{"id":1200,"date":"2024-04-25T21:32:42","date_gmt":"2024-04-25T21:32:42","guid":{"rendered":"https:\/\/www.nicekj.com\/?p=1200"},"modified":"2024-04-25T21:33:51","modified_gmt":"2024-04-25T21:33:51","slug":"openaihanshudiaoyong2ruhejiyuopenaifunctionsdazaozijillmyingyong","status":"publish","type":"post","link":"https:\/\/www.nicekj.com\/openaihanshudiaoyong2ruhejiyuopenaifunctionsdazaozijillmyingyong.html","title":{"rendered":"OpenAI \u51fd\u6570\u8c03\u7528\uff082\uff09 – \u5982\u4f55\u57fa\u4e8e OpenAI Functions \u6253\u9020\u81ea\u5df1 LLM \u5e94\u7528\uff1f"},"content":{"rendered":"

\u6982\u8ff0<\/h2>\n

\u4e0a\u6587 OpenAI \u51fd\u6570\u8c03\u7528\uff081\uff09 – ChatGPT Plugin \u662f\u5982\u4f55\u5b9e\u73b0\u7684\uff1f<\/a> \u6211\u4eec\u63d0\u5230\u4e86 OpenAI Functions Calling\uff08\u51fd\u6570\u8c03\u7528\uff09<\/code>\u529f\u80fd\u7684\u57fa\u672c\u6982\u5ff5\u4ee5\u53ca\u4f7f\u7528\u6d41\u7a0b\uff0c\u672c\u6587\u4f1a\u57fa\u4e8e\u4e0a\u6587\u5185\u5bb9\u8fdb\u884c\u5c55\u5f00\uff0c\u4e3b\u8981\u5305\u62ec 2 \u4e2a\u65b9\u5411\uff1a<\/p>\n

    \n
  1. \u8865\u5145 Functions Calling<\/code> \u529f\u80fd\u4f7f\u7528\u4e2d\u7684\u9ad8\u7ea7\u7684\u7528\u6cd5\uff0c\u5982\u591a\u8f6e\u5bf9\u8bdd\u53ca\u7cbe\u51c6\u63a7\u5236\u51fd\u6570\u8c03\u7528\u4e0e\u5426\uff1b<\/li>\n
  2. \u8865\u5145\u66f4\u591a\u7684 Functions Calling<\/code> \u7684\u4f7f\u7528\u573a\u666f\uff0c\u5982\u672c\u5730\u6570\u636e\u5e93\u7684 SQL \u67e5\u8be2\uff1b<\/li>\n<\/ol>\n

    \u597d\u7684\uff0c\u4e0b\u9762\u5c31\u7ee7\u7eed\u4f7f\u7528\u67e5\u8be2\u5929\u6c14\u7684\u573a\u666f\u6765\u8be6\u7ec6\u8bb2\u89e3\u4e0b Functions Calling<\/code> \u7684\u4e00\u4e9b\u9ad8\u7ea7\u529f\u80fd\u3002<\/p>\n

    Functions Calling \u9ad8\u7ea7\u7528\u6cd5<\/h2>\n

    \u5728\u5f00\u59cb\u9ad8\u7ea7\u7528\u6cd5\u4e4b\u524d\uff0c\u5148\u7b80\u5355\u56de\u987e\u4e0b Functions Calling<\/code> \u7684\u4f7f\u7528\u6d41\u7a0b\uff1a<\/p>\n

      \n
    1. \u58f0\u660e\u51fd\u6570\u5e76\u8bf7\u6c42 API<\/strong>\uff1a\u5b9a\u4e49\u5f53\u524d\u51fd\u6570\u7684\u540d\u79f0\uff0c\u63cf\u8ff0\uff0c\u4ee5\u53ca\u5bf9\u5e94\u7684\u53c2\u6570\u4fe1\u606f\uff0c\u5e76\u6267\u884c OpenAI \u7684 Chat \u63a5\u53e3\uff1b<\/li>\n
    2. \u6267\u884c\u672c\u5730\u51fd\u6570<\/strong>\uff1a\u63a5\u53d7\u63a5\u53e3\u8fd4\u56de\uff0c\u5e76\u89e3\u6790\u5bf9\u5e94\u7684\u51fd\u6570\u53c2\u6570\u4fe1\u606f\uff0c\u6839\u636e\u5bf9\u5e94\u7684\u53c2\u6570\u4fe1\u606f\u8c03\u7528\u672c\u5730\u51fd\u6570\uff1b<\/li>\n
    3. \u4e0a\u62a5\u6267\u884c\u7ed3\u679c<\/strong>\uff1a\u5c06\u672c\u5730\u51fd\u6570\u6267\u884c\u7684\u7ed3\u679c\u4e0a\u62a5\u7ed9 OpenAI Chat \u63a5\u53e3\u8fdb\u884c\u6c47\u603b\u3001\u603b\u7ed3\uff1b<\/li>\n<\/ol>\n

      \u6574\u4f53\u7684\u65f6\u5e8f\u56fe\u5927\u81f4\u5982\u4e0b\uff1a<\/p>\n

      \n

      \"OpenAI<\/figure>\n<\/p>\n

      \u4e0b\u9762\u4f1a\u4f7f\u7528\u67e5\u8be2\u5929\u6c14\u7684\u8fd9\u4e2a\u6848\u4f8b\u6765\u8bb2\u89e3\u4e0b\u4e00\u4e9b\u9ad8\u7ea7\u7528\u6cd5\uff0c\u9996\u5148\u6211\u4eec\u5148\u5c01\u88c5\u4e00\u4e9b\u57fa\u7840\u7684\u51fd\u6570\u3002<\/p>\n

      \u57fa\u7840\u51fd\u6570\u5c01\u88c5<\/h3>\n

      \u9996\u5148\u5c01\u88c5\u4e00\u4e2a chat \u51fd\u6570\uff0c\u53ef\u4ee5\u4f20\u5165\u5bf9\u8bdd\u5185\u5bb9\uff0cfunctions\uff0cfunction_call<\/code> \u4fe1\u606f\uff0c\u5b9a\u4e49\u5927\u81f4\u5982\u4e0b\uff1a<\/p>\n

      <\/div>
      python<\/span>
      \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>@retry(wait=wait_random_exponential(min<\/span>=1<\/span>, max<\/span>=40<\/span><\/span>), stop=stop_after_attempt(3<\/span><\/span>)<\/span>)<\/span><\/span>\ndef<\/span> chat_completion_request<\/span>(messages, functions=None<\/span>, function_call=None<\/span>, model=GPT_MODEL<\/span>):<\/span>\n    headers = {<\/span>\n        \"Content-Type\"<\/span>: \"application\/json\"<\/span>,<\/span>\n        \"Authorization\"<\/span>: \"Bearer \"<\/span> + openai.api_key,<\/span>\n    }<\/span>\n    json_data = {\"model\"<\/span>: model, \"messages\"<\/span>: messages}<\/span>\n    if<\/span> functions is<\/span> not<\/span> None<\/span>:<\/span>\n        json_data.update({\"functions\"<\/span>: functions})<\/span>\n    if<\/span> function_call is<\/span> not<\/span> None<\/span>:<\/span>\n        json_data.update({\"function_call\"<\/span>: function_call})<\/span>\n    try<\/span>:<\/span>\n        response = requests.post(<\/span>\n            \"https:\/\/api.openai.com\/v1\/chat\/completions\"<\/span>,<\/span>\n            headers=headers,<\/span>\n            json=json_data,<\/span>\n        )<\/span>\n        return<\/span> response<\/span>\n    except<\/span> Exception as<\/span> e:<\/span>\n        print<\/span>(\"Unable to generate ChatCompletion response\"<\/span>)<\/span>\n        print<\/span>(f\"Exception: {e}<\/span>\"<\/span>)<\/span>\n        return<\/span> e<\/span>\n<\/code><\/pre>\n
