大模型从入门到应用——LangChain：代理（Agents）-[代理执行器（Agent Executor）：结合使用Agent和VectorStore]

分类目录：《大模型从入门到应用》总目录

LangChain系列文章：

• 基础知识
• 快速入门
  • 安装与环境配置
  • 链（Chains）、代理（Agent:）和记忆（Memory）
  • 快速开发聊天模型
• 模型（Models）
  • 基础知识
  • 大型语言模型（LLMs）
    • 基础知识
    • LLM的异步API、自定义LLM包装器、虚假LLM和人类输入LLM（Human Input LLM）
    • 缓存LLM的调用结果
    • 加载与保存LLM类、流式传输LLM与Chat Model响应和跟踪tokens使用情况
  • 聊天模型（Chat Models）
    • 基础知识
    • 使用少量示例和响应流式传输
  • 文本嵌入模型
    • Aleph Alpha、Amazon Bedrock、Azure OpenAI、Cohere等
    • Embaas、Fake Embeddings、Google Vertex AI PaLM等
• 提示（Prompts）
  • 基础知识
  • 提示模板
    • 基础知识
    • 连接到特征存储
    • 创建自定义提示模板和含有Few-Shot示例的提示模板
    • 部分填充的提示模板和提示合成
    • 序列化提示信息
  • 示例选择器（Example Selectors）
  • 输出解析器（Output Parsers）
• 记忆（Memory）
  • 基础知识
  • 记忆的类型
    • 会话缓存记忆、会话缓存窗口记忆和实体记忆
    • 对话知识图谱记忆、对话摘要记忆和会话摘要缓冲记忆
    • 对话令牌缓冲存储器和基于向量存储的记忆
  • 将记忆添加到LangChain组件中
  • 自定义对话记忆与自定义记忆类
  • 聊天消息记录
  • 记忆的存储与应用
• 索引（Indexes）
  • 基础知识
  • 文档加载器（Document Loaders）
  • 文本分割器（Text Splitters）
  • 向量存储器（Vectorstores）
  • 检索器（Retrievers）
• 链（Chains）
  • 基础知识
  • 通用功能
    • 自定义Chain和Chain的异步API
    • LLMChain和RouterChain
    • SequentialChain和TransformationChain
    • 链的保存（序列化）与加载（反序列化）
  • 链与索引
    • 文档分析和基于文档的聊天
    • 问答的基础知识
    • 图问答（Graph QA）和带来源的问答（Q&A with Sources）
    • 检索式问答
    • 文本摘要（Summarization）、HyDE和向量数据库的文本生成
• 代理（Agents）
  • 基础知识
  • 代理类型
  • 自定义代理（Custom Agent）
  • 自定义MRKL代理
  • 带有ChatModel的LLM聊天自定义代理和自定义多操作代理（Custom MultiAction Agent）
  • 工具
    • 基础知识
    • 自定义工具（Custom Tools）
    • 多输入工具和工具输入模式
    • 人工确认工具验证和Tools作为OpenAI函数
  • 工具包（Toolkit）
  • 代理执行器（Agent Executor）
    • 结合使用Agent和VectorStore
    • 使用Agents的异步API和创建ChatGPT克隆
    • 处理解析错误、访问中间步骤和限制最大迭代次数
    • 为代理程序设置超时时间和限制最大迭代次数和为代理程序和其工具添加共享内存
  • 计划与执行
• 回调函数（Callbacks）

---

代理执行器接受一个代理和工具，并使用代理来决定调用哪些工具以及以何种顺序调用。本文将参数如何结合使用Agent和VectorStore。这种用法是将数据加载到VectorStore中，并希望以Agent的方式与之进行交互。

推荐的方法是创建一个RetrievalQA，然后将其作为整体Agent中的工具来使用。让我们在下面看一下如何实现，我们可以使用多个不同的vectordbs，将Agent作为它们之间的路由器。有两种不同的方法可以实现这一点：

