Python langchain ReAct 使用范例詳解

更新時間：2024年12月21日 16:05:52 作者：愛學習的小道長

這篇文章主要介紹了Python langchain ReAct 使用范例,本文給大家介紹的非常詳細,感興趣的朋友一起看看吧

0. 介紹

ReAct: Reasoning + Acting ，ReAct Prompt 由 few-shot task-solving trajectories 組成，包括人工編寫的文本推理過程和動作，以及對動作的環(huán)境觀察。

1. 范例

langchain version 0.3.7

$ pip show langchain
Name: langchain
Version: 0.3.7
Summary: Building applications with LLMs through composability
Home-page: https://github.com/langchain-ai/langchain
Author: 
Author-email: 
License: MIT
Location: /home/xjg/.conda/envs/langchain/lib/python3.10/site-packages
Requires: aiohttp, async-timeout, langchain-core, langchain-text-splitters, langsmith, numpy, pydantic, PyYAML, requests, SQLAlchemy, tenacity
Required-by: langchain-community

1.1 使用第三方工具

Google 搜索對接
第三方平臺：https://serpapi.com
LangChain API 封裝：SerpAPI

1.1.1 簡單使用工具

from langchain_community.utilities import SerpAPIWrapper
import os
# 刪除all_proxy環(huán)境變量
if 'all_proxy' in os.environ:
    del os.environ['all_proxy']
# 刪除ALL_PROXY環(huán)境變量
if 'ALL_PROXY' in os.environ:
    del os.environ['ALL_PROXY']
os.environ["SERPAPI_API_KEY"] = "xxx"
params = {
    "engine": "bing",
    "gl": "us",
    "hl": "en",
}
search = SerpAPIWrapper(params=params)
result = search.run("Obama's first name?")
print(result)

輸出結果：

['In 1975, when Obama started high school in Hawaii, teacher Eric Kusunoki read the roll call and stumbled on Obama\'s first name. "Is Barack here?" he asked, pronouncing it BAR-rack .', "Barack Obama, the 44th president of the United States, was born on August 4, 1961, in Honolulu, Hawaii to Barack Obama, Sr. (1936–1982) (born in Oriang' Kogelo of Rachuonyo North District, Kenya) and Stanley Ann Dunham, known as Ann (1942–1995) (born in Wichita, Kansas, United States). Obama spent most of his childhood years in Honolulu, where his mother attended the University of Hawai?i at Mānoa", 'Barack Obama is named after his father, who was a Kenyan economist (called under the same name). He’s first real given name is “Barak”, also spelled Baraq (Not to be confused with Barack which is is a building or group of buildings …', 'Nevertheless, he was proud enough of his formal name that after he and Ann Dunham married in 1961, they named their son, Barack Hussein Obama II. As a youngster, the former president likely never...', 'https://www.britannica.com/biography/Barack-Obama', 'The name Barack means "one who is blessed" in Swahili. Obama was the first African-American U.S. president. Obama was the first president born outside of the contiguous United States. Obama was the eighth left-handed …', 'Barack Obama is the first Black president of the United States. Learn facts about him: his age, height, leadership legacy, quotes, family, and more.', 'Barack and Ann’s son, Barack Hussein Obama Jr., was born in Honolulu on August 4, 1961. Did you know? Not only was Obama the first African American president, he was also the first to be...', "President Obama's full name is Barack Hussein Obama. His full, birth name is Barack Hussein Obama, II. He was named after his father, Barack Hussein Obama, Sr., who …", 'When Barack Obama was elected president in 2008, he became the first African American to hold the office. The framers of the Constitution always hoped that our leadership would not be limited...']

