ReAct

class ReActAgent:
    def __init__(self, llm, tools):
        self.llm = llm
        self.tools = tools
        self.history = []
    
    def run(self, question):
        """执行ReAct循环"""
        max_iterations = 5
        
        for i in range(max_iterations):
            # 1. 生成思考
            thought = self.generate_thought(question, self.history)
            self.history.append(f"Thought {i+1}: {thought}")
            
            # 2. 决定行动
            action = self.decide_action(thought)
            if action == "FINISH":
                return self.generate_final_answer()
            
            self.history.append(f"Action {i+1}: {action}")
            
            # 3. 执行行动
            observation = self.execute_action(action)
            self.history.append(f"Observation {i+1}: {observation}")
            
            # 4. 检查是否完成
            if self.is_complete(observation):
                break
        
        return self.generate_final_answer()
    
    def generate_thought(self, question, history):
        prompt = f"""
        问题：{question}
        历史：{history}
        
        基于以上信息，思考下一步应该做什么。
        Thought:
        """
        return self.llm.generate(prompt)

from langchain.agents import initialize_agent, Tool
from langchain.agents import AgentType

# 定义工具
tools = [
    Tool(
        name="Wikipedia",
        func=wikipedia_search,
        description="搜索维基百科获取信息"
    ),
    Tool(
        name="Calculator",
        func=calculator,
        description="执行数学计算"
    ),
    Tool(
        name="Weather",
        func=get_weather,
        description="获取天气信息"
    )
]

# 创建ReAct智能体
agent = initialize_agent(
    tools,
    llm,
    agent=AgentType.REACT_DOCSTORE,
    verbose=True
)

# 执行任务
result = agent.run("2024年诺贝尔物理学奖得主是谁？他们的主要贡献是什么？")

# 现代LLM的原生函数调用
response = openai.ChatCompletion.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "获取苹果股价"}],
    functions=[
        {
            "name": "get_stock_price",
            "description": "获取股票实时价格",
            "parameters": {
                "type": "object",
                "properties": {
                    "symbol": {"type": "string"}
                },
                "required": ["symbol"]
            }
        }
    ],
    function_call="auto"
)

# LLM自动决定调用函数
if response.choices[0].message.get("function_call"):
    function_name = response.choices[0].message["function_call"]["name"]
    arguments = json.loads(response.choices[0].message["function_call"]["arguments"])
    result = execute_function(function_name, arguments)