357 lines
12 KiB
Plaintext
357 lines
12 KiB
Plaintext
{
|
||
"cells": [
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"# 03 大模型 API 调用\n",
|
||
"\n",
|
||
"## 学习目标\n",
|
||
"1. 理解大模型 API 的基本概念（OpenAI、Anthropic 等）\n",
|
||
"2. 学会使用 Python 代码调用大模型 API 进行对话\n",
|
||
"3. 掌握 API 密钥的安全管理和环境变量配置\n",
|
||
"4. 能够区分 OpenAI 格式与 Anthropic 格式的 API 调用方式"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"## 1. 大模型 API 概述\n",
|
||
"\n",
|
||
"大模型（Large Language Model, LLM）通常通过 **HTTP API** 提供服务。你发送一段文本（Prompt），模型返回生成的回复。\n",
|
||
"\n",
|
||
"### 主流 API 提供商\n",
|
||
"\n",
|
||
"| 提供商 | 代表模型 | 特点 |\n",
|
||
"| --- | --- | --- |\n",
|
||
"| **OpenAI** | GPT-4o、GPT-4o-mini | 生态最成熟，很多第三方平台兼容其格式 |\n",
|
||
"| **Anthropic** | Claude 3.5 Sonnet、Claude 3 Opus | 长文本处理能力强，安全性设计突出 |\n",
|
||
"| **国内厂商** | 文心一言、通义千问、智谱 GLM | 中文优化好，无需翻墙 |\n",
|
||
"\n",
|
||
"> 💡 **注意**：很多国内平台（如硅基流动、智谱 AI、DashScope）提供兼容 OpenAI 格式的 API，可以直接使用 OpenAI SDK 调用。"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"## 2. API 调用核心概念\n",
|
||
"\n",
|
||
"调用大模型 API 时，你需要了解以下几个核心要素：\n",
|
||
"\n",
|
||
"- **Base URL**：API 服务的基础地址，例如 `https://api.openai.com/v1` 或 `https://api.anthropic.com`\n",
|
||
"- **API Key**：身份验证密钥，类似于密码，**绝对不能泄露**\n",
|
||
"- **Model**：模型名称，本课程统一使用 `qwen3.6-35b-A3b`\n",
|
||
"- **Message / Prompt**：发送给模型的输入内容\n",
|
||
"- **Temperature**：控制输出随机性（0~2，越低越确定，越高越有创意）\n",
|
||
"- **Max Tokens**：限制模型输出的最大长度"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"## 3. API 密钥安全管理\n",
|
||
"\n",
|
||
"**切勿将 API Key 直接写入代码并上传到 GitHub！** 推荐做法是将密钥存放在环境变量中。\n",
|
||
"\n",
|
||
"### 方法一：使用 .env 文件（推荐）\n",
|
||
"\n",
|
||
"1. 在项目根目录创建 `.env` 文件（注意文件名以点开头）\n",
|
||
"2. 添加以下内容（填入老师提供的实际值）：\n",
|
||
"\n",
|
||
"```env\n",
|
||
"# 请使用老师提供的 base_url 和 api_key\n",
|
||
"OPENAI_BASE_URL=https://your-openai-compatible-base-url.com/v1\n",
|
||
"OPENAI_API_KEY=sk-your-openai-api-key\n",
|
||
"\n",
|
||
"ANTHROPIC_BASE_URL=https://api.anthropic.com\n",
|
||
"ANTHROPIC_API_KEY=sk-ant-your-anthropic-api-key\n",
|
||
"```\n",
|
||
"\n",
|
||
"> ⚠️ **重要**：`.env` 文件已加入 `.gitignore`，不会上传到 Git，切勿手动删除该忽略配置。\n",
|
||
"\n",
|
||
"### 方法二：手动设置系统环境变量（Windows）\n",
|
||
"\n",
|
||
"```powershell\n",
|
||
"# 当前终端会话有效\n",
|
||
"$env:OPENAI_API_KEY=\"sk-your-api-key\"\n",
|
||
"$env:OPENAI_BASE_URL=\"https://your-base-url.com/v1\"\n",
|
||
"```"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": null,
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"# 加载 .env 文件中的环境变量\n",
|
||
"from dotenv import load_dotenv\n",
|
||
"import os\n",
|
||
"\n",
|
||
"load_dotenv() # 默认加载当前目录下的 .env 文件\n",
|
||
"\n",
|
||
"# 验证环境变量是否加载成功\n",
|
||
"openai_base_url = os.getenv(\"OPENAI_BASE_URL\")\n",
|
||
"openai_api_key = os.getenv(\"OPENAI_API_KEY\")\n",
|
||
"anthropic_base_url = os.getenv(\"ANTHROPIC_BASE_URL\")\n",
|
||
"anthropic_api_key = os.getenv(\"ANTHROPIC_API_KEY\")\n",
|
||
"\n",
|
||
"print(\"✅ OPENAI_BASE_URL 已设置\" if openai_base_url else \"❌ OPENAI_BASE_URL 未设置\")\n",
|
||
"print(\"✅ OPENAI_API_KEY 已设置\" if openai_api_key else \"❌ OPENAI_API_KEY 未设置\")\n",
|
||
"print(\"✅ ANTHROPIC_BASE_URL 已设置\" if anthropic_base_url else \"❌ ANTHROPIC_BASE_URL 未设置\")\n",
