For now, i can only provide terminal version. fastapi and web's version are not done yet since i'm quite busy these days.
but the interfaces are the same.
Leave a ⭐ if you like it🥺
Hugging-chat's web api. The current model is OpenAssistant/oasst-sft-6-llama-30b.
This video shows how to actually use it: https://www.bilibili.com/video/BV12h4y1G7Zc
You can also check the Documentation for how to use it
since the model only replies in english even if it's able to process chinese, i decided to add a plugin for it, which is a translator provided by Youdao.
After reverse engineered its web api and reconstructed it using pure python, i managed to finish this.
here's the repo: YouDao-Translater
import requests.sessions
from HuggingChat.Login import Login
email = "your email"
passwd = "password"
sign = Login(email=email, passwd=passwd, mysql=False)
# Login and save cookies
cookies: requests.sessions.RequestsCookieJar = sign.main()
# Load cookies from saved file or database
cookies: requests.sessions.RequestsCookieJar = sign.loadCookies()python main_cmd.py -u <email> -p| Args | |
|---|---|
| -u | Email(optional) |
| -p | Whether to input password(option) |
| --mysql | Whether to connect to mysql(optional) |
| -f | Ignore the saved cookies and login(optional) |
You need to do some configuration in a file named mysqlconf.json before connecting to MySQL
And also, create a database named open-assistant or change the database in the configuration file to the one you use.
{
"sqluser": "",
"sqlpass": "",
"database": "open-assistant",
"host": "127.0.0.1",
"port": 3306
}Format: /+command
| Command | |
|---|---|
| exit/q | exit the program |
| ls | list all conversations |
| cd | cd into the chosen conversation |
| web | activate Search web or not |
| old | check conversation history |
| new | get into new conversation |
| rm | remove the chosen conversation |
| eng | print out last reply's original text |
Example:
(None) > /ls
#* Conversations that have been established:
#
# 0. Assistant: "It's February 24th."
# 1. Today is Wednesday, July 12th, 2034
# 2. "What is today's date?"
# 3. "April 2nd."
#
(None) > /cd 0
(647e09ccabd9de3d82d6fba0) > hi
#(user): hi
#(Open-Assistant): ...
(647e09ccabd9de3d82d6fba0) > /web
#WEB_SEARCH is set to `True`
(647e09ccabd9de3d82d6fba0) > hi
#{'type': 'web_search', 'data': {'type': 'update', 'message': 'Injecting summary', 'args': ['"This is a search result page from iStock, a website that offers stock photography, illu
#strations, and videos. It appears to be related to Memorial Day trending searches, but it\'s not clear how or why that relates to German shepherd puppies. There are some links unde
#r the header \\"Explore\\" which offer curated collections such as signature collection and essentials collection; however these do not seem to have any specific relation with germ
#an shepherd puppies either.\\nThe queries mentioned on this page include Fireworks, Pride Data Timelapse Beach, Aerial views of nature etc . This suggest the user searched at wrong
# timeframe, there might have been other pages available at different point of time containing more accurate results."']}, 'conversation_id': '647f33d14f9cfed1cb9c9b01'}
#{'type': 'web_search', 'data': {'type': 'result', 'id': '647f33e74f9cfed1cb9c9b03'}, 'conversation_id': '647f33d14f9cfed1cb9c9b01'}
#(user): hi
#
#(Open-Assistant): 你好!今天我能为您做些什么?你需要我帮你找到某个主题的信息吗?或者你有问题要问我吗?如果有什么需要我帮忙的，请尽管开口。
Type /new to get into a new conversation, it only supports /exit to close the new conversation before you start it or /web to activate or deactivate Search web
Example of creating a new conversation:
(None) > /new
#Input the first message you want to send (use `/exit` to get back):
#输入创建对话的第一个消息 (使用`/exit`退出新建对话):
(new) (None) > hi
(user): hi
(Open-Assistant): ...
(647e09ccabd9de3d82d6fba0) >
Days before, I noticed that the chat window has added a button to enable web search. Then I decided to reconstruct the code and add a web search interface.
But only a few days passed, however, it gets overloaded due to high demands. It was pretty fast the day I found it.
If you're actually using this, be ready since it may cost a lot of time and even fail to work due to model overloads.
Before this, I was just simply using requests.get(stream=True) to send requests, but it doesn't work this time.
Search web's api is not like chat which is just a EventStream.
It sometimes raises an exception about chunks, I haven't dug deeper into it, what I found online was just something about http1.0 and http1.1, and problems about requests have trouble handling it.
But finally i managed to get away with it. (At least there isn't a single error occurred besides from model overloads)
Every single response will be translated into Chinese.
But for histories, it will get histories instantly but without translation if it's not connected to mysql.
On the other hand, if you do, then it will synchronize and translate histories every 15s and save them into the database.