• 让Agent像正常工具一样使用vectorstores
• 设置return_direct=True来将Agent真正用作路由

创建VectorStore

from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.vectorstores import Chroma
from langchain.text_splitter import CharacterTextSplitter
from langchain.llms import OpenAI
from langchain.chains import RetrievalQA
llm = OpenAI(temperature=0)
from pathlib import Path
relevant_parts = []
for p in Path(".").absolute().parts:
    relevant_parts.append(p)
    if relevant_parts[-3:] == ["langchain", "docs", "modules"]:
        break
doc_path = str(Path(*relevant_parts) / "state_of_the_union.txt")
from langchain.document_loaders import TextLoader
loader = TextLoader(doc_path)
documents = loader.load()
text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
texts = text_splitter.split_documents(documents)

embeddings = OpenAIEmbeddings()
docsearch = Chroma.from_documents(texts, embeddings, collection_name="state-of-union")

日志输出：

Running Chroma using direct local API.
Using DuckDB in-memory for database. Data will be transient.
state_of_union = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=docsearch.as_retriever())

输入：

from langchain.document_loaders import WebBaseLoader
loader = WebBaseLoader("https://beta.ruff.rs/docs/faq/")
docs = loader.load()
ruff_texts = text_splitter.split_documents(docs)
ruff_db = Chroma.from_documents(ruff_texts, embeddings, collection_name="ruff")
ruff = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=ruff_db.as_retriever())

日志输出：

Running Chroma using direct local API.
Using DuckDB in-memory for database. Data will be transient.

创建代理

# Import things that are needed generically
from langchain.agents import initialize_agent, Tool
from langchain.agents import AgentType
from langchain.tools import BaseTool
from langchain.llms import OpenAI
from langchain import LLMMathChain, SerpAPIWrapper
tools = [
    Tool(
        name = "State of Union QA System",
        func=state_of_union.run,
        description="useful for when you need to answer questions about the most recent state of the union address. Input should be a fully formed question."
    ),
    Tool(
        name = "Ruff QA System",
        func=ruff.run,
        description="useful for when you need to answer questions about ruff (a python linter). Input should be a fully formed question."
    ),
]
# Construct the agent. We will use the default agent type here.
# See documentation for a full list of options.
agent = initialize_agent(tools, llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True)
agent.run("What did biden say about ketanji brown jackson in the state of the union address?")

日志输出：

Entering new AgentExecutor chain...
I need to find out what Biden said about Ketanji Brown Jackson in the State of the Union address.
Action: State of Union QA System
Action Input: What did Biden say about Ketanji Brown Jackson in the State of the Union address?
Observation:  Biden said that Jackson is one of the nation's top legal minds and that she will continue Justice Breyer's legacy of excellence.
Thought:I now know the final answer
Final Answer: Biden said that Jackson is one of the nation's top legal minds and that she will continue Justice Breyer's legacy of excellence.

Finished chain.

输出：

"Biden said that Jackson is one of the nation's top legal minds and that she will continue Justice Breyer's legacy of excellence."

输入：

agent.run("Why use ruff over flake8?")

输出：

Entering new AgentExecutor chain...
I need to find out the advantages of using ruff over flake8
Action: Ruff QA System
Action Input: What are the advantages of using ruff over flake8?
Observation:  Ruff can be used as a drop-in replacement for Flake8 when used (1) without or with a small number of plugins, (2) alongside Black, and (3) on Python 3 code. It also re-implements some of the most popular Flake8 plugins and related code quality tools natively, including isort, yesqa, eradicate, and most of the rules implemented in pyupgrade. Ruff also supports automatically fixing its own lint violations, which Flake8 does not.
Thought:I now know the final answer
Final Answer: Ruff can be used as a drop-in replacement for Flake8 when used (1) without or with a small number of plugins, (2) alongside Black, and (3) on Python 3 code. It also re-implements some of the most popular Flake8 plugins and related code quality tools natively, including isort, yesqa, eradicate, and most of the rules implemented in pyupgrade. Ruff also supports automatically fixing its own lint violations, which Flake8 does not.

Finished chain.

输出：

'Ruff can be used as a drop-in replacement for Flake8 when used (1) without or with a small number of plugins, (2) alongside Black, and (3) on Python 3 code. It also re-implements some of the most popular Flake8 plugins and related code quality tools natively, including isort, yesqa, eradicate, and most of the rules implemented in pyupgrade. Ruff also supports automatically fixing its own lint violations, which Flake8 does not.'

