ChatGPT_WeChat

Introduction:

Uncertified WeChat public account is connected to chatgpt, new voice chat (English conversation) is added, based on Flask, personal WeChat public account [without authentication] is connected to ChatGPT

--Update instructions:

V1.1.0: (2023.04.13)

-Added streaming response (stream) to alleviate the request timeout problem to a certain extent. You need to install the python package: sseclient-py==1.7.2;

After the streaming response is enabled, a connection (myrequest) will be established first, and then (SSEClient) will be used to obtain the generated text character by character, and finally the obtained text list will be spliced ​​into the reply text. The time to establish a connection will still be affected by max_tokens, so it is not recommended to set max_tokens too large. The key to alleviating request timeouts is that the time consumption of establishing a connection is less than the time consumption of one-time return. Therefore, as long as the connection is successfully established within a given time, the content can basically be returned. The length of the returned content will be affected by the connection time. .

-Delete the most recent messages sent by the user after the request fails or times out to avoid errors in the next reply.

V1.0.1:

- Added option to enable IP detection (to prevent doss attacks);

-Note: If any of wechat-ip_detection and azure-trans_to_voice is true, both appid and secret need to be filled in.

V1.0:

-New voice chat function, access to Microsoft cloud text-to-speech service (free access), to achieve voice conversations (Chinese and English);

-Built-in English learning templates, you can have English-themed conversations by replying to template messages;

- Added WeChat background whitelist IP detection to prevent doss attacks, etc.;

- Added user message frequency limit to prevent malicious message spamming;

- Automatically clean up temporary voice files;

- Automatically clean up temporary voice materials uploaded in the WeChat background;

-Optimize performance and fix bugs.

background:

I recently saw that ChatGPT provides an API interface. I happen to have a server and a public account, so I want to write a chatbot? Give it a try. However, there is only one uncertified personal official account (resources are limited?). The limitations of this official account are:

1. It can only passively reply to user messages. The user sends a message to the official account. The server can only reply to one message for this request and cannot reply to additional messages (customer service messages);

2. Each message must be replied within 15 seconds. After the official account platform sends a request to the server, if it does not receive a reply within 5 seconds, it will send another request and wait for 5 seconds. If the request is still not received, it will send a final request. Therefore, the server must process the complete message within 15 seconds.

See the code for specific processing methods. This is a newbie project, please include any shortcomings and welcome corrections, thank you~

need:

A server (needs to be able to access the openai interface, may need to be overseas~)

If you need to enable the text-to-speech service, you need to register Azure's text-to-speech service. Registration and use of this service are free. Please refer to the website for details: AZURE

WeChat public account: Personal type is enough

Demo:

Public account: Tory's laboratory, follow and send messages to experience it.

Introduction to public account tweets:

1. Introduction to getting started: ChatGPT has been connected

2. Introduction to the use of voice service: The voice service has been connected

How to use:

Set the config.yml parameters in config:

 # 微信相关设置
wechat :
  token : " **** "
  # 是否获取微信公众平台的ip白名单(用于防止doss检测)
  ip_detection : false
  # 如果上面的选项为true，下面两项内容必填；如要开启后面文本转语音服务，下面两项内容必填
  appid : " **** "
  secret : " **** "

# openai相关设置
openai :
  #填写openai的api_keys时，要注意前面要加上：Bearer, 可以填写多个,因为单个账号有速率的限制
  api_keys :
   - " Bearer sk-**** "
   # - "Bearer sk-****"
   # - "Bearer sk-****"
  # 单条消息的长度，这个参数对回复速度有非常大的影响，请不要填太大~
  max_tokens : 120
  # 模型
  model : " gpt-3.5-turbo-0301 "
  # temperature，越大随机性越强
  temperature : 0.8
  # 有时候文本长度超过150，用该参数限制长度避免超过微信能发送的最长消息
  rsize : 500
  # 对话的保存历史
  save_history : 21
  
# azure文本转语音设置
azure :
  # 是否开启文本转语音服务
  trans_to_voice : false
  # 如上面的选项为false，下面的内容不用填写
  # 新定义文本长度，开启后增加处理时间，避免文本太长，处理时间过久，超过15s
  max_token : 80
  # 是否开启流式响应
  stream_response : true
  # 密钥
  subscription : " **** "
  region : " koreacentral "
  # 中文语音模型
  zh_model : " zh-CN-XiaoyanNeural "
  # 英文语音模型
  en_model : " en-US-AriaNeural "

Start flask

 export FLASK_APP=myflask

flask run --host=0.0.0.0 --port=80
# 或者
nohup flask run --host=0.0.0.0 --port=80 >> /home/jupyter/flask/log/wechat.log 2>&1 &

Notice:

1. When filling in openai’s api_keys, be sure to add: Bearer in front. You can fill in multiple api_keys because a single account has a rate limit;

2.max_tokens has a great impact on the reply speed, please do not fill it in too large.