【2024最全最细LangChain教程-13】Agent智能体（二）

【2024最全最细LangChain教程-12】Agent智能体（一）-CSDN博客

     B站视频：【2024最全最细】Langchain教学之Agent with memory_哔哩哔哩_bilibili

        上节课我们介绍了Agent，当时我们构造的是一个没有memory的Agent，下面我们看看如何构造一个有memory的Agent：

1. 构造一个google搜索工具

        google搜索工具需要自己去google的网站去申请账号，google每个月只有100次的免费调用额度，真抠：

from langchain_community.utilities import GoogleSearchAPIWrapper


search = GoogleSearchAPIWrapper()

# 注意tools是一个列表，只不过这里只有一个元素，回想一下上一节课的内容
tools = [
    Tool(
        name="Google Search",
        description="useful when you need to answer questions about current events",
        func=search.run,
    )
]
tools[0].run("what is the performance of NASDAQ today？")

2. 构造一个ZeroShotAgent with memory

        这里注意在构造prompt的时候，用了一个ZeroShotAgent.create_prompt（）方法，这种方法我们之前好像是没有见过的，input_variables注意除了input,chat_history还要有agent_scratchpad，这个是agent添加memory所需的：

import os

from langchain_core.prompts import PromptTemplate
from langchain_openai import OpenAI
from langchain.agents import AgentExecutor,Tool,ZeroShotAgent
from langchain.chains import LLMChain
from langchain.memory import ConversationBufferMemory


prefix = """
    Have a conversation with a human, answering the following questions as best you can. 
    If you can answer the question by yourself,dont' use the tools.
    You have access to the following tools: """


suffix = """Begin!"

Previous Conversation:{chat_history}
Question: {input}
Thought: {agent_scratchpad}"""

prompt = ZeroShotAgent.create_prompt(
    tools,
    prefix=prefix,
    suffix=suffix,
    input_variables=["input", "chat_history", "agent_scratchpad"],
)


memory = ConversationBufferMemory(memory_key="chat_history")
memory.clear()

memory.load_memory_variables({})

                然后我们和之前一样构造agent和 agent_chain，注意在agent_chain里添加memory：

llm = OpenAI(
    temperature=0,
    openai_api_key = os.getenv("OPENAI_API_KEY"),
    base_url = os.getenv("OPENAI_BASE_URL")
)

llm_chain = LLMChain(llm=llm, prompt=prompt)

agent = ZeroShotAgent(llm_chain=llm_chain, tools=tools, verbose=True)

agent_chain = AgentExecutor.from_agent_and_tools(
    agent=agent, tools=tools, verbose=True, memory=memory, max_iterations = 5
)

agent_chain.invoke({"input":"How many people live in China?"})

        我们来调用看下结果：

        继续提问，也可以得到答案：

        但是我发现了一个问题就是，如果你再问他一个问题，who are you ，Agent就困惑了，他不知道需不需要去使用工具，结果进入了一个死循环，这是Agent的一个问题。后面我们看可否通过提示词去优化这个问题：

3. 构造一个create_react_agent with memory

        原理相同，构造细节步骤有不同，不再赘述了，直接看代码即可：

import os

from langchain_openai import OpenAI
from langchain import hub
from langchain_core.prompts import PromptTemplate
from langchain.agents import AgentExecutor, create_react_agent

template='''
Answer the following questions as best you can. 
Try to answer the question by yourself.
If not necessary, don't use the tools.

You have access to the following tools:{tools}

Use the following format:

Question: the input question you must answer
Thought: you should always think about what to do
Action: the action to take, should be one of [{tool_names}]
Action Input: the input to the action
Observation: the result of the action
... (this Thought/Action/Action Input/Observation can repeat N times)
Thought: I now know the final answer
Final Answer: the final answer to the original input question

Begin!

Previous Conversation:{chat_history}
Question: {input}
Thought:{agent_scratchpad}
'''


prompt = PromptTemplate.from_template(template)
print(prompt)

llm = OpenAI(
    temperature=0,
    openai_api_key = os.getenv("OPENAI_API_KEY"),
    base_url = os.getenv("OPENAI_BASE_URL")
)

from langchain_community.utilities import GoogleSearchAPIWrapper


search = GoogleSearchAPIWrapper()


tools = [
    Tool(
        name="Google Search",
        description="A search engine you can use when you need to know latest news or data",
        func=search.run,
    ),
]

memory = ConversationBufferMemory(memory_key="chat_history")

agent = create_react_agent(llm, tools, prompt)
agent_executor = AgentExecutor(agent=agent, tools=tools, memory = memory, max_iterations = 3,handle_parsing_errors=True,verbose= True)

agent_executor.invoke({"input":"hi,how are you?"})

        这个agent在运行的时候，也会有不知道何时使用工具的问题。