|
||
"print(\"✅ ANTHROPIC_API_KEY 已设置\" if anthropic_api_key else \"❌ ANTHROPIC_API_KEY 未设置\")"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"## 4. 方式一：OpenAI 格式调用（兼容 OpenAI SDK）\n",
|
||
"\n",
|
||
"OpenAI 格式的 API 是目前最通用的标准。很多国内厂商（如硅基流动、智谱、DeepSeek）也提供兼容此格式的接口。\n",
|
||
"\n",
|
||
"### 安装依赖\n",
|
||
"\n",
|
||
"```powershell\n",
|
||
"pip install openai python-dotenv -i https://pypi.tuna.tsinghua.edu.cn/simple\n",
|
||
"```"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": null,
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"from openai import OpenAI\n",
|
||
"import os\n",
|
||
"\n",
|
||
"# 初始化客户端（使用环境变量中的配置）\n",
|
||
"client = OpenAI(\n",
|
||
" base_url=os.getenv(\"OPENAI_BASE_URL\"),\n",
|
||
" api_key=os.getenv(\"OPENAI_API_KEY\")\n",
|
||
")\n",
|
||
"\n",
|
||
"# 发送对话请求\n",
|
||
"response = client.chat.completions.create(\n",
|
||
" model=\"qwen3.6-35b-A3b\", # 请根据老师提供的实际模型名称修改\n",
|
||
" messages=[\n",
|
||
" {\"role\": \"system\", \"content\": \"你是一个 helpful 的 AI 助手。\"},\n",
|
||
" {\"role\": \"user\", \"content\": \"请用一句话解释什么是大语言模型。\"}\n",
|
||
" ],\n",
|
||
" temperature=0.7,\n",
|
||
" max_tokens=200\n",
|
||
")\n",
|
||
"\n",
|
||
"# 输出模型回复\n",
|
||
"print(\"🤖 AI 回复：\")\n",
|
||
"print(response.choices[0].message.content)"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"### OpenAI 格式调用要点\n",
|
||
"\n",
|
||
"| 参数 | 说明 |\n",
|
||
"| --- | --- |\n",
|
||
"| `model` | 模型名称，本课程默认使用 `qwen3.6-35b-A3b` |\n",
|
||
"| `messages` | 消息列表，每条消息包含 `role`（system/user/assistant）和 `content` |\n",
|
||
"| `temperature` | 随机性（0~2），0 最确定，2 最有创意 |\n",
|
||
"| `max_tokens` | 最大输出 token 数 |\n",
|
||
"| `response.choices[0].message.content` | 获取模型回复的文本内容 |"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"## 5. 扩展阅读：Anthropic 格式调用（Claude API）\n",
|
||
"\n",
|
||
"Anthropic 的 Claude 系列模型使用独立的 API 格式，与 OpenAI 格式不同。需要安装 `anthropic` Python SDK。\n",
|
||
"\n",
|
||
"> 课程主线统一使用 OpenAI 兼容格式调用 `qwen3.6-35b-A3b`。以下 Anthropic/Claude 内容仅用于了解不同 API 格式，不作为默认运行示例。\n",
|
||
"\n",
|
||
"### 安装依赖\n",
|
||
"\n",
|
||
"```powershell\n",
|
||
"pip install anthropic python-dotenv -i https://pypi.tuna.tsinghua.edu.cn/simple\n",
|
||
"```\n"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"### Anthropic/Claude 调用格式示意（扩展阅读）\n",
|
||
"\n",
|
||
"课程默认不执行 Claude 调用，主线统一使用 OpenAI 兼容格式调用 `qwen3.6-35b-A3b`。如需了解 Anthropic SDK 的写法，可参考下面的非执行示例：\n",
|
||
"\n",
|
||
"```python\n",
|
||
"from anthropic import Anthropic\n",
|
||
"import os\n",
|
||
"\n",
|
||
"client = Anthropic(\n",
|
||
" base_url=os.getenv(\"ANTHROPIC_BASE_URL\"),\n",
|
||
" api_key=os.getenv(\"ANTHROPIC_API_KEY\"),\n",
|
||
")\n",
|
||
"\n",
|
||
"response = client.messages.create(\n",
|
||
" model=\"<your-claude-model>\",\n",
|
||
" max_tokens=200,\n",
|
||
" temperature=0.7,\n",
|
||
" system=\"你是一个 helpful 的 AI 助手。\",\n",
|
||
" messages=[\n",
|
||
" {\"role\": \"user\", \"content\": \"请用一句话解释什么是大语言模型。\"}\n",
|
||
" ],\n",
|
||
")\n",
|
||
"\n",
|
||
"print(response.content[0].text)\n",
|
||
"```\n"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"### Anthropic 格式与 OpenAI 格式的区别\n",
|
||
"\n",
|
||