1.1.2 使用第三方工具時ReAct

提示詞 hwchase17/self-ask-with-search

from langchain_community.utilities import SerpAPIWrapper
from langchain.agents import create_self_ask_with_search_agent, AgentType,Tool,AgentExecutor
from langchain import hub
from langchain_openai import ChatOpenAI
import os
from dotenv import load_dotenv, find_dotenv
# 刪除all_proxy環(huán)境變量
if 'all_proxy' in os.environ:
    del os.environ['all_proxy']
# 刪除ALL_PROXY環(huán)境變量
if 'ALL_PROXY' in os.environ:
    del os.environ['ALL_PROXY']
_ = load_dotenv(find_dotenv())
os.environ["SERPAPI_API_KEY"] = "xxx"
chat_model = ChatOpenAI(model="gpt-4o-mini", temperature=0)
# 實例化查詢工具
search = SerpAPIWrapper()
tools = [
    Tool(
        name="Intermediate Answer",
        func=search.run,
        description="useful for when you need to ask with search",
    )
]
prompt = hub.pull("hwchase17/self-ask-with-search")
self_ask_with_search = create_self_ask_with_search_agent(
     chat_model,tools,prompt
)
agent_executor = AgentExecutor(agent=self_ask_with_search, tools=tools,verbose=True,handle_parsing_errors=True)
reponse = agent_executor.invoke({"input": "成都舉辦的大運會是第幾屆大運會？2023年大運會舉辦地在哪里？"})
print(reponse)
print(chat_model.invoke("成都舉辦的大運會是第幾屆大運會？").content)
print(chat_model.invoke("2023年大運會舉辦地在哪里？").content)

輸出:

> Entering new AgentExecutor chain...
Could not parse output: Yes.
Follow up: 成都舉辦的大運會是由哪個組織舉辦的？

1. **成都舉辦的大運會是第幾屆大運會？**
- The 2023 Chengdu Universiade was the 31st Summer Universiade.

2. **2023年大運會舉辦地在哪里？**
- The 2023 Summer Universiade was held in Chengdu, China.

So the final answers are:
- 成都舉辦的大運會是第31屆大運會。
- 2023年大運會舉辦地是成都，China。

如果你還有其他問題或需要進一步的澄清，請隨時問我！
For troubleshooting, visit: https://python.langchain.com/docs/troubleshooting/errors/OUTPUT_PARSING_FAILUREInvalid or incomplete response

> Finished chain.
{'input': '成都舉辦的大運會是第幾屆大運會？2023年大運會舉辦地在哪里？', 'output': '31屆，成都，China'}
成都舉辦的世界大學生運動會是第31屆大運會。該屆大運會于2023年在中國成都舉行。
2023年大運會（世界大學生運動會）將于2023年在中國成都舉辦。

1.2 使用langchain內(nèi)置的工具

hwchase17/react

from langchain.agents import create_react_agent, AgentType,Tool,AgentExecutor
from langchain import hub
from langchain_community.agent_toolkits.load_tools import load_tools
from langchain.prompts import PromptTemplate
from dotenv import load_dotenv, find_dotenv
from langchain_openai import ChatOpenAI
import os
# 刪除all_proxy環(huán)境變量
if 'all_proxy' in os.environ:
    del os.environ['all_proxy']
# 刪除ALL_PROXY環(huán)境變量
if 'ALL_PROXY' in os.environ:
    del os.environ['ALL_PROXY']
_ = load_dotenv(find_dotenv())
os.environ["SERPAPI_API_KEY"] = "xxx"
chat_model = ChatOpenAI(model="gpt-4o-mini", temperature=0)
prompt = hub.pull("hwchase17/react")
tools = load_tools(["serpapi", "llm-math"], llm=chat_model)
agent = create_react_agent(chat_model, tools, prompt)
agent_executor = AgentExecutor(agent=agent, tools=tools,verbose=True)
reponse =agent_executor.invoke({"input": "誰是萊昂納多·迪卡普里奧的女朋友？她現(xiàn)在年齡的0.43次方是多少?"})
print(reponse)

輸出：

> Entering new AgentExecutor chain...
我需要先找到萊昂納多·迪卡普里奧目前的女朋友是誰，然后獲取她的年齡以計算年齡的0.43次方。
Action: Search
Action Input: "Leonardo DiCaprio girlfriend 2023"
Vittoria Ceretti我找到萊昂納多·迪卡普里奧的女朋友是維多利亞·切雷提（Vittoria Ceretti）。接下來，我需要找到她的年齡以計算0.43次方。
Action: Search
Action Input: "Vittoria Ceretti age 2023"