      \u4e3a\u4e86\u65b9\u4fbf\u67e5\u770b\u7ed3\u679c\uff0c\u6211\u4eec\u5728\u5c01\u88c5\u4e00\u4e2a\u65e5\u5fd7\u6253\u5370\u7684\u51fd\u6570\uff0c\u5982\u4e0b\uff1a<\/p>\n
      <\/div>
      python<\/span>
      \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>def<\/span> pretty_print_conversation<\/span>(messages<\/span>):<\/span>\n    role_to_color = {<\/span>\n        \"system\"<\/span>: \"red\"<\/span>,<\/span>\n        \"user\"<\/span>: \"green\"<\/span>,<\/span>\n        \"assistant\"<\/span>: \"blue\"<\/span>,<\/span>\n        \"function\"<\/span>: \"magenta\"<\/span>,<\/span>\n    }<\/span>\n    formatted_messages = []<\/span>\n    for<\/span> message in<\/span> messages:<\/span>\n        if<\/span> message[\"role\"<\/span>] == \"system\"<\/span>:<\/span>\n            formatted_messages.append(f\"system: {message['content'<\/span>]}<\/span>n\"<\/span>)<\/span>\n        elif<\/span> message[\"role\"<\/span>] == \"user\"<\/span>:<\/span>\n            formatted_messages.append(f\"user: {message['content'<\/span>]}<\/span>n\"<\/span>)<\/span>\n        elif<\/span> message[\"role\"<\/span>] == \"assistant\"<\/span> and<\/span> message.get(\"function_call\"<\/span>):<\/span>\n            formatted_messages.append(f\"assistant: {message['function_call'<\/span>]}<\/span>n\"<\/span>)<\/span>\n        elif<\/span> message[\"role\"<\/span>] == \"assistant\"<\/span> and<\/span> not<\/span> message.get(\"function_call\"<\/span>):<\/span>\n            formatted_messages.append(f\"assistant: {message['content'<\/span>]}<\/span>n\"<\/span>)<\/span>\n        elif<\/span> message[\"role\"<\/span>] == \"function\"<\/span>:<\/span>\n            formatted_messages.append(f\"function ({message['name'<\/span>]}<\/span>): {message['content'<\/span>]}<\/span>n\"<\/span>)<\/span>\n    for<\/span> formatted_message in<\/span> formatted_messages:<\/span>\n        print<\/span>(<\/span>\n            colored(<\/span>\n                formatted_message,<\/span>\n                role_to_color[messages[formatted_messages.index(formatted_message)][\"role\"<\/span>]],<\/span>\n            )<\/span>\n        )<\/span>\n<\/code><\/pre>\n
      \u5176\u6253\u5370\u6548\u679c\u5927\u81f4\u5982\u4e0b\uff0c\u6bcf\u4e00\u79cd\u89d2\u8272\u90fd\u4f7f\u7528\u4e0d\u540c\u7684\u989c\u8272\u6253\u5370\u51fa\u6765\uff08\u4e0d\u7528\u5173\u6ce8\u5177\u4f53\u7684\u6253\u5370\u5185\u5bb9\uff0c\u4e0b\u6587\u4f1a\u8be6\u7ec6\u8bb2\u89e3\uff09\uff1a<\/p>\n
      \n
      \"OpenAI<\/figure>\n<\/p>\n
      \u591a\u8f6e\u5bf9\u8bdd<\/h3>\n
      \u5f53\u6211\u4eec\u5b9a\u4e49\u7684\u51fd\u6570\u9700\u8981\u591a\u4e2a\u5fc5\u4f20<\/strong>\u5b57\u6bb5\uff0c\u5e76\u4e14\u7528\u6237\u5728\u4e00\u8f6e\u5bf9\u8bdd\u4e2d\u6ca1\u6709\u63d0\u4f9b\u8db3\u591f\u591a\u7684\u4fe1\u606f\u7684\u65f6\u5019\uff0cGPT \u4f1a\u8be2\u95ee\u7528\u6237\uff0c\u76f4\u5230\u7528\u6237\u63d0\u4f9b\u4e86\u5fc5\u8981\u7684\u4fe1\u606f\u540e\u624d\u4f1a\u89e6\u53d1\u51fd\u6570\u8c03\u7528\u3002<\/p>\n
      \u4e0b\u9762\u8fd8\u662f\u7528\u67e5\u8be2\u5929\u6c14\u4e3a\u4f8b\u6765\u8be6\u7ec6\u8bf4\u660e\u4e0b\uff0c\u9996\u5148\u9700\u8981\u5148\u5b9a\u4e49\u4e0b\u5bf9\u5e94\u67e5\u8be2\u5929\u6c14\u7684\u51fd\u6570\uff1a<\/p>\n
        \n
      1. get_current_weather<\/code>\uff1a\u67e5\u8be2\u5f53\u524d\u7684\u5929\u6c14\u60c5\u51b5\uff0c\u9700\u6c42\u4f20\u5165\u5bf9\u5e94\u7684\u4f4d\u7f6e\u4fe1\u606f\u4ee5\u53ca\u5929\u6c14\u7684\u683c\u5f0f\uff1b<\/li>\n
      2. get_n_day_weather_forecast<\/code>\uff1a\u83b7\u53d6\u6700\u8fd1 N \u5929\u7684\u5929\u6c14\u60c5\u51b5\uff0c\u9700\u6c42\u4f20\u5165\u5bf9\u5e94\u7684\u4f4d\u7f6e\u4fe1\u606f\u3001\u5929\u6c14\u7684\u683c\u5f0f\u4ee5\u53ca\u67e5\u8be2\u7684\u5929\u6570\uff1b<\/li>\n<\/ol>\n
        \u51fd\u6570\u58f0\u660e\u5982\u4e0b\uff1a<\/p>\n
        <\/div>
        python<\/span>
        \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>functions = [<\/span>\n    {<\/span>\n        \"name\"<\/span>: \"get_current_weather\"<\/span>,<\/span>\n        \"description\"<\/span>: \"Get the current weather\"<\/span>,<\/span>\n        \"parameters\"<\/span>: {<\/span>\n            \"type\"<\/span>: \"object\"<\/span>,<\/span>\n            \"properties\"<\/span>: {<\/span>\n                \"location\"<\/span>: {<\/span>\n                    \"type\"<\/span>: \"string\"<\/span>,<\/span>\n                    \"description\"<\/span>: \"The city and state, e.g. San Francisco, CA\"<\/span>,<\/span>\n                },<\/span>\n                \"format\"<\/span>: {<\/span>\n                    \"type\"<\/span>: \"string\"<\/span>,<\/span>\n                    \"enum\"<\/span>: [\"celsius\"<\/span>, \"fahrenheit\"<\/span>],<\/span>\n                    \"description\"<\/span>: \"The temperature unit to use. Infer this from the users location.