07-08-周三_10-09-42

04_LangChain概述.ipynb

@@ -0,0 +1,303 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# 04 LangChain 概述\n",
+    "\n",
+    "## 学习目标\n",
+    "1. 理解 LangChain 产生的背景和解决的问题\n",
+    "2. 掌握 LangChain 的核心架构与组件组成\n",
+    "3. 能够运行第一个 LangChain 程序，体验链式调用的便利性\n",
+    "4. 了解 LangChain 生态中的相关工具（LangServe、LangSmith 等）"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 1. 为什么需要 LangChain\n",
+    "\n",
+    "前两节课中，我们直接通过 openai 或 anthropic SDK 调用大模型 API。这种方式简单直接，但当应用变得复杂时，会遇到以下问题：\n",
+    "\n",
+    "- **Prompt 管理混乱**：硬编码在代码中，难以维护和复用\n",
+    "- **代码重复**：每次调用都要写重复的 API 初始化、参数配置\n",
+    "- **模型切换困难**：换一个模型提供商需要重写大量代码\n",
+    "- **缺乏标准流程**：没有统一的输入处理、输出解析规范\n",
+    "\n",
+    "**LangChain** 就是为了解决这些问题而生的框架。它将大模型应用的开发流程抽象为标准化组件，让开发者可以像搭积木一样构建复杂的 AI 应用。"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 2. LangChain 是什么\n",
+    "\n",
+    "LangChain 是一个用于构建基于大语言模型（LLM）应用的 Python/JS 框架，由 Harrison Chase 于 2022 年创建。\n",
+    "\n",
+    "### 核心设计思想\n",
+    "\n",
+    "- **组件化**：将大模型应用拆解为可复用的标准组件\n",
+    "- **链式组合**：通过管道（|）将多个组件串联成处理流程\n",
+    "- **模型无关**：同一套代码可以切换不同厂商的模型\n",
+    "- **生态丰富**：提供大量预集成工具（向量库、搜索引擎、数据库等）"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 3. LangChain 核心架构\n",
+    "\n",
+    "LangChain 将大模型应用抽象为四大核心模块：\n",
+    "\n",
+    "### 四大核心模块\n",
+    "\n",
+    "| 模块 | 作用 | 本课程对应章节 |\n",
+    "| --- | --- | --- |\n",
+    "| **Model I/O** | 模型输入输出管理（Prompt、Model、Output Parser） | 05-07 节 |\n",
+    "| **Retrieval** | 检索增强生成（RAG）：文档加载、向量化、检索 | 11-13 节 |\n",
+    "| **Chains** | 链式组合，将多个组件串联成工作流 | 05、08 节 |\n",
+    "| **Agents** | 智能体：让模型自主决策、调用工具完成任务 | 09-10 节 |\n",
+    "\n",
+    "### 组件关系图\n",
+    "\n",
+    "用户输入 -> Prompt模板 -> LLM模型 -> 输出解析 -> 最终结果\n",
+    "\n",
+    "例如：\n",
+    "\n",
+    "chain = prompt | llm | parser\n",
+    "result = chain.invoke({'input': '用户问题'})"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 4. 第一个 LangChain 程序\n",
+    "\n",
+    "我们先看一个对比示例：直接调用 API vs 使用 LangChain。"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### 方式一：直接调用 API（传统方式）"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from openai import OpenAI\n",
+    "import os\n",
+    "from dotenv import load_dotenv\n",
+    "\n",
+    "load_dotenv()\n",
+    "\n",
+    "\n",
+    "client = OpenAI(\n",
+    "    base_url=os.getenv('OPENAI_BASE_URL'),\n",
+    "    api_key=os.getenv('OPENAI_API_KEY')\n",
+    ")\n",
+    "\n",
+    "response = client.chat.completions.create(\n",
+    "    model='qwen3.6-35b-A3b',\n",
+    "    messages=[\n",
+    "        {'role': 'system', 'content': '你是一个翻译助手，将中文翻译成英文。'},\n",
+    "        {'role': 'user', 'content': '你好，世界'}\n",
+    "    ]\n",
+    ")\n",
+    "print(response.choices[0].message.content)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### 方式二：使用 LangChain（推荐方式）\n",
+    "\n",
+    "LangChain 将 Prompt、模型调用、输出解析封装成标准组件，代码更简洁、更易维护。"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from langchain_openai import ChatOpenAI\n",
+    "from langchain_core.prompts import ChatPromptTemplate\n",
+    "import os\n",
+    "\n",
+    "# 1. 创建模型（自动读取环境变量 OPENAI_BASE_URL 和 OPENAI_API_KEY）\n",
+    "llm = ChatOpenAI(\n",
+    "    model='qwen3.6-35b-A3b',\n",
+    "    temperature=0.7\n",
+    ")\n",
+    "\n",
+    "# 2. 创建 Prompt 模板\n",
+    "prompt = ChatPromptTemplate.from_messages([\n",
+    "    ('system', '你是一个翻译助手，将中文翻译成英文。'),\n",
