利用OpenAI GPT-3写一个Telegram聊天机器人 (Cloudflare Tunnel & GCP AppEngine)

目录展开

最近ChatGPT很红，想说可以试着把OpenAI的API给接上Telegram的群组来玩玩看，顺便记录一下GCP的AppEngine Deploy方法！

代码我放在：https://github.com/stevenyu113228/OpenAI-GPT-3-Telegram-Chatbot

效果

OpenAI

先到 https://beta.openai.com/account/api-keys

申请一组 API Token

在Playground上面随意玩一下，复制他的Example来微调

import os
import openai

openai.api_key = os.getenv("OPENAI_API_KEY")

start_sequence = "\nA:"
restart_sequence = "\n\nQ: "

response = openai.Completion.create(
  model="text-davinci-003",
  prompt="Q: ",
  temperature=0,
  max_tokens=100,
  top_p=1,
  frequency_penalty=0,
  presence_penalty=0,
  stop=["\n"]
)

Telegram API

接下来到Telegram的BotFather来新增Bot

Botfather会响应一组API Token，把它抄下来

写到我们的Config（config.ini）档案里面

[Telegram]
token=xxxxx:xxxxxxxx

[OPENAI]
key=xxxxxxxxxxxxxxxxxxxxx

接下来回到Botfather设定/setjoingroups为Enable，以及/setprivacy为Disable

完整代码

再来增加一点细节，串上Telegram的API

import openai
from flask import Flask, request
import telegram
from telegram.ext import Dispatcher, MessageHandler, Filters
import configparser

config = configparser.ConfigParser()
config.read("config.ini")

bot = telegram.Bot(token=(config['Telegram']['token']))
app = Flask(__name__)


@app.route('/hook', methods=['POST'])
def webhook_handler():
    if request.method == "POST":
        update = telegram.Update.de_json(request.get_json(force=True), bot)
        dispatcher.process_update(update)
    return 'ok'


openai.api_key = config['OPENAI']['key']


def chat_ai(input_str):
    response = openai.Completion.create(
    model="text-davinci-003",
    prompt=f"Human: {input_str} \n AI:",
    temperature=0.9,
    max_tokens=999,
    top_p=1,
    frequency_penalty=0,
    presence_penalty=0.6,
    stop=[" Human:", " AI:"]
    )

    print(response['choices'])
    res = response['choices'][0]['text']
    if res.startswith("？") or res.startswith("?"):
        res = res[1:]
    return res.strip()


def reply_handler(update ,bot):
    """Reply message."""
    try:
        text = update.message.text
        if text.startswith("機器人："):
          text = text[4:]
          print(text)
          res = chat_ai(text)
          print(res)
          update.message.reply_text(res)

    except Exception as e:
        print(e)

dispatcher = Dispatcher(bot, None)
dispatcher.add_handler(MessageHandler(Filters.text, reply_handler))

if __name__ == "__main__":
    # Running server
    app.run()

CloudFlare Tunnel

接下来透过cloudflared开启一个tunnel到127.0.0.1:5000

cloudflared tunnel --url 127.0.0.1:5000

并复制它吐回来的网址，如上图是

https://baptist-freight-criteria-forums.trycloudflare.com

用curl或浏览器戳一次以下网址

https://api.telegram.org/bot{API_TOKEN}/setWebhook?url=https://{CLOUD_FLARE_TUNNEL}/hook

接下来使用python3 main.py跑起来即可

GCP

为了需要一个干净的环境取得requirements.txt

我使用docker

docker run -it --rm python:3.8 bash
pip3 install python-telegram-bot
pip3 install openai
pip3 install flask

pip3 freeze

这样我们就能取得所有相关的lib，建立一个requirements.txt把它存取来

APScheduler==3.6.3
backports.zoneinfo==0.2.1
cachetools==4.2.2
certifi==2022.9.24
charset-normalizer==2.1.1
click==8.1.3
et-xmlfile==1.1.0
Flask==2.2.2
idna==3.4
importlib-metadata==5.1.0
itsdangerous==2.1.2
Jinja2==3.1.2
MarkupSafe==2.1.1
numpy==1.23.5
openai==0.25.0
openpyxl==3.0.10
pandas==1.5.2
pandas-stubs==1.5.2.221124
python-dateutil==2.8.2
python-telegram-bot==13.14
pytz==2022.6
pytz-deprecation-shim==0.1.0.post0
requests==2.28.1
six==1.16.0
tornado==6.1
tqdm==4.64.1
types-pytz==2022.6.0.1
typing_extensions==4.4.0
tzdata==2022.7
tzlocal==4.2
urllib3==1.26.13
Werkzeug==2.2.2
zipp==3.11.0

再来建立一个app.yaml

里面输入

service: openai-chatbot
runtime: python38

GCP AppEngine (PaaS)

开启GCP Cloud Shell的编辑器

把所有档案都贴进来

Cd到文件夹后，输入gcloud init

接下来输入gcloud app deploy

等待它Deploy完毕，再跟前面一样的方法设定Webhook，就能直接玩了ㄛ！

https://api.telegram.org/bot{API_TOKEN}/setWebhook?url=https://{APP Engine URL}/hook