feat: Cohere telegra.ph

- Enable Cohere web search
- Output the entire LLM message with its source link on Telegraph
This commit is contained in:
Alter-xyz 2024-06-22 07:29:00 -04:00
parent 90fd74c403
commit dbed5f08b3
4 changed files with 567 additions and 0 deletions

View File

@ -75,6 +75,22 @@ Note, if you are using third party service, you need to `export OPENAI_API_BASE=
Note, currently its support dify Chatbot with instructions(System prompt) and different MODEL with its parameters.
## Bot -> Cohere
1. visit https://dashboard.cohere.com/api-keys get the key
2. export COHERE_API_KEY=${the_key}
3. use `cohere: ${message}` to ask
## Bot -> `Telegra.ph`
1. https://t.me/telegraph Create or login Telegraph account
2. `Log in as ${Account} on this device`
3. On Browser at https://telegra.ph/, press F12 or right click and inspect
4. Go to Application -> Storage -> Cookies -> https://telegra.ph/
5. The token at `tph_token` is the token for telegra.ph API
Do not share the token with others, it's like a password.
## HOW TO Install and Run
### Manually install

View File

@ -200,10 +200,261 @@ def image_to_data_uri(file_path):
return f"data:image/png;base64,{encoded_image}"
import requests
import json
import markdown # pip install Markdown
from bs4 import BeautifulSoup # pip install beautifulsoup4
class TelegraphAPI:
def __init__(self, access_token):
self.access_token = access_token
self.base_url = "https://api.telegra.ph"
# Get account info on initialization
account_info = self.get_account_info()
self.short_name = account_info.get("short_name")
self.author_name = account_info.get("author_name")
self.author_url = account_info.get("author_url")
def create_page(
self, title, content, author_name=None, author_url=None, return_content=False
):
"""
Creates a new Telegraph page.
Args:
title (str): Page title (1-256 characters).
content (list): Content of the page as a list of Node dictionaries.
author_name (str, optional): Author name (0-128 characters). Defaults to account's author_name.
author_url (str, optional): Profile link (0-512 characters). Defaults to account's author_url.
return_content (bool, optional): If True, return the content field in the response.
Returns:
str: URL of the created page.
Raises:
requests.exceptions.RequestException: If the request fails.
"""
url = f"{self.base_url}/createPage"
data = {
"access_token": self.access_token,
"title": title,
"content": json.dumps(content),
"return_content": return_content,
# Use provided author info or fall back to account info
"author_name": author_name if author_name else self.author_name,
"author_url": author_url if author_url else self.author_url,
}
response = requests.post(url, data=data)
response.raise_for_status()
response = response.json()
page_url = response["result"]["url"]
return page_url
def get_account_info(self):
"""
Gets information about the Telegraph account.
Returns:
dict: Account information including short_name, author_name, and author_url.
Returns None if there's an error.
"""
url = f"{self.base_url}/getAccountInfo?access_token={self.access_token}" # &fields=[\"author_name\",\"author_url\"] for specific fields
response = requests.get(url)
if response.status_code == 200:
return response.json()["result"]
else:
print(f"Fail getting telegra.ph token info: {response.status_code}")
return None
def edit_page(
self,
path,
title,
content,
author_name=None,
author_url=None,
return_content=False,
):
"""
Edits an existing Telegraph page.
Args:
path (str): Path of the page to edit.
title (str): New page title (1-256 characters).
content (list): New content of the page as a list of Node dictionaries.
author_name (str, optional): Author name (0-128 characters). Defaults to account's author_name.
author_url (str, optional): Profile link (0-512 characters). Defaults to account's author_url.
return_content (bool, optional): If True, return the content field in the response.
Returns:
str: URL of the edited page.
Raises:
requests.exceptions.RequestException: If the request fails.
"""
url = f"{self.base_url}/editPage"
data = {
"access_token": self.access_token,
"path": path,
"title": title,
"content": json.dumps(content),
"return_content": return_content,
# Use provided author info or fall back to account info
"author_name": author_name if author_name else self.author_name,
"author_url": author_url if author_url else self.author_url,
}
response = requests.post(url, data=data)
response.raise_for_status()
response = response.json()
page_url = response["result"]["url"]
return page_url
def get_page(self, path):
"""
Gets information about a Telegraph page.
Args:
path (str): Path of the page to get.
Returns:
dict: Information about the page.
"""
url = f"{self.base_url}/getPage/{path}?return_content=true"
response = requests.get(url)
response.raise_for_status()
return response.json()["result"]
def create_page_md(
self,
title,
markdown_text,
author_name=None,
author_url=None,
return_content=False,
):
"""
Creates a new Telegraph page from markdown text.
Args:
title (str): Page title (1-256 characters).
markdown_text (str): Markdown text to convert to HTML.
author_name (str, optional): Author name (0-128 characters). Defaults to account's author_name.
author_url (str, optional): Profile link (0-512 characters). Defaults to account's author_url.
return_content (bool, optional): If True, return the content field in the response.
Returns:
str: URL of the created page.
Raises:
requests.exceptions.RequestException: If the request fails.
"""
content = md_to_dom(markdown_text)
return self.create_page(title, content, author_name, author_url, return_content)
def edit_page_md(
self,
path,
title,
markdown_text,
author_name=None,
author_url=None,
return_content=False,
):
content = md_to_dom(markdown_text)
return self.edit_page(
path, title, content, author_name, author_url, return_content
)
def md_to_dom(markdown_text):
"""Converts markdown text to a Python dictionary representing the DOM,
limiting heading levels to h3 and h4.
Args:
markdown_text: The markdown text to convert.
Returns:
A Python list representing the DOM, where each element is a dictionary
with the following keys:
- 'tag': The tag name of the element.
- 'attributes': A dictionary of attributes for the element (optional).
- 'children': A list of child elements (optional).
"""
# Convert markdown to HTML
html = markdown.markdown(
markdown_text,
extensions=["markdown.extensions.extra", "markdown.extensions.sane_lists"],
)
# Parse the HTML with BeautifulSoup
soup = BeautifulSoup(html, "html.parser")
def parse_element(element):
tag_dict = {"tag": element.name}
if element.name in ["h1", "h2", "h3", "h4", "h5", "h6"]:
if element.name == "h1":
tag_dict["tag"] = "h3"
elif element.name == "h2":
tag_dict["tag"] = "h4"
else:
tag_dict["tag"] = "p"
tag_dict["children"] = [{"tag": "strong", "children": element.contents}]
# Correctly handle children for h1-h6
if element.attrs:
tag_dict["attributes"] = element.attrs
if element.contents:
children = []
for child in element.contents:
if isinstance(child, str):
# Remove leading/trailing whitespace from text nodes
children.append(child.strip())
else: # it's another tag
children.append(parse_element(child))
tag_dict["children"] = children
else:
if element.attrs:
tag_dict["attributes"] = element.attrs
if element.contents:
children = []
for child in element.contents:
if isinstance(child, str):
# Remove leading/trailing whitespace from text nodes
children.append(child.strip())
else: # it's another tag
children.append(parse_element(child))
if children:
tag_dict["children"] = children
return tag_dict
new_dom = []
for element in soup.contents:
if isinstance(element, str) and not element.strip():
# Skip empty text nodes
continue
elif isinstance(element, str):
# Treat remaining text nodes as separate elements for clarity
new_dom.append({"tag": "text", "content": element.strip()})
else:
new_dom.append(parse_element(element))
return new_dom
# `import *` will give you these
__all__ = [
"bot_reply_first",
"bot_reply_markdown",
"enrich_text_with_urls",
"image_to_data_uri",
"TelegraphAPI",
]