"| 特性 | OpenAI 格式 | Anthropic 格式 |\n",
|
||
"| --- | --- | --- |\n",
|
||
"| 系统提示 | 放入 `messages` 列表，role=\"system\" | 独立的 `system` 参数 |\n",
|
||
"| 回复获取 | `response.choices[0].message.content` | `response.content[0].text` |\n",
|
||
"| SDK 包名 | `openai` | `anthropic` |\n",
|
||
"| 客户端类 | `OpenAI()` | `Anthropic()` |\n",
|
||
"| 方法名 | `chat.completions.create()` | `messages.create()` |"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"## 6. 多轮对话示例\n",
|
||
"\n",
|
||
"多轮对话需要保留历史消息，让模型理解上下文。以下以 OpenAI 格式为例："
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": null,
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"from openai import OpenAI\n",
|
||
"import os\n",
|
||
"\n",
|
||
"client = OpenAI(\n",
|
||
" base_url=os.getenv(\"OPENAI_BASE_URL\"),\n",
|
||
" api_key=os.getenv(\"OPENAI_API_KEY\")\n",
|
||
")\n",
|
||
"\n",
|
||
"# 维护对话历史\n",
|
||
"messages = [\n",
|
||
" {\"role\": \"system\", \"content\": \"你是一个 helpful 的 AI 助手。\"}\n",
|
||
"]\n",
|
||
"\n",
|
||
"# 第一轮对话\n",
|
||
"messages.append({\"role\": \"user\", \"content\": \"你好，我叫小明。\"})\n",
|
||
"response = client.chat.completions.create(\n",
|
||
" model=\"qwen3.6-35b-A3b\",\n",
|
||
" messages=messages,\n",
|
||
" temperature=0.7\n",
|
||
")\n",
|
||
"reply = response.choices[0].message.content\n",
|
||
"print(f\"🤖: {reply}\")\n",
|
||
"\n",
|
||
"# 将模型回复加入历史\n",
|
||
"messages.append({\"role\": \"assistant\", \"content\": reply})\n",
|
||
"\n",
|
||
"# 第二轮对话（模型应该记得我叫小明）\n",
|
||
"messages.append({\"role\": \"user\", \"content\": \"你还记得我的名字吗？\"})\n",
|
||
"response = client.chat.completions.create(\n",
|
||
" model=\"qwen3.6-35b-A3b\",\n",
|
||
" messages=messages,\n",
|
||
" temperature=0.7\n",
|
||
")\n",
|
||
"reply = response.choices[0].message.content\n",
|
||
"print(f\"🤖: {reply}\")"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"## 7. 流式输出（Streaming）\n",
|
||
"\n",
|
||
"流式输出让模型内容逐字返回，提升用户体验："
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": null,
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"from openai import OpenAI\n",
|
||
"import os\n",
|
||
"\n",
|
||
"client = OpenAI(\n",
|
||
" base_url=os.getenv(\"OPENAI_BASE_URL\"),\n",
|
||
" api_key=os.getenv(\"OPENAI_API_KEY\")\n",
|
||
")\n",
|
||
"\n",
|
||
"stream = client.chat.completions.create(\n",
|
||
" model=\"qwen3.6-35b-A3b\",\n",
|
||
" messages=[{\"role\": \"user\", \"content\": \"写一首关于春天的短诗。\"}],\n",
|
||
" stream=True # 开启流式输出\n",
|
||
")\n",
|
||
"\n",
|
||
"print(\"🤖: \", end=\"\")\n",
|
||
"for chunk in stream:\n",
|
||
" # 某些 chunk 的 choices 为空，需要先判断\n",
|
||
" if chunk.choices and chunk.choices[0].delta.content is not None:\n",
|
||
" print(chunk.choices[0].delta.content, end=\"\")\n",
|
||
"print()"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"## 8. 本节课练习\n",
|
||
"\n",
|
||
"1. 根据老师提供的 `base_url` 和 `api_key`，在 `.env` 文件中正确配置 OpenAI 和 Anthropic 两种环境变量\n",
|
||
"2. 运行 OpenAI 格式的示例代码，成功获取模型回复\n",
|
||
"3. 阅读 Anthropic 格式的扩展说明，对比两种 API 的差异\n",
|
||
"4. 尝试修改 `temperature` 和 `max_tokens` 参数，观察输出变化\n",
|
||
"5. 使用流式输出模式，让模型生成一段关于人工智能的短文"
|
||
]
|
||
}
|
||
],
|
||
"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
|
||
}
|