\"<\/span>,<\/span>\n                },<\/span>\n            },<\/span>\n            \"required\"<\/span>: [\"location\"<\/span>, \"format\"<\/span>],<\/span>\n        },<\/span>\n    },<\/span>\n    {<\/span>\n        \"name\"<\/span>: \"get_n_day_weather_forecast\"<\/span>,<\/span>\n        \"description\"<\/span>: \"Get an N-day weather forecast\"<\/span>,<\/span>\n        \"parameters\"<\/span>: {<\/span>\n            \"type\"<\/span>: \"object\"<\/span>,<\/span>\n            \"properties\"<\/span>: {<\/span>\n                \"location\"<\/span>: {<\/span>\n                    \"type\"<\/span>: \"string\"<\/span>,<\/span>\n                    \"description\"<\/span>: \"The city and state, e.g. San Francisco, CA\"<\/span>,<\/span>\n                },<\/span>\n                \"format\"<\/span>: {<\/span>\n                    \"type\"<\/span>: \"string\"<\/span>,<\/span>\n                    \"enum\"<\/span>: [\"celsius\"<\/span>, \"fahrenheit\"<\/span>],<\/span>\n                    \"description\"<\/span>: \"The temperature unit to use. Infer this from the users location.\"<\/span>,<\/span>\n                },<\/span>\n                \"num_days\"<\/span>: {<\/span>\n                    \"type\"<\/span>: \"integer\"<\/span>,<\/span>\n                    \"description\"<\/span>: \"The number of days to forecast\"<\/span>,<\/span>\n                }<\/span>\n            },<\/span>\n            \"required\"<\/span>: [\"location\"<\/span>, \"format\"<\/span>, \"num_days\"<\/span>]<\/span>\n        },<\/span>\n    },<\/span>\n]<\/span>\n<\/code><\/pre>\n
        \u4f46\u7528\u6237\u8be2\u95ee\u5f53\u524d\u5929\u6c14\uff0c\u5e76\u4e14\u6ca1\u6709\u544a\u77e5\u8be2\u95ee\u54ea\u91cc\u7684\u5929\u6c14\u65f6\uff0cGPT \u56de\u8be2\u95ee\u9700\u8981\u67e5\u8be2\u54ea\u91cc\u7684\u5929\u6c14\uff0c\u4ee3\u7801\u5982\u4e0b\uff1a<\/p>\n
        <\/div>
        python<\/span>
        \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>messages = []<\/span>\nmessages.append({\"role\"<\/span>: \"system\"<\/span>, \"content\"<\/span>: \"Don't make assumptions about what values to plug into functions. Ask for clarification if a user request is ambiguous.\"<\/span>})<\/span>\nmessages.append({\"role\"<\/span>: \"user\"<\/span>, \"content\"<\/span>: \"What's the weather like today\"<\/span>})<\/span>\nchat_response = chat_completion_request(<\/span>\n    messages, functions=functions<\/span>\n)<\/span>\nassistant_message = chat_response.json()[\"choices\"<\/span>][0<\/span>][\"message\"<\/span>]<\/span>\nmessages.append(assistant_message)<\/span>\npretty_print_conversation(messages)<\/span>\n<\/code><\/pre>\n
        \u5bf9\u8bdd\u5185\u5bb9\u5927\u81f4\u5982\u4e0b\uff1a<\/p>\n
        \n
        \"OpenAI<\/figure>\n<\/p>\n
        \u8fd9\u5c31\u662f\u6240\u8c13\u7684\u591a\u8f6e\u5bf9\u8bdd\uff0c\u5f53\u9700\u8981\u4f20\u5165\u7684\u53c2\u6570\u6ca1\u6709\u63d0\u4f9b\u65f6\uff0c\u4f1a\u8be2\u95ee\u7528\u6237\u5bf9\u5e94\u7684\u4fe1\u606f\uff0c\u76f4\u5230\u7528\u6237\u56de\u7b54\u4e86\u6240\u6709\u7684\u4fe1\u606f\u3002\u8fd9\u662f Chat \u63a5\u53e3\u5c31\u4f1a\u8fd4\u56de\u5bf9\u5e94\u7684 function_call<\/code> \u5b57\u6bb5\uff0c\u8fd4\u56de\u5185\u5bb9\u5927\u81f4\u5982\u4e0b\uff1a<\/p>\n
        <\/div>
        json<\/span>
        \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>{<\/span><\/span>\n    \"role\"<\/span>:<\/span> \"assistant\"<\/span>,<\/span><\/span>\n    \"content\"<\/span>:<\/span> null<\/span><\/span>,<\/span><\/span>\n    \"function_call\"<\/span>:<\/span> {<\/span><\/span>\n        \"name\"<\/span>:<\/span> \"get_current_weather\"<\/span>,<\/span><\/span>\n        \"arguments\"<\/span>:<\/span> \"{n  \"location\": \"Glasgow, Scotland\",n  \"format\": \"celsius\"n}\"<\/span><\/span>\n    }<\/span><\/span>\n}<\/span><\/span>\n<\/code><\/pre>\n
        \u5b8c\u6574\u7684\u5bf9\u8bdd\u8fc7\u7a0b\u5927\u81f4\u5982\u4e0b\uff1a<\/p>\n
        \n
        \"OpenAI<\/figure>\n<\/p>\n
        \u6211\u4eec\u5b9a\u4e49\u7684 get_n_day_weather_forecast<\/code> \u51fd\u6570\u4e5f\u662f\u901a\u7528\u7684\u903b\u8f91\uff0c\u5927\u5bb6\u53ef\u4ee5\u81ea\u884c\u5c1d\u8bd5\u3002<\/p>\n
        \u7cbe\u51c6\u63a7\u5236\u51fd\u6570\u8c03\u7528<\/h3>\n
        \u4e0a\u8ff0\u7684\u5bf9\u8bdd\u8fc7\u7a0b\u4e2d\uff0c\u6211\u4eec\u8be2\u95ee\u4e86\u548c\u5929\u6c14\u76f8\u5173\u5185\u5bb9\u5e76\u4e14\u8865\u5168\u4e86\u5bf9\u5e94\u7684\u4fe1\u606f\uff0cGPT \u6700\u7ec8\u4f1a\u8c03\u7528\u5bf9\u5e94\u7684\u51fd\u6570\u3002\u6b63\u5e38\u60c5\u51b5\u4e0b\uff0c\u6211\u4eec\u8be2\u95ee\u65e0\u5173\u7684\u95ee\u9898\u65f6\u5e76\u4e0d\u4f1a\u51fa\u53d1\u5bf9\u5e94\u7684\u51fd\u6570\u5b9a\u4e49\uff0c\u5982\u4e0b\u5bf9\u8bdd\u5185\u5bb9\uff1a<\/p>\n
        \n
        \"OpenAI<\/figure>\n<\/p>\n
        \u6211\u4eec\u53ef\u4ee5\u5728\u8bf7\u6c42\u63a5\u53e3\u7684\u65f6\u5019\u4f7f\u7528 function_call<\/code> \u5b57\u6bb5\u5f3a\u5236 GPT \u8c03\u7528\u5bf9\u5e94\u7684\u51fd\u6570\uff0c\u4ee3\u7801\u5982\u4e0b\uff1a<\/p>\n
        <\/div>
        json<\/span>
