2025.10.30
This commit is contained in:
@@ -5,7 +5,7 @@ LLM 调用|Prompt 设计|Chain 构建|Memory 记忆|实战练习
|
||||
|
||||
---
|
||||
|
||||
### 1. `01_Models.py`(模型调用)
|
||||
### 1. `01_models.py`(模型调用)
|
||||
封装 LLM 实例,实现标准调用流程。
|
||||
|
||||
✅ 掌握点:
|
||||
@@ -14,7 +14,7 @@ LLM 调用|Prompt 设计|Chain 构建|Memory 记忆|实战练习
|
||||
|
||||
---
|
||||
|
||||
### 2. `02_Prompt.py`(提示词构建)
|
||||
### 2. `02_prompt.py`(提示词构建)
|
||||
使用 `ChatPromptTemplate` 构建可复用的 Prompt 模板。
|
||||
|
||||
✅ 掌握点:
|
||||
@@ -24,7 +24,7 @@ LLM 调用|Prompt 设计|Chain 构建|Memory 记忆|实战练习
|
||||
|
||||
---
|
||||
|
||||
### 3. `03_Chain.py`(链式调用)
|
||||
### 3. `03_chain.py`(链式调用)
|
||||
将 Prompt + LLM + Parser 组合成 Chain。
|
||||
|
||||
✅ 掌握点:
|
||||
@@ -34,7 +34,7 @@ LLM 调用|Prompt 设计|Chain 构建|Memory 记忆|实战练习
|
||||
|
||||
---
|
||||
|
||||
### 4. `04_Memory.py`(记忆功能)
|
||||
### 4. `04_memory.py`(记忆功能)
|
||||
添加会话记忆,实现多轮对话。
|
||||
|
||||
✅ 掌握点:
|
||||
@@ -52,4 +52,6 @@ LLM 调用|Prompt 设计|Chain 构建|Memory 记忆|实战练习
|
||||
- 实际运行体验
|
||||
- 可直接扩展为 Web 应用
|
||||
|
||||
---
|
||||
|
||||
💡 建议:跑通后,试试让 AI 记住你喜欢的颜色,并在后续对话中提及。
|
||||
@@ -0,0 +1,23 @@
|
||||
import os
|
||||
from dotenv import load_dotenv
|
||||
|
||||
load_dotenv()
|
||||
api_key = os.getenv("OPENAI_API_KEY")
|
||||
|
||||
|
||||
from langchain_core.tools import tool
|
||||
|
||||
@tool
|
||||
def get_weather(location):
|
||||
"""模拟获得天气信息"""
|
||||
return f"{location}当前天气:23℃,晴,风力2级"
|
||||
|
||||
@tool
|
||||
def get_user_name(user):
|
||||
"""模拟获得用户名字"""
|
||||
return f'用户名字是:{user}'
|
||||
|
||||
|
||||
# 封装好我们要用的工具
|
||||
tools = [get_weather,get_user_name]
|
||||
print('工具箱已封装完毕!')
|
||||
@@ -0,0 +1,54 @@
|
||||
import os
|
||||
from dotenv import load_dotenv
|
||||
|
||||
load_dotenv()
|
||||
api_key = os.getenv("OPENAI_API_KEY")
|
||||
|
||||
|
||||
import os
|
||||
from langchain_openai import ChatOpenAI
|
||||
from langchain.agents import AgentExecutor, create_tool_calling_agent
|
||||
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
|
||||
from langchain_core.tools import tool # 导入 @tool
|
||||
|
||||
# 配置LLM
|
||||
llm = ChatOpenAI(
|
||||
model="deepseek-chat",
|
||||
api_key=api_key,
|
||||
base_url="https://api.deepseek.com"
|
||||
)
|
||||
|
||||
# 配置prompt
|
||||
prompt = ChatPromptTemplate.from_messages([
|
||||
("system","你是一个聪明的智能助手。当你遇到解决不了的问题时,会调用工具来解决问题。"),
|
||||
("human","{input}"),
|
||||
MessagesPlaceholder(variable_name="agent_scratchpad") # 必加,Agent的思考过程
|
||||
])
|
||||
|
||||
# 配置tool
|
||||
@tool
|
||||
def get_weather(location):
|
||||
"""模拟获得天气信息"""
|
||||
return f"{location}当前天气:23℃,晴,风力2级"
|
||||
|
||||
@tool
|
||||
def get_user_name(user):
|
||||
"""模拟获得用户名字"""
|
||||
return f'用户名字是:{user}'
|
||||
|
||||
|
||||
tools = [get_weather,get_user_name]
|
||||
|
||||
# 创建Agent(大脑)
|
||||
agent = create_tool_calling_agent(llm=llm,prompt=prompt,tools=tools)
|
||||
# 创建AgentExecutor(执行器)--负责运行ReAct循环
|
||||
agent_executor = AgentExecutor(agent=agent,tools=tools,verbose=True) # 开启verbose以看到ai思考链
|
||||
# 运行
|
||||
response = agent_executor.invoke({
|
||||
'input':"今天北京的天气怎么样?"
