大模型通过tool_calls调用外部方法

import json
import os
from openai import OpenAI
from dotenv import load_dotenv

# 加载 .env 文件
load_dotenv()
client = OpenAI(
                # 若没有配置环境变量，请用阿里云百炼API Key将下行替换为：api_key="sk-xxx",
                api_key=os.getenv("DASHSCOPE_API_KEY"),
                base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
            )
messages = [
        # {'role': 'system', 'content': ''},
        {'role': 'user', 'content': '我的手上有10个苹果  又买了3个  还剩多少个?'}
        ]

tools =  [
            {
                "type": "function",
                "function": {
                    "name": "sum",
                    "description": "使列表中n个数字相加",
                    "parameters": {
                        "type": "object",
                        "properties": {
                            "query": {
                                "description": "列表",
                                "type": "array",
                                "items": {
                                    "type": "number"
                                }
                            }
                        },
                        "required": ["query"]
                    }
                }
            }
        ]
def call_model_with_tool_response(messages):
    # 继续与模型对话
    response = client.chat.completions.create(
        model="qwen-plus",
        messages=messages,
        temperature=0, max_tokens=1024, top_p=1, frequency_penalty=0, presence_penalty=0,
        tools=tools
    )
    return response
message = call_model_with_tool_response(messages).choices[0].message
print(message)
#定义一个方法使调用工具回参

while message.tool_calls and message.tool_calls[0].function is not None:
    arguments = json.loads(message.tool_calls[0].function.arguments)["query"]
    name = message.tool_calls[0].function.name
    sumReturn  = sum(arguments)
    print(name)
    print(arguments)
    if name == "sum":
        sumReturn =  sum(arguments)
        messages.append(message)
        messages.append({
            'tool_call_id': message.tool_calls[0].id,
            'role': 'tool',
            'name': name,
            'content': str(sumReturn)
        })
    response = call_model_with_tool_response(messages)
    message = response.choices[0].message
    print(message)

调用tool_calls的步骤实际上是访问了两次大模型

1.大模型根据用户定义的tools识别出本次要用的tool 并在message内部返回本次调用的tool信息

2. 用户把大模型给的tool运行出来

3.用户把tool结果和tool_call_id   附加大模型第一次返回的message信息在第二次发送给大模型

        第三条需要特别注意不然不会被大模型识别为二次调用:

        抛出异常:openai.BadRequestError: Error code: 400 - {'error': {'code': 'invalid_parameter_error', 'param': None, 'message': '<400> InternalError.Algo.InvalidParameter: messages with role "tool" must be a response to a preceeding message with "tool_calls".', 'type': 'invalid_request_error'}, 'id': 'chatcmpl-c6979e5a-22b5-92ea-80ad-b18e820b279f', 'request_id': 'c6979e5a-22b5-92ea-80ad-b18e820b279f'}
4.如果需要更多的工具调用 例如  func1->func2->func3...->funN   大模型会逐步返回,真实与大模型发生N+1次交互