        \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>messages = [<\/span>]<\/span><\/span>\nmessages.append({<\/span>\"role\"<\/span>:<\/span> \"system\"<\/span>,<\/span> \"content\"<\/span>:<\/span> \"Don't make assumptions about what values to plug into functions. Ask for clarification if a user request is ambiguous.\"<\/span>}<\/span>)<\/span>\nmessages.append({<\/span>\"role\"<\/span>:<\/span> \"user\"<\/span>,<\/span> \"content\"<\/span>:<\/span> \"hello\"<\/span>}<\/span>)<\/span>\nchat_response = chat_completion_request(<\/span>\n    messages,<\/span> functions=functions,<\/span> function_call={<\/span>\"name\"<\/span>:<\/span> \"get_n_day_weather_forecast\"<\/span>}<\/span><\/span>\n)<\/span>\nassistant_message = chat_response.json()[<\/span>\"choices\"<\/span>]<\/span>[<\/span>0<\/span>]<\/span>[<\/span>\"message\"<\/span>]<\/span><\/span>\nmessages.append(assistant_message)<\/span>\npretty_print_conversation(messages)<\/span>\n<\/code><\/pre>\n
        \u81f3\u4e8e\u7f3a\u5931\u7684\u53c2\u6570 GPT \u4f1a\u81ea\u52a8\u751f\u6210\uff0c\u8fd4\u56de\u7684\u7ed3\u679c\u5927\u81f4\u5982\u4e0b\uff1a<\/p>\n
        <\/div>
        json<\/span>
        \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>{<\/span><\/span>\n    \/\/...<\/span><\/span>\n    \"choices\"<\/span>:<\/span> [<\/span><\/span>\n      {<\/span><\/span>\n        \"index\"<\/span>:<\/span> 0<\/span>,<\/span><\/span>\n        \"message\"<\/span>:<\/span> {<\/span><\/span>\n          \"role\"<\/span>:<\/span> \"assistant\"<\/span>,<\/span><\/span>\n          \"content\"<\/span>:<\/span> null<\/span><\/span>,<\/span><\/span>\n          \"function_call\"<\/span>:<\/span> {<\/span><\/span>\n            \"name\"<\/span>:<\/span> \"get_n_day_weather_forecast\"<\/span>,<\/span><\/span>\n            \"arguments\"<\/span>:<\/span> \"{n  \"location\": \"San Francisco, CA\",n  \"format\": \"celsius\",n  \"num_days\": 5n}\"<\/span><\/span>\n          }<\/span><\/span>\n        }<\/span>,<\/span><\/span>\n        \"finish_reason\"<\/span>:<\/span> \"stop\"<\/span><\/span>\n      }<\/span><\/span>\n    ]<\/span><\/span>\n  }<\/span><\/span>\n<\/code><\/pre>\n
        \u5b8c\u6574\u7684\u5bf9\u8bdd\u6d41\u7a0b\u5982\u4e0b\uff1a<\/p>\n
        \n
        \"OpenAI<\/figure>\n<\/p>\n
        \u53ef\u77e5\uff0cGPT \u81ea\u52a8\u751f\u6210\u9700\u8981\u7684\u5730\u5740\uff08San Francisco, CA<\/code>\uff09\u3001\u5929\u6c14\u683c\u5f0f\uff08celsius<\/code>\uff09\u4ee5\u53ca\u9700\u8981\u67e5\u8be2\u7684\u5929\u6570\uff083<\/code>\uff09\u3002<\/p>\n
        \u5173\u4e8e\u59cb\u7ec8\u88ab\u8c03\u7528\u7684\u8fd9\u4e2a\u7279\u6027\u5728\u4e00\u4e9b LLM \u5e94\u7528\u4e2d\u5c06\u4f1a\u53d8\u5f97\u975e\u5e38\u5b9e\u7528\uff0c\u6211\u4eec\u53ef\u4ee5\u59cb\u7ec8\u8981\u6c42\u8fd4\u56de\u5bf9\u5e94\u7684 JSON \u7ed3\u6784\uff0c\u800c\u4e0d\u518d\u662f\u5904\u7406\u4e00\u4e9b\u7eaf\u6587\u672c\u5bf9\u8bdd\u4e86\u3002\u6bd4\u5982\uff0c\u8fd4\u56de command \u548c payload \u5f62\u5f0f\u5185\u5bb9\uff0c\u6211\u4eec\u5c31\u53ef\u4ee5\u5f88\u597d\u7684\u53bb\u5904\u7406\u5bf9\u5e94\u7684\u903b\u8f91\u4e86\u3002\u5f53\u7136\uff0c\u8fd9\u5c5e\u4e8e\u51fd\u6570\u8c03\u7528\u7684 hack \u7528\u6cd5\u4e86\uff0c\u8fd9\u79cd\u573a\u666f\u8fd4\u56de\u7684\u5185\u5bb9\u4e0d\u662f\u53c2\u6570\uff0c\u800c\u662f\u5bf9\u5e94\u7684 Response \u4e86\u3002<\/p>\n
        \u7cbe\u51c6\u63a7\u5236\u51fd\u6570\u4e0d\u88ab\u8c03\u7528<\/h3>\n
        \u53ef\u4ee5\u901a\u77e5 GPT \u59cb\u7ec8\u8c03\u7528\u51fd\u6570\uff0c\u540c\u6837\u7684\u4e5f\u53ef\u4ee5\u8bbe\u7f6e\u59cb\u7ec8\u4e0d\u8c03\u7528\u51fd\u6570\uff0c\u5c06\u5bf9\u5e94\u7684 function_call<\/code> \u8bbe\u7f6e\u4e3a none<\/code> \u5373\u53ef\u3002\u5982\u4e0b\uff1a<\/p>\n
        <\/div>
        python<\/span>
        \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>messages = []<\/span>\nmessages.append({\"role\"<\/span>: \"system\"<\/span>, \"content\"<\/span>: \"Don't make assumptions about what values to plug into functions. Ask for clarification if a user request is ambiguous.\"<\/span>})<\/span>\nmessages.append({\"role\"<\/span>: \"user\"<\/span>, \"content\"<\/span>: \"Give me the current weather (use Celcius) for Toronto, Canada.\"<\/span>})<\/span>\nchat_response = chat_completion_request(<\/span>\n    messages, functions=functions, function_call=\"none\"<\/span><\/span>\n)<\/span>\nassistant_message = chat_response.json()[\"choices\"<\/span>][0<\/span>][\"message\"<\/span>]<\/span>\nmessages.append(assistant_message)<\/span>\npretty_print_conversation(messages)<\/span>\n<\/code><\/pre>\n
        assistant_message<\/code> \u5bf9\u5e94\u7684 JOSN \u5185\u5bb9\u5982\u4e0b\uff1a<\/p>\n
        <\/div>
        python<\/span>
        \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>{<\/span>\n\"role\"<\/span>: \"assistant\"<\/span>,<\/span>\n\"content\"<\/span>: \"{n \"<\/span>location\": \"<\/span>Toronto, Canada\",n \"<\/span>format<\/span>\": \"<\/span>celsius\"n}\"<\/span><\/span>\n}<\/span>\n<\/code><\/pre>\n