|
||||
})
|
||||
|
||||
print(response)
|
||||
print()
|
||||
print(response['output'])
|
||||
|
||||
@@ -0,0 +1,50 @@
|
||||
import os
|
||||
from dotenv import load_dotenv
|
||||
|
||||
load_dotenv()
|
||||
api_key = os.getenv("OPENAI_API_KEY")
|
||||
|
||||
import sqlite3
|
||||
from langchain_openai import ChatOpenAI
|
||||
from langchain_community.agent_toolkits import create_sql_agent
|
||||
from langchain_community.utilities import SQLDatabase # 导入 SQLDatabase
|
||||
|
||||
# 配置llm
|
||||
llm = ChatOpenAI(
|
||||
model="deepseek-chat",
|
||||
api_key=api_key,
|
||||
base_url="https://api.deepseek.com"
|
||||
)
|
||||
|
||||
# 创建一个临时的数据库--用于演示
|
||||
db_file = "test_sql.db"
|
||||
if os.path.exists(db_file):
|
||||
os.remove(db_file)
|
||||
conn = sqlite3.connect(db_file)
|
||||
cursor = conn.cursor()
|
||||
cursor.execute("CREATE TABLE users (id INT,name TEXT,age INT);")
|
||||
cursor.execute("INSERT INTO users (id,name,age) VALUES (1,'Alice',30);")
|
||||
cursor.execute("INSERT INTO users (id,name,age) VALUES (2,'Bob',25);")
|
||||
conn.commit()
|
||||
conn.close()
|
||||
|
||||
# 连接数据库 -- LangChain 使用 SQLAlchemy URI (连接方式)
|
||||
db_uri = f'sqlite:///{db_file}'
|
||||
db = SQLDatabase.from_uri(db_uri)
|
||||
|
||||
# 创建sqlAgent -- 一键完成,无需定义tools,仅告诉它使用openai-tools,即Tool Calling(工具调用)模式
|
||||
agent_executor = create_sql_agent(
|
||||
llm=llm,
|
||||
db=db,
|
||||
agent_type="openai-tools",
|
||||
verbose=True
|
||||
)
|
||||
|
||||
# 运行
|
||||
response = agent_executor.invoke({"input":"告诉我Alice多大了?"})
|
||||
print(response['output'])
|
||||
|
||||
# 清理
|
||||
db._engine.dispose() # 关闭连接池,避免文件被占用
|
||||
if os.path.exists(db_file):
|
||||
os.remove(db_file)
|
||||
@@ -0,0 +1,81 @@
|
||||
import os
|
||||
from dotenv import load_dotenv
|
||||
|
||||
load_dotenv()
|
||||
api_key = os.getenv("OPENAI_API_KEY")
|
||||
|
||||
|
||||
import os
|
||||
from langchain_openai import ChatOpenAI
|
||||
from langchain.agents import AgentExecutor, create_tool_calling_agent
|
||||
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
|
||||
from langchain_core.tools import tool # 导入 @tool
|
||||
from langchain_community.chat_message_histories import ChatMessageHistory
|
||||
from langchain_core.runnables import RunnableWithMessageHistory
|
||||
|
||||
|
||||
# 配置llm
|
||||
llm = ChatOpenAI(
|
||||
model="deepseek-chat",
|
||||
api_key=api_key,
|
||||
base_url="https://api.deepseek.com"
|
||||
)
|
||||
# 配置prompt(新增俩占位符 一个为对话历史记录,一个为agent的思考过程)
|
||||
prompt = ChatPromptTemplate.from_messages([
|
||||
('system','你是小智,一个帮助他人的智能助手。当你无法解答当前问题时,会调用工具来解决问题。'),
|
||||
MessagesPlaceholder(variable_name="history"),
|
||||
('human','{input}'),
|
||||
MessagesPlaceholder(variable_name="agent_scratchpad")
|
||||
])
|
||||
|
||||
# 配置tool
|