About 25 years維多利亞·切雷提（Vittoria Ceretti）大約25歲。接下來，我將計算25的0.43次方。
Action: Calculator
Action Input: 25 ** 0.43

Answer: 3.991298452658078我現(xiàn)在知道最終答案
Final Answer: 萊昂納多·迪卡普里奧的女朋友是維多利亞·切雷提，她的年齡0.43次方約為3.99。

> Finished chain.
{'input': '誰是萊昂納多·迪卡普里奧的女朋友？她現(xiàn)在年齡的0.43次方是多少?', 'output': '萊昂納多·迪卡普里奧的女朋友是維多利亞·切雷提，她的年齡0.43次方約為3.99。'}

1.3 使用自定義的工具

hwchase17/openai-functions-agent

1.3.1 簡單使用

from langchain_openai import ChatOpenAI
from  langchain.agents import tool,AgentExecutor,create_openai_functions_agent
from langchain import hub
import os
from dotenv import load_dotenv, find_dotenv
# 刪除all_proxy環(huán)境變量
if 'all_proxy' in os.environ:
    del os.environ['all_proxy']
# 刪除ALL_PROXY環(huán)境變量
if 'ALL_PROXY' in os.environ:
    del os.environ['ALL_PROXY']
_ = load_dotenv(find_dotenv())
chat_model = ChatOpenAI(model="gpt-4o-mini",temperature=0)
@tool
def get_word_length(word: str) -> int:
    """Returns the length of a word."""
    return len(word)
tools = [get_word_length]
prompt = hub.pull("hwchase17/openai-functions-agent")
agent = create_openai_functions_agent(llm=chat_model, tools=tools, prompt=prompt)
agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True,handle_parsing_errors=True)
agent_executor.invoke({"input": "單詞“educati”中有多少個字母?"})

輸出：

> Entering new AgentExecutor chain...

Invoking: `get_word_length` with `{'word': 'educati'}`

7單詞“educati”中有7個字母。

> Finished chain.

1.3.1 帶有記憶功能

from langchain_openai import ChatOpenAI
from  langchain.agents import tool,AgentExecutor,create_openai_functions_agent
from langchain_core.prompts.chat import ChatPromptTemplate
from langchain.prompts import MessagesPlaceholder
from langchain.memory import ConversationBufferMemory
from langchain import hub
import os
from dotenv import load_dotenv, find_dotenv
# 刪除all_proxy環(huán)境變量
if 'all_proxy' in os.environ:
    del os.environ['all_proxy']
# 刪除ALL_PROXY環(huán)境變量
if 'ALL_PROXY' in os.environ:
    del os.environ['ALL_PROXY']
_ = load_dotenv(find_dotenv())
chat_model = ChatOpenAI(model="gpt-4o-mini",temperature=0)
@tool
def get_word_length(word: str) -> int:
    """Returns the length of a word."""
    return len(word)
tools = [get_word_length]
prompt = ChatPromptTemplate.from_messages(
    [
        ("placeholder", "{chat_history}"),
        ("human", "{input}"),
        ("placeholder", "{agent_scratchpad}"),
    ]
)
memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)
agent = create_openai_functions_agent(llm=chat_model, tools=tools, prompt=prompt)
agent_executor = AgentExecutor(agent=agent, tools=tools, memory=memory, verbose=True)
agent_executor.invoke({"input":"單詞“educati”中有多少個字母?"})
agent_executor.invoke({"input":"那是一個真實的單詞嗎？"})

輸出：

> Entering new AgentExecutor chain...

Invoking: `get_word_length` with `{'word': 'educati'}`

7

單詞“educati”中有7個字母。

> Finished chain.

> Entering new AgentExecutor chain...
“educati”并不是一個標準的英語單詞。它可能是“education”的一個變形或拼寫錯誤。標準英語中的相關詞是“education”，意為“教育”。

> Finished chain.