.5) + 0.438 * 0.8 \u2248 -0.168 + 0.350 =\u00a00.182<\/td>\n\u8f83\u4f4e\uff08\u4f8b\u5982 15%\uff09<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

        \u6700\u7ec8\u6a21\u578b\u9009\u62e9\u6700\u9ad8\u6982\u7387\u7684 \u201c\u809a\u5b50\u75bc\u201d \u4f5c\u4e3a\u4e0b\u4e00\u4e2a token\u3002<\/p>\n

        \u6ce8\u610f\u5728\u5b9e\u9645\u573a\u666f\u4e2d, \u9884\u6d4b\u7684\u4e0b\u4e00\u4e2atoken\u662f\u4e0d\u786e\u5b9a\u7684, \u662f\u56e0\u4e3a\u6709\u4e00\u4e2a\u63b7\u9ab0\u5b50\u7684\u64cd\u4f5c, \u5927\u6a21\u578b\u4f1a\u5728\u6982\u7387\u6700\u5927\u7684\u51e0\u4e2atoken\u4e2d\u968f\u673a\u6311\u9009\u4e00\u4e2a\u4f5c\u4e3a\u6700\u7ec8\u8f93\u51fa\u3002<\/p>\n

        \u4e09\u3001Prompt (\u63d0\u793a\u8bcd)<\/h3>\n

        \u5bf9\u4e8e\u8fd9\u4e2a\u8bcd\u5927\u5bb6\u5e76\u4e0d\u964c\u751f. \u6211\u4eec\u7528chatGPT\u65f6\u7ecf\u5e38\u4f1a\u7528\u5230, “\u4f60\u662f\u4e00\u4e2a…….”<\/p>\n

        \u4f46\u4f60\u771f\u7684\u7406\u89e3\u5b83\u5417?<\/p>\n

        \u4e0eai\u5bf9\u8bdd\u65f6\u7684\u8fd9\u79cd\u9884\u8bbe\u89d2\u8272, \u5176\u5b9e\u5e76\u4e0d\u662f\u4e25\u683c\u610f\u4e49\u4e0a\u7684 prompt<\/p>\n

        \u4e3a\u4ec0\u4e48\u8fd9\u4e48\u8bf4\u5462? \u5148\u770b\u4e00\u4e0bAPI<\/p>\n

        \u56db\u3001\u7406\u89e3API<\/h3>\n

         <\/p>\n

        \u6211\u4eec\u524d\u9762\u63d0\u5230\u8fc7\u5927\u8bed\u8a00\u6a21\u578b\u7684\u00a0\u672c\u8d28\u5c31\u662f\u6587\u5b57\u63a5\u9f99\u00a0, \u76f8\u5bf9\u5e94\u7684\u4f7f\u7528\u5927\u6a21\u578b\u4e5f\u6bd4\u8f83\u7b80\u5355. \u53ef\u4ee5\u53c2\u89c1deepseek\u7684\u6587\u5b57\u63a5\u9f99 api\u00a0\u8bf7\u6c42:<\/p>\n

        https:\/\/api-docs.deepseek.com\/zh-cn\/api\/create-chat-completion<\/p>\n

        <\/p>\n

        \u8fd9\u91cc\u6bd4\u8f83\u91cd\u8981\u7684\u51e0\u4e2a\u90e8\u5206, \u9700\u8981\u7406\u89e3\uff1a<\/p>\n

        1. temperature \u6e29\u5ea6<\/h4>\n

        Temperature\uff08\u6e29\u5ea6\uff09 \u662f\u4e00\u4e2a\u63a7\u5236\u751f\u6210\u6587\u672c\u968f\u673a\u6027\u548c\u591a\u6837\u6027\u7684\u5173\u952e\u53c2\u6570\u3002\u5b83\u901a\u8fc7\u8c03\u6574\u6a21\u578b\u8f93\u51fa\u7684\u6982\u7387\u5206\u5e03\uff0c\u76f4\u63a5\u5f71\u54cd\u751f\u6210\u5185\u5bb9\u7684\u201c\u4fdd\u5b88\u201d\u6216\u201c\u5192\u9669\u201d\u7a0b\u5ea6. \u770b\u51e0\u4e2a\u5178\u578b\u573a\u666f:<\/p>\n\n\n\n<\/colgroup>\n\n\n\n\n\n\n\n
        \u573a\u666f<\/td>\n\u6e29\u5ea6<\/td>\n<\/tr>\n
        \u4ee3\u7801\u751f\u6210\/\u6570\u5b66\u89e3\u9898<\/td>\n0<\/td>\n<\/tr>\n
        \u6570\u636e\u62bd\u53d6\/\u5206\u6790<\/td>\n1<\/td>\n<\/tr>\n
        \u901a\u7528\u5bf9\u8bdd<\/td>\n1.3<\/td>\n<\/tr>\n
        \u7ffb\u8bd1<\/td>\n1.3<\/td>\n<\/tr>\n
        \u521b\u610f\u7c7b\u5199\u4f5c\/\u8bd7\u6b4c\u521b\u4f5c<\/td>\n1.5<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

        2. tools \u5de5\u5177\u652f\u6301<\/h4>\n

        \u5927\u6a21\u578b\u5bf9 function calling \u7684\u652f\u6301, \u540e\u9762\u518d\u8be6\u7ec6\u4ecb\u7ecd\u3002<\/p>\n