||||
@tool
|
||||
def get_weather(location):
|
||||
"""模拟获得天气信息"""
|
||||
return f"{location}当前天气:23℃,晴,风力2级"
|
||||
|
||||
|
||||
tools = [get_weather]
|
||||
# 配置agent
|
||||
agent = create_tool_calling_agent(llm=llm,prompt=prompt,tools=tools)
|
||||
# 配置AgentExecutor
|
||||
agent_executor = AgentExecutor(agent=agent,tools=tools) # 这里没加verbose=True,想打印日志看思考链的可以自行打印
|
||||
|
||||
|
||||
# 记忆存储--包装agent_executor
|
||||
store = {}
|
||||
def get_session_history(session_id:str):
|
||||
if session_id not in store:
|
||||
store[session_id] = ChatMessageHistory()
|
||||
return store[session_id]
|
||||
|
||||
|
||||
agent_with_memory = RunnableWithMessageHistory(
|
||||
runnable=agent_executor,
|
||||
get_session_history=get_session_history,
|
||||
input_messages_key="input",
|
||||
history_messages_key="history"
|
||||
)
|
||||
# 打印测试
|
||||
session_id = 'user123'
|
||||
|
||||
if __name__ == '__main__':
|
||||
while 1:
|
||||
user_input = input('\n你:')
|
||||
if user_input == 'quit':
|
||||
print('拜拜~')
|
||||
break
|
||||
response = agent_with_memory.invoke(
|
||||
{'input': user_input},
|
||||
config={'configurable': {'session_id': session_id}}
|
||||
)
|
||||
print(f"AI:{response['output']}")
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
@@ -0,0 +1,52 @@
|
||||
import os
|
||||
from dotenv import load_dotenv
|
||||
|
||||
load_dotenv()
|
||||
api_key = os.getenv("OPENAI_API_KEY")
|
||||
|
||||
|
||||
import time
|
||||
from langchain_openai import ChatOpenAI
|
||||
from langchain.globals import set_llm_cache
|
||||
from langchain_community.cache import InMemoryCache
|
||||
|
||||
# 配置llm
|
||||
llm = ChatOpenAI(
|
||||
model="deepseek-chat",
|
||||
api_key=api_key,
|
||||
base_url="https://api.deepseek.com"
|
||||
)
|
||||
# 设置全局缓存
|
||||
set_llm_cache(InMemoryCache())
|
||||
|
||||
# 第一次调用llm(会远程请求)
|
||||
query = "用中文写一句关于猫的五言诗。"
|
||||
start_time = time.time()
|
||||
response1 = llm.invoke(query).content
|
||||
print(f"第一次调用结果: {response1}")
|
||||
print(f"第一次运行时间: {time.time() - start_time:.4f} 秒")
|
||||
print('')
|
||||
|
||||
# 第二次调用llm(会命中缓存)
|
||||
start_time = time.time()
|
||||
response2 = llm.invoke(query).content
|
||||
print(f"第二次调用结果: {response2}")
|
||||
print(f"第二次运行时间 (已缓存): {time.time() - start_time:.4f} 秒")
|
||||
|
||||
# 清理
|
||||
set_llm_cache(None) # 关闭缓存,以免影响后续实例
|
||||
print('缓存清理完成')
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
@@ -0,0 +1,78 @@
|
||||
import os
|
||||
from dotenv import load_dotenv
|
||||
|
||||
load_dotenv()
|
||||
api_key = os.getenv("OPENAI_API_KEY")
|
||||
|
||||
|
||||
from langchain_openai import ChatOpenAI
|
||||
from langchain.agents import AgentExecutor, create_tool_calling_agent
|
||||
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
|
||||