仅将Agent用作路由器

如果我们打算将Agent用作路由，并且只想直接返回RetrievalQAChain的结果，我们还可以设置return_direct=True。

需要注意的是，在上面的示例中，Agent在查询RetrievalQAChain之后还做了一些额外的工作，我们可以避免这样做，直接返回结果。

tools = [
    Tool(
        name = "State of Union QA System",
        func=state_of_union.run,
        description="useful for when you need to answer questions about the most recent state of the union address. Input should be a fully formed question.",
        return_direct=True
    ),
    Tool(
        name = "Ruff QA System",
        func=ruff.run,
        description="useful for when you need to answer questions about ruff (a python linter). Input should be a fully formed question.",
        return_direct=True
    ),
]
agent = initialize_agent(tools, llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True)
agent.run("What did biden say about ketanji brown jackson in the state of the union address?")

日志输出：

Entering new AgentExecutor chain...
I need to find out what Biden said about Ketanji Brown Jackson in the State of the Union address.
Action: State of Union QA System
Action Input: What did Biden say about Ketanji Brown Jackson in the State of the Union address?
Observation:  Biden said that Jackson is one of the nation's top legal minds and that she will continue Justice Breyer's legacy of excellence.


Finished chain.

输出：

" Biden said that Jackson is one of the nation's top legal minds and that she will continue Justice Breyer's legacy of excellence."

输入：

agent.run("Why use ruff over flake8?")

日志输出：

Entering new AgentExecutor chain...
I need to find out the advantages of using ruff over flake8
Action: Ruff QA System
Action Input: What are the advantages of using ruff over flake8?
Observation:  Ruff can be used as a drop-in replacement for Flake8 when used (1) without or with a small number of plugins, (2) alongside Black, and (3) on Python 3 code. It also re-implements some of the most popular Flake8 plugins and related code quality tools natively, including isort, yesqa, eradicate, and most of the rules implemented in pyupgrade. Ruff also supports automatically fixing its own lint violations, which Flake8 does not.


Finished chain.

输出：

' Ruff can be used as a drop-in replacement for Flake8 when used (1) without or with a small number of plugins, (2) alongside Black, and (3) on Python 3 code. It also re-implements some of the most popular Flake8 plugins and related code quality tools natively, including isort, yesqa, eradicate, and most of the rules implemented in pyupgrade. Ruff also supports automatically fixing its own lint violations, which Flake8 does not.'

多跳向量存储推理

由于vectorstores可以很容易地作为Agent中的工具使用，因此可以轻松使用现有的Agent框架回答依赖于vectorstores的多跳问题。

tools = [
    Tool(
        name = "State of Union QA System",
        func=state_of_union.run,
        description="useful for when you need to answer questions about the most recent state of the union address. Input should be a fully formed question, not referencing any obscure pronouns from the conversation before."
    ),
    Tool(
        name = "Ruff QA System",
        func=ruff.run,
        description="useful for when you need to answer questions about ruff (a python linter). Input should be a fully formed question, not referencing any obscure pronouns from the conversation before."
    ),
]
# Construct the agent. We will use the default agent type here.
# See documentation for a full list of options.
agent = initialize_agent(tools, llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True)
agent.run("What tool does ruff use to run over Jupyter Notebooks? Did the president mention that tool in the state of the union?")

日志输出：

Entering new AgentExecutor chain...
I need to find out what tool ruff uses to run over Jupyter Notebooks, and if the president mentioned it in the state of the union.
Action: Ruff QA System
Action Input: What tool does ruff use to run over Jupyter Notebooks?
Observation:  Ruff is integrated into nbQA, a tool for running linters and code formatters over Jupyter Notebooks. After installing ruff and nbqa, you can run Ruff over a notebook like so: > nbqa ruff Untitled.ipynb
Thought:I now need to find out if the president mentioned this tool in the state of the union.
Action: State of Union QA System
Action Input: Did the president mention nbQA in the state of the union?
Observation:  No, the president did not mention nbQA in the state of the union.
Thought:I now know the final answer.
Final Answer: No, the president did not mention nbQA in the state of the union.

Finished chain.

输出：

'No, the president did not mention nbQA in the state of the union.'

参考文献：
[1] LangChain官方网站：https://www.langchain.com/
[2] LangChain ️ 中文网，跟着LangChain一起学LLM/GPT开发：https://www.langchain.com.cn/
[3] LangChain中文网 - LangChain 是一个用于开发由语言模型驱动的应用程序的框架：http://www.cnlangchain.com/