83 lines
2.9 KiB
Python
83 lines
2.9 KiB
Python
|
from langchain_openai import ChatOpenAI
|
|||
|
|
from langchain_core.messages import HumanMessage, ToolMessage
|
|||
|
|
"""
|
|||
|
|
Pydantic 是一个 Python 库，用于数据验证和序列化。
|
|||
|
|
它通过使用 Python 类型注解（type hints）来定义数据模型，
|
|||
|
|
并提供强大的数据验证功能。Pydantic 基于 Python 的 dataclasses 和 typing 模块，
|
|||
|
|
允许开发者定义结构化的数据模型，并自动验证输入数据是否符合指定的类型和约束。
|
|||
|
|
"""
|
|||
|
|
from pydantic.v1 import BaseModel, Field
|
|||
|
|
|
|||
|
|
from conf import settings
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
# todo: 第一步：定义工具函数
|
|||
|
|
class Add(BaseModel):
|
|||
|
|
"""
|
|||
|
|
将两个数字相加
|
|||
|
|
"""
|
|||
|
|
a: int = Field(..., description="第一个数字")
|
|||
|
|
b: int = Field(..., description="第二个数字")
|
|||
|
|
|
|||
|
|
def invoke(self, args):
|
|||
|
|
# 验证参数
|
|||
|
|
tool_instance = self.__class__(**args) # 自动验证 a 和 b
|
|||
|
|
return tool_instance.a + tool_instance.b
|
|||
|
|
|
|||
|
|
class Multiply(BaseModel):
|
|||
|
|
"""
|
|||
|
|
将两个数字相乘
|
|||
|
|
"""
|
|||
|
|
a: int = Field(..., description="第一个数字")
|
|||
|
|
b: int = Field(..., description="第二个数字")
|
|||
|
|
|
|||
|
|
def invoke(self, args):
|
|||
|
|
# 验证参数
|
|||
|
|
tool_instance = self.__class__(**args) # 自动验证 a 和 b
|
|||
|
|
return tool_instance.a * tool_instance.b
|
|||
|
|
|
|||
|
|
# 定义 JSON 格式的工具 schema
|
|||
|
|
tools = [Add, Multiply]
|
|||
|
|
|
|||
|
|
|
|||
|
|
# todo: 第二步：初始化模型
|
|||
|
|
llm = ChatOpenAI(
|
|||
|
|
base_url=settings.base_url,
|
|||
|
|
api_key=settings.api_key,
|
|||
|
|
model=settings.model_name,
|
|||
|
|
temperature=0.1
|
|||
|
|
)
|
|||
|
|
# 绑定工具，允许模型自动选择工具
|
|||
|
|
llm_with_tools = llm.bind_tools(tools, tool_choice="auto")
|
|||
|
|
|
|||
|
|
# todo: 第三步：调用回复
|
|||
|
|
query = "2+1等于多少？"
|
|||
|
|
messages = [HumanMessage(query)]
|
|||
|
|
|
|||
|
|
try:
|
|||
|
|
# todo: 第一次调用
|
|||
|
|
ai_msg = llm_with_tools.invoke(messages)
|
|||
|
|
messages.append(ai_msg)
|
|||
|
|
print(f"\n第一轮调用后结果：\n{messages}")
|
|||
|
|
|
|||
|
|
# 处理工具调用
|
|||
|
|
# 判断消息中是否有tool_calls，以判断工具是否被调用
|
|||
|
|
if hasattr(ai_msg, 'tool_calls') and ai_msg.tool_calls:
|
|||
|
|
for tool_call in ai_msg.tool_calls:
|
|||
|
|
# todo: 处理工具调用
|
|||
|
|
selected_tool = {"add": Add, "multiply": Multiply}[tool_call["name"].lower()]
|
|||
|
|
# 实例化工具类并调用 invoke
|
|||
|
|
tool_instance = selected_tool(**tool_call["args"])
|
|||
|
|
tool_output = tool_instance.invoke(tool_call["args"])
|
|||
|
|
messages.append(ToolMessage(content=tool_output, tool_call_id=tool_call["id"]))
|
|||
|
|
print(f"\n第二轮 message中增加tool_output 之后：\n{messages}")
|
|||
|
|
|
|||
|
|
# todo: 第二次调用，将工具结果传回模型以生成最终回答
|
|||
|
|
final_response = llm_with_tools.invoke(messages)
|
|||
|
|
print(f"\n最终模型响应：\n{final_response.content}")
|
|||
|
|
else:
|
|||
|
|
print("模型未生成工具调用，直接返回文本:")
|
|||
|
|
print(ai_msg.content)
|
|||
|
|
except Exception as e:
|
|||
|
|
print(f"模型调用失败: {str(e)}")
|