from langchain_core.tools import tool # 导入 @tool
|
||||
from langchain_community.chat_message_histories import ChatMessageHistory
|
||||
from langchain_core.runnables import RunnableWithMessageHistory
|
||||
from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
|
||||
|
||||
|
||||
# 配置llm
|
||||
llm = ChatOpenAI(
|
||||
model="deepseek-chat",
|
||||
api_key=api_key,
|
||||
base_url="https://api.deepseek.com",
|
||||
streaming=True,
|
||||
callbacks=[StreamingStdOutCallbackHandler()]
|
||||
)
|
||||
# 配置prompt(新增俩占位符 一个为对话历史记录,一个为agent的思考过程)
|
||||
prompt = ChatPromptTemplate.from_messages([
|
||||
('system','你是小智,一个帮助他人的智能助手。当你无法解答当前问题时,会调用工具来解决问题。'),
|
||||
MessagesPlaceholder(variable_name="history"),
|
||||
('human','{input}'),
|
||||
MessagesPlaceholder(variable_name="agent_scratchpad")
|
||||
])
|
||||
|
||||
# 配置tool
|
||||
@tool
|
||||
def get_weather(location):
|
||||
"""模拟获得天气信息"""
|
||||
return f"{location}当前天气:23℃,晴,风力2级"
|
||||
|
||||
|
||||
tools = [get_weather]
|
||||
# 配置agent
|
||||
agent = create_tool_calling_agent(llm=llm,prompt=prompt,tools=tools)
|
||||
# 配置AgentExecutor
|
||||
agent_executor = AgentExecutor(agent=agent,tools=tools) # 这里没加verbose=True,想打印日志看思考链的可以自行打印
|
||||
|
||||
|
||||
# 记忆存储--包装agent_executor
|
||||
store = {}
|
||||
def get_session_history(session_id:str):
|
||||
if session_id not in store:
|
||||
store[session_id] = ChatMessageHistory()
|
||||
return store[session_id]
|
||||
|
||||
|
||||
agent_with_memory = RunnableWithMessageHistory(
|
||||
runnable=agent_executor,
|
||||
get_session_history=get_session_history,
|
||||
input_messages_key="input",
|
||||
history_messages_key="history"
|
||||
)
|
||||
# 打印测试
|
||||
session_id = 'user123'
|
||||
|
||||
if __name__ == '__main__':
|
||||
while 1:
|
||||
user_input = input('\n你:')
|
||||
if user_input == 'quit':
|
||||
print('拜拜~')
|
||||
break
|
||||
# 用于标记"AI:"这个内容
|
||||
# flush=True保证"AI:"立即输出,而不是等缓存区存满再输出
|
||||
print("AI: ", end="", flush=True)
|
||||
response = agent_with_memory.invoke(
|
||||
{'input': user_input},
|
||||
config={'configurable': {'session_id': session_id}}
|
||||
)
|
||||
print()
|
||||
|
||||
@@ -0,0 +1,69 @@
|
||||
## 🧩 模块说明:LangChain Agents 进阶核心组件
|
||||
|
||||
📌 **核心知识点**:
|
||||
Function Calling|@tool 工具封装|ReAct 循环|Agent 构建|SQL Agent|记忆+流式|开发优化
|
||||
|
||||
---
|
||||
|
||||
### 1. `01_define_toolbox.py`(工具函数定义)
|
||||
使用 `@tool` 装饰器将 Python 函数封装为 LangChain 可调用的工具。
|
||||
|
||||
✅ 掌握点:
|
||||
- 如何用 `@tool` 定义工具
|
||||
- 必须添加三引号描述(作为提示词)
|
||||
- 支持任意自定义逻辑函数
|
||||
|
||||
---
|
||||
|
||||
### 2. `02_general_agent.py`(通用 Agent 构建)
|
||||
基于 `create_tool_calling_agent` 创建支持工具调用的智能体。
|
||||
|
||||
✅ 掌握点:
|
||||
- ReAct 思维循环:思考 → 行动 → 观察 → 再思考
|
||||
- 使用 `MessagesPlaceholder("agent_scratchpad")` 记录推理过程
