前言

最近公司裁员风波，很不幸成为了裁员名单中的一员；此时又恰逢一波AIGC浪潮，首当其冲的就是GPT这样的大语言模型的诞生，是整个AI领域的一个质的飞跃。正好在这样一个空挡期，我就基于Chat-GPT 做了一些深入的实践，并将整个过程记录下来。

准备工作

• 需要在OpenAI官方平台注册一个账号
  首先访问官网需要梯子，不然无法访问；
  账号注册时，最好使用谷歌邮箱，用国内的邮箱注册会返回一些异常的错误；
  注册第二步，需要接受一个短信验证，这里我使用的是sms-activate平台（可以百度一下使用方式），主要就是获取一个临时的国际号码，来获取验证码，我买的印尼🇮🇩的号码，比较便宜，充值1$能用好几次；

  

• 获取API-Key
  注册成功后，点击头像，选择Views API Keys，然后+ Create New Secret Key

  

接入

OpenAI 提供了多种接入方式，包括Python,Node,云API等等

1.Python方式

//安装OpenAI模块
pip install openai

import os
import openai
//组织信息（如果存在多组织）
openai.organization = "org-CfErNk1yqg8HSj5mzWA6AUwA"
//OpenAI-Key
openai.api_key = os.getenv("OPENAI_API_KEY")
//获取训练模型列表
openai.Model.list()
//Completion方式互动
response = openai.Completion.create(
      model="text-davinci-003", 
      prompt="Say this is a test", 
      temperature=0, 
      max_tokens=7
)

2.Node.js 方式

npm install openai

const { Configuration, OpenAIApi } = require("openai");
const configuration = new Configuration({
  apiKey: process.env.OPENAI_API_KEY,
});
const openai = new OpenAIApi(configuration);
//Completion方式互动
const response = await openai.createCompletion({
  model: "text-davinci-003",
  prompt: "Say this is a test",
  temperature: 0,
  max_tokens: 7,
});
//ChatCompletion方式互动
const response1 = await openai.ChatCompletion.create(
  model="gpt-3.5-turbo",
  messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Who won the world series in 2020?"},
        {"role": "assistant", "content": "The Los Angeles Dodgers won the World Series in 2020."},
        {"role": "user", "content": "Where was it played?"}
    ]
)

//ChatCompletion Respones
/*
{
 'id': 'chatcmpl-6p9XYPYSTTRi0xEviKjjilqrWU2Ve',
 'object': 'chat.completion',
 'created': 1677649420,
 'model': 'gpt-3.5-turbo',
 'usage': {'prompt_tokens': 56, 'completion_tokens': 31, 'total_tokens': 87},
 'choices': [
   {
    'message': {
      'role': 'assistant',
      'content': 'The 2020 World Series was played in Arlington, Texas at the Globe Life Field, which was the new home stadium for the Texas Rangers.'},
    'finish_reason': 'stop',
    'index': 0
   }
  ]
}
*/

3.云API

• Completions 方式
  POST https://api.openai.com/v1/completions

header:
{
  "Content-Type: application/json",
  "Authorization: Bearer $OPENAI_API_KEY"
}
body：
{
  "model": "text-davinci-003",    //使用的模型id
  "prompt": "Say this is a test",  //用户输入的内容(提示词)
  "max_tokens": 7,    //最大token数(消费token 来计费)
  "temperature": 0,  //采样精度(0~2) 值越小，精度越高
  "top_p": 1, //概率质量系数， 默认为1
  "n": 1,  //为prompt生成的Completions个数
  "stream": false,  //是否采用流式响应
  "stop": "\n" //生成结果 停止标识符
}
respones:
{
  "id": "cmpl-uqkvlQyYK7bGYrRHQ0eXlWi7",
  "object": "text_completion",
  "created": 1589478378,
  "model": "text-davinci-003",
  "choices": [
    {
      "text": "\n\nThis is indeed a test", //互动返回的结果
      "index": 0,
      "logprobs": null,
      "finish_reason": "length"
    }
  ],
  "usage": {                          // token 消费信息
    "prompt_tokens": 5,
    "completion_tokens": 7,
    "total_tokens": 12
  }
}

• Chat 方式
  POST https://api.openai.com/v1/chat/completions

header:
{
  "Content-Type: application/json",
  "Authorization: Bearer $OPENAI_API_KEY"
}
body：
{
  "model": "gpt-3.5-turbo",
  "messages": [      //消息对象列表,每个对象都包含一个角色(user,system,assistant)和内容字段,支持填充默认/预设的上下文内容
      {"role": "system", "content": "你是一个得力助手，无所不能"},
      {"role": "assistant", "content": "我是得力小助手"},
      {"role": "user", "content": "说你好"}
  ]
}
response：
{
  "id": "chatcmpl-123",
  "object": "chat.completion",
  "created": 1677652288,
  "choices": [{
    "index": 0,
    "message": {
      "role": "assistant",
      "content": "\n\nHello there, how may I assist you today?",
    },
    "finish_reason": "stop"
  }],
  "usage": {
    "prompt_tokens": 9,
    "completion_tokens": 12,
    "total_tokens": 21
  }
}

上面是关于Chat-GPT相关的内容铺垫，接下看微信相关

WeChaty

GitHub地址：https://github.com/wechaty/wechaty
官方介绍：Wechaty is a RPA (Robotic Process Automation) SDK for Chatbot Makers which can help you create a bot in 6 lines of JavaScript, Python, Go, and Java, with cross-platform support including Linux, Windows, MacOS, and Docker.
更多

参考：什么是无头浏览器，它的用途是什么？
总之，使用WeChaty,可以模拟一个微信客户端，用户通过扫码登陆后，它会捕获登陆用户的所以会话，并可以通过提供的API 完成一些自动回复的功能（思考：微信同意这样搞吗？难道没有安全隐患吗？），当前在WeChaty社区也看到了很多被封号的情况，说明在微信的生态下还是不能随便玩的。