        \u6ce8\u610f\uff0c\u4e0a\u8ff0\u5e76\u6ca1\u6709\u8c03\u7528\u51fd\u6570\uff0c\u53ea\u662f content \u7684\u5185\u5bb9\u8fd4\u56de\u7684 JSON \u683c\u5f0f\u3002\u5bf9\u8bdd\u6d41\u7a0b\u5927\u81f4\u5982\u4e0b\uff1a<\/p>\n
        \n
        \"OpenAI<\/figure>\n<\/p>\n
        \u5f53\u7136\uff0c\u4e0d\u4f20\u5165 functions<\/code> \u53c2\u6570\u4e5f\u53ef\u4ee5\u8fbe\u5230\u540c\u6837\u7684\u6548\u679c\u3002<\/p>\n
        \u4e0a\u8ff0\u7684\u5185\u5bb9\u90fd\u5728\u56f4\u7ed5\u7740 OpenAI \u7684 Chat \u63a5\u53e3\u5982\u4f55\u8c03\u7528\uff0c\u5173\u4e8e get_current_weather<\/code> \u4e0e get_n_day_weather_forecast<\/code> \u51fd\u6570\u7684\u5177\u4f53\u5b9a\u4e49\u4ee5\u53ca\u6267\u884c\u7684\u5e76\u6ca1\u6709\u5c55\u5f00\u9610\u8ff0\uff0c\u8fd9\u90e8\u5206\u5176\u5b9e\u5c31\u662f\u672c\u5730\u7684\u51fd\u6570\u5b9a\u4e49\u90e8\u5206\uff0c\u4e0e\u5177\u4f53\u7684\u4e1a\u52a1\u903b\u8f91\u6709\u5173\uff0c\u4e0e OpenAI \u63a5\u53e3\u5173\u7cfb\u4e0d\u5927\u3002<\/p>\n
        \u7531\u4e8e\u67e5\u8be2\u5929\u6c14\u672c\u8d28\u4e0a\u5c31\u662f\u7f51\u7edc\u63a5\u53e3\u8bf7\u6c42\uff0c\u5927\u5bb6\u5e94\u8be5\u90fd\u6bd4\u8f83\u719f\u6089\uff0c\u8fd9\u90e8\u4e0d\u518d\u63cf\u8ff0\u8be6\u7ec6\u7684\u5b9e\u73b0\u7ec6\u8282\uff0c\u4e0b\u9762\u4f1a\u4ee5\u67e5\u8be2\u672c\u5730\u6570\u636e\u5e93\u7684\u573a\u666f\uff0c\u6765\u8be6\u7ec6\u9610\u8ff0\u4e0b OpenAI Factions Calling \u4e8e\u672c\u5730\u51fd\u6570\u6267\u884c\u7684\u5b8c\u6574\u903b\u8f91\u3002<\/p>\n
        \u6848\u4f8b\u8bb2\u89e3\uff1a SQL \u67e5\u8be2<\/h2>\n
        \u6570\u636e\u5e93\u5b9a\u4e49<\/h3>\n
        \u5047\u8bbe\u73b0\u5728\u6709\u4e00\u4e2a\u540d\u4e3a Chinook.db<\/code> \u7684\u6570\u636e\u5e93\u6587\u4ef6\uff0c\u5176\u4e2d\u5305\u542b\u591a\u5f20\u8868\uff0c\u6bcf\u5f20\u8868\u4e2d\u53c8\u5305\u542b\u591a\u4e2a\u5b57\u6bb5\u4fe1\u606f\u3002\u5927\u81f4\u5b9a\u4e49\u5982\u4e0b\uff1a<\/p>\n
        <\/div>
        yaml<\/span>
        \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>Table:<\/span> Album<\/span><\/span>\n       Columns:<\/span> AlbumId,<\/span> Title,<\/span> ArtistId<\/span><\/span>\nTable:<\/span> Artist<\/span><\/span>\n       Columns:<\/span> ArtistId,<\/span> Name<\/span><\/span>\nTable:<\/span> Customer<\/span><\/span>\n       Columns:<\/span> CustomerId,<\/span> FirstName,<\/span> LastName,<\/span> Company,<\/span> Address,<\/span> City,<\/span> State,<\/span> Country,<\/span> PostalCode,<\/span> Phone,<\/span> Fax,<\/span> Email,<\/span> SupportRepId<\/span><\/span>\nTable:<\/span> Employee<\/span><\/span>\n       Columns:<\/span> EmployeeId,<\/span> LastName,<\/span> FirstName,<\/span> Title,<\/span> ReportsTo,<\/span> BirthDate,<\/span> HireDate,<\/span> Address,<\/span> City,<\/span> State,<\/span> Country,<\/span> PostalCode,<\/span> Phone,<\/span> Fax,<\/span> Email<\/span><\/span>\nTable:<\/span> Genre<\/span><\/span>\n       Columns:<\/span> GenreId,<\/span> Name<\/span><\/span>\nTable:<\/span> Invoice<\/span><\/span>\n       Columns:<\/span> InvoiceId,<\/span> CustomerId,<\/span> InvoiceDate,<\/span> BillingAddress,<\/span> BillingCity,<\/span> BillingState,<\/span> BillingCountry,<\/span> BillingPostalCode,<\/span> Total<\/span><\/span>\nTable:<\/span> InvoiceLine<\/span><\/span>\n       Columns:<\/span> InvoiceLineId,<\/span> InvoiceId,<\/span> TrackId,<\/span> UnitPrice,<\/span> Quantity<\/span><\/span>\nTable:<\/span> MediaType<\/span><\/span>\n       Columns:<\/span> MediaTypeId,<\/span> Name<\/span><\/span>\nTable:<\/span> Playlist<\/span><\/span>\n       Columns:<\/span> PlaylistId,<\/span> Name<\/span><\/span>\nTable:<\/span> PlaylistTrack<\/span><\/span>\n       Columns:<\/span> PlaylistId,<\/span> TrackId<\/span><\/span>\nTable:<\/span> Track<\/span><\/span>\n       Columns:<\/span> TrackId,<\/span> Name,<\/span> AlbumId,<\/span> MediaTypeId,<\/span> GenreId,<\/span> Composer,<\/span> Milliseconds,<\/span> Bytes,<\/span> UnitPrice<\/span><\/span>\n<\/code><\/pre>\n
        \u60f3\u8981 GPT \u5e2e\u6211\u4eec\u751f\u6210\u4e00\u4e9b\u67e5\u8be2\u6570\u636e\u5e93\u7684\u53c2\u6570\uff0c\u5fc5\u987b\u8981\u544a\u8bc9 GPT \u6211\u4eec\u6570\u636e\u5e93\u7684\u4e00\u4e9b\u7ed3\u6784\u4fe1\u606f\uff0c\u5426\u5219\u5b83\u5e76\u4e0d\u4f1a\u751f\u6210\u51c6\u786e\u7684\u4fe1\u606f\u3002\u6240\u4ee5\u4e0b\u9762\u5c31\u662f\u5b9a\u4e49\u4e00\u4e9b\u57fa\u7840\u51fd\u6570\u6765\u751f\u6210\u4e00\u4e9b\u6570\u636e\u5e93\u7684\u63cf\u8ff0\u4fe1\u606f\uff0c\u5e76\u5c06\u5176\u4f20\u9012\u7ed9 GPT \u3002<\/p>\n
        \u6570\u636e\u4fe1\u606f\u751f\u6210<\/h3>\n
        \u9996\u5148\u6211\u4eec\u9700\u8981\u5b9a\u4e49\u4e00\u4e9b\u51fd\u6570\u6765\u751f\u6210\u7c7b\u4f3c\u4e0a\u8ff0\u7684\u6570\u636e\u5e93\u63cf\u8ff0\u4fe1\u606f\uff08\u8868\u7ed3\u6784\uff09\uff0c\u9996\u5148\u5b9a\u4e49\u83b7\u53d6\u6570\u636e\u5e93\u4fe1\u606f\u7684\u51fd\u6570\uff0c\u5927\u81f4\u4ee3\u7801\u5982\u4e0b\uff1a<\/p>\n
        <\/div>
        python<\/span>