+    "    ('user', '{input}')\n",
+    "])\n",
+    "\n",
+    "# 3. 构建链：Prompt -> LLM\n",
+    "chain = prompt | llm\n",
+    "\n",
+    "# 4. 运行链\n",
+    "result = chain.invoke({'input': '你好，世界'})\n",
+    "print(result.content)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### 对比总结\n",
+    "\n",
+    "| 特性 | 直接 API | LangChain |\n",
+    "| --- | --- | --- |\n",
+    "| Prompt 管理 | 硬编码在 messages 中 | 模板化，支持变量复用 |\n",
+    "| 代码可读性 | 每次都要写完整 API 调用 | 语义清晰：prompt \\| llm |\n",
+    "| 模型切换 | 重写 SDK 初始化代码 | 换一行 model= 即可 |\n",
+    "| 输出处理 | 手动解析 choices[0] | 自动解析，支持结构化输出 |\n",
+    "| 扩展性 | 难以添加后续处理步骤 | 可继续追加 \\| parser 等组件 |"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 5. LangChain 核心组件速览\n",
+    "\n",
+    "本课程后续章节会逐一深入讲解，这里先建立整体印象。"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### 5.1 Model（模型）\n",
+    "\n",
+    "LangChain 封装了各大厂商的模型，统一接口：\n",
+    "\n",
+    "```python\n",
+    "from langchain_openai import ChatOpenAI      # OpenAI 兼容格式\n",
+    "from langchain_anthropic import ChatAnthropic # Anthropic Claude\n",
+    "from langchain_community.llms import Tongyi    # 阿里通义千问\n",
+    "\n",
+    "llm = ChatOpenAI(model='qwen3.6-35b-A3b')\n",
+    "```\n",
+    "\n",
+    "所有模型都支持 .invoke()、.stream()、.batch() 等统一方法。"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### 5.2 Prompt（提示模板）\n",
+    "\n",
+    "将 Prompt 模板化，支持变量插入和复用：\n",
+    "\n",
+    "```python\n",
+    "from langchain_core.prompts import ChatPromptTemplate\n",
+    "\n",
+    "prompt = ChatPromptTemplate.from_messages([\n",
+    "    ('system', '你是一个{role}。'),\n",
+    "    ('user', '请{task}：{content}')\n",
+    "])\n",
+    "\n",
+    "prompt.format(role='翻译助手', task='翻译', content='Hello')\n",
+    "```"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### 5.3 Output Parser（输出解析器）\n",
+    "\n",
+    "将模型输出的文本解析为结构化数据（JSON、列表、Pydantic 对象等）：\n",
+    "\n",
+    "```python\n",
+    "from langchain_core.output_parsers import JsonOutputParser\n",
+    "\n",
+    "parser = JsonOutputParser()\n",
+    "chain = prompt | llm | parser  # 模型输出自动解析为 JSON\n",
+    "```"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### 5.4 Chain（链）\n",
+    "\n",
+    "链是 LangChain 的核心概念，表示数据的处理流程。使用 | 运算符组合：\n",
+    "\n",
+    "```python\n",
+    "chain = prompt | llm | parser\n",
+    "result = chain.invoke({'input': '用户输入'})\n",
+    "```\n",
+    "\n",
+    "数据流：用户输入 -> Prompt模板 -> LLM -> 输出解析 -> 最终结果"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 6. LangChain 生态工具\n",
+    "\n",
+    "| 工具 | 作用 |\n",
+    "| --- | --- |\n",
+    "| **LangGraph** | 构建复杂状态图和多智能体工作流（本课程核心内容） |\n",
+    "| **LangServe** | 将 LangChain 链部署为 REST API 服务 |\n",
+    "| **LangSmith** | 调试、监控和评估 LangChain 应用的可观测性平台 |\n",
+    "| **LangChain Hub** | 社区共享的 Prompt 模板市场 |\n",
+    "\n",
+    "本课程重点讲解 LangChain 核心组件 + LangGraph 智能体，LangSmith 会在最后的部署评估环节介绍。"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 7. 本节课练习\n",
+    "\n",
+    "1. 对比直接调用 API 和 LangChain 的代码，体会 LangChain 的便利性\n",
+    "2. 修改上面的 LangChain 示例，将 system 提示改为「你是一个诗人」，将输入改为「写一句关于月亮的诗」\n",
+    "3. 尝试在 ChatOpenAI 中切换 temperature（0.0、1.0、2.0），观察输出变化\n",
+    "4. 尝试将 prompt | llm 链保存为一个变量，多次调用 invoke() 传入不同输入"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": ".venv",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.14.4"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 4
+}