在这个基础上，小助理的实现逻辑思路就很清晰了，也非常简单

开始模块组装

核心依赖的3个模块：
"chatgpt": "^4.8.3",    //对openai 的云api 做了上层的封装和管理
"wechaty": "^1.20.2",  //微信机器人模块
"wechaty-puppet-wechat": "^1.18.4" 

初始化Wechaty

async function initProject() {
  try {
    await initChatGPT();
    bot = WechatyBuilder.build({
      name: 'zezefamily',
      puppet: 'wechaty-puppet-wechat',
      puppetOptions: {
        uos: true,
      },
    });

    bot
      .on('scan', onScan)
      .on('login', onLogin)
      .on('logout', onLogout)
      .on('message', onMessage);
    if (config.friendShipRule) {
      bot.on('friendship', onFriendShip);
    }

    bot
      .start()
      .then(() => console.log('Start to log in WeChat...'))
      .catch((e) => console.error(e));
  } catch (error) {
    console.log('init error: ', error);
  }
}

处理bot 的回调

//扫码登陆
function onScan(qrcode) {
  qrcodeTerminal.generate(qrcode); // 在console端显示二维码
  const qrcodeImageUrl = [
    'https://api.qrserver.com/v1/create-qr-code/?data=',
    encodeURIComponent(qrcode),
  ].join('');

  console.log(qrcodeImageUrl);
}
//登陆
async function onLogin(user) {
  console.log(`${user} has logged in`);
  const date = new Date();
  console.log(`Current time:${date}`);
  if (config.autoReply) {
    console.log(`Automatic robot chat mode has been activated`);
  }
}
//登出
function onLogout(user) {
  console.log(`${user} has logged out`);
}
//添加好友
async function onFriendShip(friendship) {
  if (friendship.type() === 2) {
    if (config.friendShipRule.test(friendship.hello())) {
      await friendship.accept();
    }
  }
}

// 收到消息
async function onMessage(msg) {
  // 避免重复发送
  if (msg.date() < startTime) {
    return;
  }
  const contact = msg.talker();    //消息发送者
  const receiver = msg.to();         //消息接收者
  const content = msg.text().trim();  //消息体
  const room = msg.room();        //是否为群消息
  const alias = (await contact.alias()) || (await contact.name());
  const isText = msg.type() === bot.Message.Type.Text;   //判断是否为文本消息
  if (msg.self()) {     //如果是自己给自己发送
    return;
  }

  if (room && isText) {    //只处理文本类消息
    const topic = await room.topic();
    console.log(
      `Group name: ${topic} talker: ${await contact.name()} content: ${content}`
    );
    // console.log("receiver.name() ==>",receiver.name())
    if (await msg.mentionSelf()) {    //群消息，是否@的是自己
      // console.log("是自我提及:",content);
      if (content.startsWith("@小泽玛利亚")){
        const groupContent = content.replace("@小泽玛利亚", '');
        console.log("groupContent ==>",groupContent)
        replyMessage(room, groupContent);  //提取内容发送给Chat-gpt
        return;
      }else {
        console.log(
          'Content is not within the scope of the customizition format'
        );
      }
    }
  } else if (isText) {   //单聊消息
    console.log(`talker: ${alias} content: ${content}`);
    if (config.autoReply) {
      if (content.startsWith(config.privateKey) || config.privateKey === '') {
        let privateContent = content;
        if (config.privateKey === '') {
          privateContent = content.substring(config.privateKey.length).trim();
        }
        replyMessage(contact, privateContent);   //提取内容给Chat-gpt
      } else {
        console.log(
          'Content is not within the scope of the customizition format'
        );
      }
    }
  }
}

Chat-GPT 内部，对外导出了几个function

//chat-gpt 模块初始化
export function initChatGPT() {
  chatGPT = new ChatGPTAPI({
    apiKey: config.OPENAI_API_KEY,
    completionParams: {
      model: 'gpt-3.5-turbo',
    },
  });
}
//contact : talker实例  content：发送的内容
export async function replyMessage(contact, content) {
  const { id: contactId } = contact;
  try {
    if (
      content.trim().toLocaleLowerCase() === config.resetKey.toLocaleLowerCase()
    ) {
      chatOption = {
        ...chatOption,
        [contactId]: {},
      };
      await contact.say('Previous conversation has been reset.');
      return;
    }
    const message = await retryRequest(
      () => getChatGPTReply(content, contactId),
      config.retryTimes,
      500
    );

    if (
      (contact.topic && contact?.topic() && config.groupReplyMode) ||
      (!contact.topic && config.privateReplyMode)
    ) {
      const result = content + '\n-----------\n' + message;
      await contact.say(result);
      return;
    } else {
      await contact.say(message);
    }
  } catch (e: any) {
    console.error(e);
    if (e.message.includes('timed out')) {
      await contact.say(
        content +
          '\n-----------\nERROR: Please try again, ChatGPT timed out for waiting response.'
      );
    }
  }
}
//请求OpenAI接口
async function getChatGPTReply(content, contactId) {
  //调用封装的chatGPT模块发起请求
  const { conversationId, text, id } = await chatGPT.sendMessage(
    content,
    chatOption[contactId]
  );
  chatOption = {
    [contactId]: {
      conversationId,
      parentMessageId: id,
    },
  };
  console.log('response: ', conversationId, text);
  // response is a markdown-formatted string
  return text;
}

核心的内容基本完成，演示一下结果吧
单聊：

群聊：

至此，一个简单Chat-GPT 助理就搞定了。

注意：以上案例仅供学习，不能商业运作；