        \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>def<\/span> get_table_names<\/span>(conn<\/span>):<\/span>\n    \"\"\"Return a list of table names.\"\"\"<\/span><\/span>\n    table_names = []<\/span>\n    tables = conn.execute(\"SELECT name FROM sqlite_master WHERE type='table';\"<\/span>)<\/span>\n    for<\/span> table in<\/span> tables.fetchall():<\/span>\n        table_names.append(table[0<\/span>])<\/span>\n    return<\/span> table_names<\/span>\n<\/span>\ndef<\/span> get_column_names<\/span>(conn, table_name<\/span>):<\/span>\n    \"\"\"Return a list of column names.\"\"\"<\/span><\/span>\n    column_names = []<\/span>\n    columns = conn.execute(f\"PRAGMA table_info('{table_name}<\/span>');\"<\/span>).fetchall()<\/span>\n    for<\/span> col in<\/span> columns:<\/span>\n        column_names.append(col[1<\/span>])<\/span>\n    return<\/span> column_names<\/span>\n<\/span>\ndef<\/span> get_database_info<\/span>(conn<\/span>):<\/span>\n    \"\"\"Return a list of dicts containing the table name and columns for each table in the database.\"\"\"<\/span><\/span>\n    table_dicts = []<\/span>\n    for<\/span> table_name in<\/span> get_table_names(conn):<\/span>\n        columns_names = get_column_names(conn, table_name)<\/span>\n        table_dicts.append({\"table_name\"<\/span>: table_name, \"column_names\"<\/span>: columns_names})<\/span>\n    return<\/span> table_dicts<\/span>\n<\/code><\/pre>\n
        \u7136\u540e\u6211\u4eec\u5c06\u5bf9\u5e94\u7684\u6570\u636e\u63cf\u8ff0\u4fe1\u606f\u6574\u7406\u6210\u4e00\u6bb5\u53ef\u7406\u89e3\u7684\u63cf\u8ff0\u4fe1\u606f\uff0c\u4ee3\u7801\u5927\u81f4\u5982\u4e0b\uff1a<\/p>\n
        <\/div>
        python<\/span>
        \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>import<\/span> sqlite3<\/span>\n<\/span>\nconn = sqlite3.connect(\"data\/Chinook.db\"<\/span>)<\/span>\nprint<\/span>(\"Opened database successfully\"<\/span>)<\/span>\n<\/span>\ndatabase_schema_dict = get_database_info(conn)<\/span>\ndatabase_schema_string = \"n\"<\/span>.join(<\/span>\n    [<\/span>\n        f\"Table: {table['table_name'<\/span>]}<\/span>nColumns: {', '<\/span>.join(table['column_names'<\/span>])}<\/span>\"<\/span><\/span>\n        for<\/span> table in<\/span> database_schema_dict<\/span>\n    ]<\/span>\n)<\/span>\n<\/code><\/pre>\n
        \u6253\u5370\u51fa database_schema_string<\/code> \u4fe1\u606f\u5982\u4e0b\uff1a<\/p>\n
        <\/div>
        makefile<\/span>
        \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>Table: Album<\/span><\/span>\nColumns: AlbumId, Title, ArtistId<\/span><\/span>\nTable: Artist<\/span><\/span>\nColumns: ArtistId, Name<\/span><\/span>\nTable: Customer<\/span><\/span>\nColumns: CustomerId, FirstName, LastName, Company, Address, City, State, Country, PostalCode, Phone, Fax, Email, SupportRepId<\/span><\/span>\nTable: Employee<\/span><\/span>\nColumns: EmployeeId, LastName, FirstName, Title, ReportsTo, BirthDate, HireDate, Address, City, State, Country, PostalCode, Phone, Fax, Email<\/span><\/span>\nTable: Genre<\/span><\/span>\nColumns: GenreId, Name<\/span><\/span>\nTable: Invoice<\/span><\/span>\nColumns: InvoiceId, CustomerId, InvoiceDate, BillingAddress, BillingCity, BillingState, BillingCountry, BillingPostalCode, Total<\/span><\/span>\nTable: InvoiceLine<\/span><\/span>\nColumns: InvoiceLineId, InvoiceId, TrackId, UnitPrice, Quantity<\/span><\/span>\nTable: MediaType<\/span><\/span>\nColumns: MediaTypeId, Name<\/span><\/span>\nTable: Playlist<\/span><\/span>\nColumns: PlaylistId, Name<\/span><\/span>\nTable: PlaylistTrack<\/span><\/span>\nColumns: PlaylistId, TrackId<\/span><\/span>\nTable: Track<\/span><\/span>\nColumns: TrackId, Name, AlbumId, MediaTypeId, GenreId, Composer, Milliseconds, Bytes, UnitPrice<\/span><\/span>\n<\/code><\/pre>\n
        \u6709\u4e86\u5bf9\u5e94\u7684\u4fe1\u606f\u4e4b\u540e\uff0c\u6211\u4eec\u5c31\u53ef\u4ee5\u58f0\u660e\u5bf9\u5e94\u7684 functions \u4e86\u3002<\/p>\n
        Function Calling \u58f0\u660e\u4e0e\u5b9a\u4e49<\/h3>\n
        \u60f3\u8981 GPT \u5e2e\u6211\u4eec\u751f\u6210\u4e00\u4e9b\u67e5\u8be2\u6570\u636e\u5e93\u7684\u53c2\u6570\uff0c\u5fc5\u987b\u8981\u544a\u8bc9 GPT \u6211\u4eec\u6570\u636e\u5e93\u7684\u4e00\u4e9b\u7ed3\u6784\u4fe1\u606f\uff0c\u5bf9\u5e94\u7684\u5b9a\u4e49\u5982\u4e0b\uff1a<\/p>\n
          \n
        1. name<\/code>\uff1a\u51fd\u6570\u540d\u4e3a ask_database\uff0c\u53ef\u4ee5\u67e5\u8be2\u6570\u636e\u5e93\u7684\u4e00\u4e9b\u4fe1\u606f\uff1b<\/li>\n
        2. description<\/code>\uff1a\u8be6\u7ec6\u63cf\u8ff0\u4e0b\u51fd\u6570\u7684\u4f5c\u7528\u4ee5\u53ca\u671f\u671b\u7684\u683c\u5f0f\uff1b<\/li>\n
        3. parameters<\/code>\uff1a\u51fd\u6570\u63a5\u53d7\u4e00\u4e2a\u540d\u4e3a query<\/code> \u7684\u53c2\u6570\uff0c\u5c06\u5bf9\u5e94\u7684\u81ea\u7136\u8bed\u8a00\u8f6c\u6362\u6210\u5bf9\u5e94\u7684 SQL \u8bed\u53e5\uff0c\u800c\u5bf9 query<\/code> \u7684\u63cf\u8ff0\u5c31\u9700\u8981\u6dfb\u52a0\u4e0a\u5bf9\u5e94\u7684\u6570\u636e\u5e93\u4fe1\u606f\u4e86\uff1b<\/li>\n<\/ol>\n