        3. \u89d2\u8272\u548c\u4fe1\u606f<\/h4>\n

        \u8fd9\u4e00\u90e8\u5206\u662fai\u5bf9\u8bdd\u7684\u4e3b\u4f53.\u00a0\u5176\u4e2drole \u5b9a\u4e49\u4e86\u56db\u4e2a\u89d2\u8272\uff1a<\/p>\n

          \n
        • system \u7cfb\u7edf\u8bbe\u5b9a<\/li>\n
        • user \u7528\u6237\u56de\u590d<\/li>\n
        • assistant \u6a21\u578b\u56de\u7b54<\/li>\n
        • tool \u662f\u914d\u5408function call\u5de5\u4f5c\u7684\u89d2\u8272, \u53ef\u4ee5\u8c03\u7528\u5916\u90e8\u5de5\u5177<\/li>\n<\/ul>\n

          \u56de\u5230\u524d\u4e00\u7ae0\u7684\u95ee\u9898, ai\u5bf9\u8bdd\u65f6\u5176\u5b9e\u662fuser\u90e8\u5206\u8f93\u5165\u7684\u5185\u5bb9, \u6240\u4ee5system\u89d2\u8272\u7684\u8bbe\u5b9a\u5185\u5bb9\u624d\u5e94\u8be5\u662f\u4e25\u683c\u610f\u4e49\u4e0a\u7684Prompt\u3002<\/p>\n

          \u8fd9\u6709\u5565\u533a\u522b\u5462? (user \u4e0e system?)<\/p>\n

          \u4e2a\u4eba\u4e00\u4e2a\u5408\u7406\u7684\u731c\u6d4b: system\u7684\u5185\u5bb9\u5728Transformer\u63a8\u7406\u4e2d\u62e5\u6709\u8f83\u9ad8\u7684\u6743\u91cd\uff0c\u6240\u4ee5\u62e5\u6709\u8f83\u9ad8\u7684\u54cd\u5e94\u4f18\u5148\u7ea7\u3002<\/p>\n

          4. \u5173\u4e8e\u591a\u8f6e\u5bf9\u8bdd<\/h4>\n

          \u56e0\u4e3aLLM\u662f\u65e0\u72b6\u6001\u7684. \u6211\u4eec\u8981\u65f6\u523b\u8bb0\u5f97\u6587\u5b57\u63a5\u9f99\u7684\u6e38\u620f, \u56e0\u6b64\u5728\u5b9e\u9645\u64cd\u4f5c\u4e2d\u4e5f\u662f\u8fd9\u6837\u7684\uff1a<\/p>\n

          (1) \u5728\u7b2c\u4e00\u8f6e\u8bf7\u6c42\u65f6\uff0c\u4f20\u9012\u7ed9 API \u7684 messages \u4e3a<\/p>\n

          <\/p>\n

          (2) \u5927\u6a21\u578b\u56de\u7b54<\/p>\n

          <\/p>\n

          (3) \u5f53\u7528\u6237\u53d1\u8d77\u7b2c\u4e8c\u8f6e\u95ee\u9898\u65f6, \u8bf7\u6c42\u53d8\u6210\u4e86\u8fd9\u6837<\/p>\n

          <\/p>\n

          \u4e94\u3001Function Calling (\u51fd\u6570\u8c03\u7528)<\/h3>\n

          \u4ec5\u4ec5\u4e00\u4e2a\u53ef\u4ee5\u56de\u7b54\u95ee\u9898\u7684\u673a\u5668\u4eba, \u4f5c\u7528\u5e76\u4e0d\u592a\u5927\u3002<\/p>\n

          \u8981\u5b8c\u6210\u590d\u6742\u7684\u4efb\u52a1, \u5c31\u9700\u8981\u5927\u6a21\u578b\u7684\u8f93\u51fa\u662f\u7a33\u5b9a\u7684, \u800c\u4e14\u662f\u53ef\u7f16\u7a0b\u7684\u3002<\/p>\n

          \u56e0\u6b64OpenAI \u63a8\u51fa\u4e86 function calling\u7684\u652f\u6301. \u4e5f\u5c31\u662f\u524d\u9762\u63d0\u5230\u7684 tools\u53c2\u6570\u76f8\u5173\u5185\u5bb9\u3002<\/p>\n

          1. \u57fa\u672c\u6d41\u7a0b<\/h4>\n

          (1) \u5de5\u5177\u58f0\u660e\u53ca\u7528\u6237\u8f93\u5165<\/p>\n

          <\/p>\n

          (2) \u6a21\u578b\u68c0\u6d4b\u5230\u9700\u8981\u4f7f\u7528\u5de5\u5177, \u8fd4\u56de\u76f8\u5173\u5de5\u5177\u53c2\u6570<\/p>\n

          <\/p>\n

          (3) \u5f00\u53d1\u8005\u6839\u636e\u65b9\u6cd5\u540d\u548c\u53c2\u6570, \u8c03\u7528\u76f8\u5173\u5de5\u5177\u65b9\u6cd5<\/p>\n

          <\/p>\n

          (4) \u5c06\u5de5\u5177\u65b9\u6cd5\u7684\u8fd4\u56de\u503c, \u9644\u52a0\u5230\u8bf7\u6c42\u4e2d, \u518d\u6b21\u8bf7\u6c42\u5927\u6a21\u578b<\/p>\n

          <\/p>\n

          (5) \u5f97\u51fa\u6700\u7ec8\u7ed3\u679c<\/p>\n

          \n
          \n