Langchain学习笔记

2025-09-15   1064 words   ~3 mins
#langchain #ai #python

最近在研究langchain和langgraph，记录一下学习过程。

参考文章

  • Agent入门说明：https://medium.com/data-science/building-a-simple-agent-with-tools-and-toolkits-in-langchain-77e0f9bd1fa5
  • 工具链文档：https://python.langchain.com/docs/integrations/toolkits/
  • DeepSeek配置：https://docs.langchain.com/oss/python/integrations/chat/deepseek
  • 模型文档：https://docs.langchain.com/oss/python/integrations/chat
  • LangGraph概览：https://docs.langchain.com/oss/python/langgraph/overview

环境配置

python3 -m venv venv
source venv/bin/activate
pip install langchain langchain_openai langchain-deepseek
pip install -qU langchain-deepseek

调用DeepSeek的demo

from langchain_deepseek import ChatDeepSeek

llm = ChatDeepSeek(
    model="deepseek-chat",
    api_key="sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
    api_base="https://api.deepseek.com/v1",
    extra_body={"reasoning": {"enabled": True}},
    temperature=0,
    max_tokens=None,
    timeout=None,
    max_retries=2,
)

messages = [
    ("system", "You are a helpful assistant that translates English to Chinese. Translate the user sentence."),
    ("human", "I love programming."),
]

ai_msg = llm.invoke(messages)
print(ai_msg.content)

langchain-tool

工具定义需要保留类型声明，这样对AI更友好。

from langchain.tools import tool

@tool
def search_database(query: str, limit: int = 10) -> str:
    """Search the customer database for records matching the query.

    Args:
        query: Search terms to look for
        limit: Maximum number of results to return
    """
    return f"Found {limit} results for '{query}'"

自定义工具名称：

@tool("web_search", description="Performs web search.")
def search(query: str) -> str:
    """Search the web for information."""
    return f"Results for: {query}"

工具运行时状态获取：

from langchain.tools import tool, ToolRuntime

def get_weather(city: str) -> str:
    runtime: ToolRuntime
    print(runtime.state["messages"])

langchain-model

模型是智能体的推理引擎，驱动智能体的决策过程。

模型支持三种调用方式：直接调用、流式调用、批量调用。

# 直接调用
response = model.invoke(conversation)

# 流式调用
for chunk in model.stream("Why do parrots have colorful feathers?"):
    print(chunk.text, end="|", flush=True)

# 批量调用
responses = model.batch([
    "Why do parrots have colorful feathers?",
    "How do airplanes fly?",
    "What is quantum computing?"
])
for response in responses:
    print(response)

模型与工具绑定：

llm = llm.bind_tools([get_weather])

ai_msg = llm.invoke(messages)
for i in ai_msg.tool_calls:
    print(i)

image-20251209212919510

langchain-agent

智能体将语言模型与工具相结合，创建能够对任务进行推理、决定使用哪些工具并迭代地寻求解决方案的系统。

image-20251209214231633

from langchain.agents import create_agent

llm = ChatDeepSeek(
    model="deepseek-chat",
    api_key="sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
    api_base="https://api.deepseek.com/v1",
    extra_body={"reasoning": {"enabled": True}},
    temperature=0,
)

tools = [get_weather]

agent = create_agent(llm, tools=tools)
response = agent.invoke({"messages": [HumanMessage("今天成都的天气怎么样？")]})
print(response)

langgraph

LangGraph用于构建有状态的agent工作流。

基本步骤：

  1. 定义状态
  2. 定义模型节点
  3. 定义工具节点
  4. 定义结束条件
  5. 构建图并编译
from langgraph.graph import StateGraph, MessagesState

# 新建工作流
agent_builder = StateGraph(MessagesState)

# 新增节点
agent_builder.add_node("agent", model_node)
agent_builder.add_node("tools", tool_node)

# 新增边界连接节点
agent_builder.add_conditional_edges("agent", should_continue)
agent_builder.add_edge("tools", "agent")

# 编译图
agent = agent_builder.compile()

使用注意点：

  • 工具函数保持无状态特性
  • state的处理在tool_node中更新
  • 集中管理时间记录逻辑
  • 不依赖工具是否支持ainvoke

jupyter调试

用jupyter调试langgraph可以快速看到生成的图和代码。

source venv/bin/activate && jupyter lab
← prev next →