          \u5bf9\u5e94\u7684\u58f0\u660e\u5927\u81f4\u5982\u4e0b\uff1a<\/p>\n
          <\/div>
          python<\/span>
          \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>functions = [<\/span>\n    {<\/span>\n        \"name\"<\/span>: \"ask_database\"<\/span>,<\/span>\n        \"description\"<\/span>: \"Use this function to answer user questions about music. Output should be a fully formed SQL query.\"<\/span>,<\/span>\n        \"parameters\"<\/span>: {<\/span>\n            \"type\"<\/span>: \"object\"<\/span>,<\/span>\n            \"properties\"<\/span>: {<\/span>\n                \"query\"<\/span>: {<\/span>\n                    \"type\"<\/span>: \"string\"<\/span>,<\/span>\n                    \"description\"<\/span>: f\"\"\"<\/span><\/span>\n                            SQL query extracting info to answer the user's question.<\/span>\n                            SQL should be written using this database schema:<\/span>\n                            {database_schema_string}<\/span><\/span>\n                            The query should be returned in plain text, not in JSON.<\/span>\n                            \"\"\",<\/span>\n                }<\/span>\n            },<\/span>\n            \"required\"<\/span>: [\"query\"<\/span>],<\/span>\n        },<\/span>\n    }<\/span>\n]<\/span>\n<\/code><\/pre>\n
          \u4e0b\u9762\u5c31\u662f\u9700\u8981\u5b9a\u4e49\u4e00\u4e2a ask_database<\/code> \u8fd9\u4e2a\u51fd\u6570\u7684\u771f\u5b9e\u903b\u8f91\u4e86\u3002<\/p>\n
          \u672c\u5730\u51fd\u6570\u5b9a\u4e49<\/h3>\n
          \u4e0b\u9762\u5c31\u662f\u4e00\u4e2a\u67e5\u8be2\u672c\u5730\u6570\u636e\u5e93\u7684\u5177\u4f53\u5b9e\u73b0\uff0c\u8be5\u51fd\u6570\u4f1a\u63a5\u53d7\u5bf9\u5e94\u7684 SQL \u8bed\u53e5\uff0c\u5e76\u8fd4\u56de\u5bf9\u5e94\u7684\u67e5\u8be2\u7ed3\u679c\uff0c\u4ee3\u7801\u5982\u4e0b\uff1a<\/p>\n
          <\/div>
          python<\/span>
          \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>def<\/span> ask_database<\/span>(conn, query<\/span>):<\/span>\n    \"\"\"Function to query SQLite database with a provided SQL query.\"\"\"<\/span><\/span>\n    try<\/span>:<\/span>\n        results = str<\/span>(conn.execute(query).fetchall())<\/span>\n    except<\/span> Exception as<\/span> e:<\/span>\n        results = f\"query failed with error: {e}<\/span>\"<\/span><\/span>\n    return<\/span> results<\/span>\n<\/span>\ndef<\/span> execute_function_call<\/span>(message<\/span>):<\/span>\n    if<\/span> message[\"function_call\"<\/span>][\"name\"<\/span>] == \"ask_database\"<\/span>:<\/span>\n        query = json.loads(message[\"function_call\"<\/span>][\"arguments\"<\/span>])[\"query\"<\/span>]<\/span>\n        results = ask_database(conn, query)<\/span>\n    else<\/span>:<\/span>\n        results = f\"Error: function {message['function_call'<\/span>]['name'<\/span>]}<\/span> does not exist\"<\/span><\/span>\n    return<\/span> results<\/span>\n<\/code><\/pre>\n
          Chat \u5b8c\u6574\u6d41\u7a0b<\/h3>\n
          \u6240\u6709\u7684\u51c6\u5907\u5de5\u4f5c\u90fd\u5df2\u7ecf\u5b8c\u6210\u4e86\uff0c\u6211\u4eec\u770b\u4e00\u4e0b\u5b8c\u6210\u6d41\u7a0b\u7684\u6548\u679c\uff0c\u7528\u6237\u6b63\u5e38\u5bf9\u8bdd\u7684\u8fc7\u7a0b\u4e2d\u4f1a\u6709\u4ec0\u4e48\u6548\u679c\u3002\u6bd4\u5982\u6211\u4eec\u60f3\u8981\u67e5\u8be2\u6b4c\u66f2\u91cf\u524d 5 \u7684\u6b4c\u624b\u90fd\u6709\u54ea\u4e9b\uff0c\u4ee3\u7801\u5982\u4e0b\uff1a<\/p>\n
          <\/div>
          python<\/span>
          \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>messages = []<\/span>\nmessages.append({\"role\"<\/span>: \"system\"<\/span>, \"content\"<\/span>: \"Answer user questions by generating SQL queries against the Chinook Music Database.\"<\/span>})<\/span>\nmessages.append({\"role\"<\/span>: \"user\"<\/span>, \"content\"<\/span>: \"Hi, who are the top 5 artists by number of tracks?\"<\/span>})<\/span>\nchat_response = chat_completion_request(messages, functions)<\/span>\nassistant_message = chat_response.json()[\"choices\"<\/span>][0<\/span>][\"message\"<\/span>]<\/span>\n# \u89e3\u6790\u5bf9\u5e94\u7684\u7ed3\u679c\uff0c\u5e76\u6dfb\u52a0\u5230\u5bf9\u8bdd\u5386\u53f2\u4e2d<\/span><\/span>\nmessages.append(assistant_message)<\/span>\nif<\/span> assistant_message.get(\"function_call\"<\/span>):<\/span>\n\t\t# \u6267\u884c\u672c\u5730\u7684\u6570\u636e\u5e93\u67e5\u8be2\u903b\u8f91<\/span><\/span>\n    results = execute_function_call(assistant_message)<\/span>\n\t\t# \u5c06\u6267\u884c\u7ed3\u679c\u6dfb\u52a0\u5bf9\u8bdd\u5386\u53f2\u4e2d<\/span><\/span>\n    messages.append({\"role\"<\/span>: \"function\"<\/span>, \"name\"<\/span>: assistant_message[\"function_call\"<\/span>][\"name\"<\/span>], \"content\"<\/span>: results})<\/span>\n<\/span>\npretty_print_conversation(messages)<\/span>\n<\/code><\/pre>\n