|
||||
- `AgentExecutor` 执行完整流程,支持 `verbose=True` 查看思考链
|
||||
|
||||
---
|
||||
|
||||
### 3. `03_sql_agent.py`(SQL 专用 Agent)
|
||||
一键构建自然语言查询数据库的智能体。
|
||||
|
||||
✅ 掌握点:
|
||||
- 使用 `create_sql_agent` 快速接入 SQLite 数据库
|
||||
- 自动分析表结构、生成 SQL 并执行
|
||||
- 无需手动定义工具,开箱即用
|
||||
|
||||
---
|
||||
|
||||
### 4. `04_memory_general_agent.py`(带记忆的 Agent)
|
||||
将 Agent 与对话历史结合,实现多轮上下文感知。
|
||||
|
||||
✅ 掌握点:
|
||||
- 在 Prompt 中加入 `MessagesPlaceholder("history")`
|
||||
- 使用 `RunnableWithMessageHistory` 包装 AgentExecutor
|
||||
- 实现“记住用户身份”、“持续追问”的真实对话体验
|
||||
|
||||
---
|
||||
|
||||
### 5. `05_caching.py`(缓存优化技巧)
|
||||
在开发调试阶段避免重复调用 LLM,节省成本与时间。
|
||||
|
||||
✅ 掌握点:
|
||||
- 使用 `set_llm_cache(InMemoryCache())` 启用缓存
|
||||
- 第一次调用远程请求,后续命中缓存秒出结果
|
||||
- 开发时开启,上线前关闭(仅用于测试)
|
||||
|
||||
---
|
||||
|
||||
### 6. `06_streaming.py`(流式输出实战)
|
||||
实现 AI 回复像打字机一样逐字输出,提升用户体验。
|
||||
|
||||
✅ 掌握点:
|
||||
- 设置 `streaming=True` 启用流式模式
|
||||
- 添加 `callbacks=[StreamingStdOutCallbackHandler()]` 实时打印 token
|
||||
- 使用 `print("AI: ", end="", flush=True)` 确保提示立即显示
|
||||
|
||||
---
|
||||
|
||||
💡 建议:
|
||||
跑通所有示例后,尝试将 `get_weather` 和 `query_user_info` 组合进一个带记忆的 Agent,让 AI 能连续问:“你是谁?”、“北京天气怎么样?”、“你喜欢什么颜色?”,并记住你的回答。
|
||||
@@ -27,7 +27,7 @@
|
||||
| | [02 LLM 基础调用](https://github.com/Annyfee/agent-craft/tree/main/02_llm_fundamentals) | [🏠](https://blog.csdn.net/2401_87328929/article/details/153735431) | LLM API 调用 · prompt · 上下文记忆 | ⭐ |
|
||||
| | [03 Function Calling 与工具调用](https://github.com/Annyfee/agent-craft/tree/main/03_function_calling_tools) | [🏠](https://blog.csdn.net/2401_87328929/article/details/153866573) | Function Call · 工具函数封装 | ⭐⭐ |
|
||||
| ⚙️ **框架篇** | [04 LangChain 基础篇](https://github.com/Annyfee/agent-craft/tree/main/04_langchain_basics) | [🏠](https://blog.csdn.net/2401_87328929/article/details/153978186) | LLM · Prompt · Chain · Memory | ⭐⭐ |
|
||||
| | 05 LangChain 进阶篇 | 🚧撰写中 | Agents · 多链协作 · 缓存 | ⭐⭐⭐ |
|
||||
| | 05 LangChain 进阶篇 | [🏠](https://blog.csdn.net/2401_87328929/article/details/154064397) | Agents · 多链协作 · 缓存 | ⭐⭐⭐ |
|
||||
| | 06 LangGraph 入门 | 🚧 | Flow · Node · Edge 控制 | ⭐⭐⭐ |
|
||||
| | 07 RAG 基础篇(Embedding & 向量) | 🚧 | 向量化 · 检索匹配 | ⭐⭐ |
|
||||
| | 08 RAG 进阶篇(检索 + 生成) | 🚧 | RAG Pipeline · QA 系统 | ⭐⭐⭐ |
|
||||
@@ -67,8 +67,13 @@
|
||||
- **目标**:认识Langchain六大模块,学会用Langchain构建智能体。
|
||||
- **内容**:LLM 调用|Prompt 设计|Chain 构建|Memory 记忆|实战练习
|
||||
|
||||
### ✅ 模块 05 — LangChain 进阶篇
|
||||
|
||||
> 📌 后续模块将陆续开放(LangChain、RAG、MCP、多智能体等)
|
||||
- **目标**:掌握Langchain Agents的核心机制,构建能调用工具、持续思考、具备记忆的智能体。
|
||||
- **内容**:Function Calling|@tool 工具封装|ReAct 循环|Agent 构建|SQL Agent|记忆+流式|开发优化
|
||||
|
||||
|
||||
> 📌 后续模块将陆续开放(LangGraph、RAG、MCP、多智能体等)
|
||||
|
||||
---
|
||||
|
||||
|
||||
Reference in New Issue
Block a user