          \u8fd9\u4e2a\u8fc7\u7a0b\u4e2d\u4ea7\u751f\u7684\u5bf9\u8bdd\u5386\u53f2\u5982\u4e0b\uff1a<\/p>\n
          \n
          \"OpenAI<\/figure>\n<\/p>\n
          \u5230\u8fd9\u91cc\u6574\u4e2a\u73af\u8282\u8fd8\u6ca1\u6709\u5b8c\u6210\uff0c\u6211\u8fd8\u9700\u5c06\u4e0a\u8ff0\u7684\u5bf9\u8bdd\u5386\u53f2\u8bf7\u6c42 OpenAI \u7684 Chat \u63a5\u53e3\uff0cGPT \u4f1a\u6839\u636e\u51fd\u6570\u7684\u7ed3\u679c\u6574\u7406\u6210\u5bb9\u6613\u7406\u89e3\u7684\u81ea\u7136\u8bed\u8a00\u3002\u4ee3\u7801\u5982\u4e0b\uff1a<\/p>\n
          <\/div>
          python<\/span>
          \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div># \u4f7f\u7528\u4e0a\u8ff0\u7684\u5bf9\u8bdd\u5386\u53f2\u518d\u6b21\u8bf7\u6c42 Chat \u63a5\u53e3<\/span><\/span>\nchat_response = chat_completion_request(messages, functions)<\/span>\n# \u5c06 GPT \u8fd4\u56de\u7684\u90a3\u5929\u6dfb\u52a0\u5230\u5bf9\u8bdd\u5386\u53f2\u4e2d<\/span><\/span>\nassistant_message = chat_response.json()[\"choices\"<\/span>][0<\/span>][\"message\"<\/span>]<\/span>\nmessages.append(assistant_message)<\/span>\npretty_print_conversation(messages)<\/span>\n<\/code><\/pre>\n
          \u6574\u4e2a\u8fc7\u7a0b\u5b8c\u6574\u7684\u5bf9\u8bdd\u5386\u53f2\u5982\u4e0b\uff1a<\/p>\n
          \n
          \"OpenAI<\/figure>\n<\/p>\n
          \u53ef\u89c1\uff0cGPT \u7684\u786e\u662f\u5c06\u6570\u636e\u5e93\u7684\u67e5\u8be2\u7ed3\u679c\u8f6c\u6362\u6210\u5bb9\u6613\u7406\u89e3\u7684\u81ea\u7136\u8bed\u8a00\u4e86\u3002<\/p>\n
          \u603b\u7ed3<\/h2>\n
          \u672c\u6587\u4ecb\u7ecd\u4e86 Function Calling<\/code> \u7684\u4e00\u4e9b\u9ad8\u7ea7\u7528\u6cd5\uff0c\u5982\uff1a<\/p>\n
            \n
          1. GPT \u7684\u591a\u8f6e\u5bf9\u8bdd\u673a\u5236\uff0c\u6765\u8be2\u95ee\u7528\u6237\u63d0\u4f9b\u5fc5\u8981\u7684\u5173\u952e\u4fe1\u606f\uff08\u901a\u8fc7 Prompt \u63a7\u5236\uff09\uff1b<\/li>\n
          2. \u5f3a\u5236\u8981\u6c42 GPT \u4f7f\u7528\u6307\u5b9a\u51fd\u6570\uff0c\u5e76\u751f\u6210\u4e00\u4e9b\u53c2\u6570\u4fe1\u606f\uff1b<\/li>\n
          3. \u4e5f\u53ef\u4ee5\u8981\u6c42 GPT \u4e0d\u7528\u4f7f\u7528\u51fd\u6570\uff1b<\/li>\n<\/ol>\n
            \u4e3a\u4e86\u6f14\u793a\u5b8c\u6574\u771f\u5b9e\u7684\u4f7f\u7528\u573a\u666f\uff0c\u6211\u4eec\u4ee5\u67e5\u8be2\u672c\u5730\u6570\u636e\u5e93\u7684\u5177\u4f53\u6848\u4f8b\u8bb2\u89e3\u4e86 Function Calling<\/code> \u58f0\u660e\u7684\u7ec6\u8282\uff08\u4e00\u5b9a\u8981\u5c06\u5bf9\u5e94\u7684\u4fe1\u606f\u5b8c\u6574\u7684\u7ed9\u5230GPT\uff0c\u5426\u5219\u5b83\u5e76\u4e0d\u80fd\u5b8c\u5168\u7406\u89e3\u201c\u53c2\u6570\u201d\u4f20\u9012\u7684\u5408\u7406\u6027\uff09\uff0c\u4ee5\u53ca\u51fd\u6570\u6267\u884c\u7684\u5b8c\u6574\u6d41\u7a0b\u3002<\/p>\n
            \u540e\u9762\u4f1a\u8bb2\u89e3\u6253\u9020\u81ea\u5df1\u7684 Plugin \u5546\u5e97\u7684\u4e00\u4e9b\u601d\u8def\uff0c\u53ef\u4ee5\u65e0\u7f1d\u7684\u5c06\u73b0\u6709\u7684 ChatGPT Plugin \u63a5\u5165\u5230\u81ea\u5df1\u7684\u4e1a\u52a1\u4e2d\u3002<\/p>","protected":false},"excerpt":{"rendered":"
            \u8be6\u7ec6\u8bb2\u89e3 OpenAI Functions Calling \u529f\u80fd\u4f7f\u7528\u4e2d\u7684\u9ad8\u7ea7\u7684\u7528\u6cd5\uff0c\u5982\u591a\u8f6e\u5bf9\u8bdd\u53ca\u7cbe\u51c6\u63a7\u5236\u51fd\u6570\u8c03\u7528\u4e0e\u5426\uff0c\u8865\u5145 GPT \u8c03\u7528\u672c\u5730\u6570\u636e\u5e93\u67e5\u8be2\u7684\u5177\u4f53\u6848\u4f8b\u3002<\/p>\n","protected":false},"author":1,"featured_media":2357,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"rank_math_title":"ChatGPT Plugin\u8fdb\u9636\uff1a\u51fd\u6570\u8c03\u7528\u9ad8\u7ea7\u7528\u6cd5\u8be6\u89e3 - \u5948\u65af\u79d1\u6280\u793e\u533a","rank_math_description":"\u6df1\u5165\u89e3\u6790OpenAI\u51fd\u6570\u8c03\u7528\u548cChatGPT Plugin\u7684\u9ad8\u7ea7\u7528\u6cd5\uff0c\u5305\u62ec\u591a\u8f6e\u5bf9\u8bdd\u3001\u7cbe\u51c6\u63a7\u5236\u51fd\u6570\u8c03\u7528\u3001\u672c\u5730\u6570\u636e\u5e93SQL\u67e5\u8be2\u7b49\u573a\u666f\uff0c\u52a9\u60a8\u6253\u9020\u66f4\u5f3a\u5927\u7684AI\u5e94\u7528\u3002","rank_math_focus_keyword":"OpenAI, ChatGPT, \u51fd\u6570\u8c03\u7528, Plugin, SQL\u67e5\u8be2","views":"33","footnotes":""},"categories":[3],"tags":[129,136,126,127,128],"collection":[],"class_list":["post-1200","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-fenlei2","tag-129","tag-136","tag-gpt","tag-ai","tag-128"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1200","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/comments?post=1200"}],"version-history":[{"count":0,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1200\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media\/2357"}],"wp:attachment":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media?parent=1200"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/categories?post=1200"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/tags?post=1200"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/collection?post=1200"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}