211
handlers/cohere.py Normal file
View File

@ -0,0 +1,211 @@
from os import environ
import time
from telebot import TeleBot
from telebot.types import Message
from expiringdict import ExpiringDict
from . import *
import cohere
from telegramify_markdown import convert
from telegramify_markdown.customize import markdown_symbol
markdown_symbol.head_level_1 = "📌" # If you want, Customizing the head level 1 symbol
markdown_symbol.link = "🔗" # If you want, Customizing the link symbol
COHERE_API_KEY = environ.get("COHERE_API_KEY")
COHERE_MODEL = "command-r-plus"
TELEGRA_PH_TOKEN = environ.get("TELEGRA_PH_TOKEN")
if TELEGRA_PH_TOKEN:
ph = TelegraphAPI(TELEGRA_PH_TOKEN)
if COHERE_API_KEY:
co = cohere.Client(api_key=COHERE_API_KEY)
# Global history cache
cohere_player_dict = ExpiringDict(max_len=1000, max_age_seconds=300)
def cohere_handler_direct(message: Message, bot: TeleBot) -> None:
"""cohere : /cohere <question>"""
m = message.text.strip()
player_message = []
if str(message.from_user.id) not in cohere_player_dict:
cohere_player_dict[str(message.from_user.id)] = player_message
else:
player_message = cohere_player_dict[str(message.from_user.id)]
if m.strip() == "clear":
bot.reply_to(
message,
"Just cleared your Cohere messages history",
)
player_message.clear()
return
if m[:4].lower() == "new ":
m = m[4:].strip()
player_message.clear()
m = enrich_text_with_urls(m)
who = "Command R Plus"
reply_id = bot_reply_first(message, who, bot)
player_message.append({"role": "User", "message": m})
# keep the last 5, every has two ask and answer.
if len(player_message) > 10:
player_message = player_message[2:]
try:
stream = co.chat_stream(
model=COHERE_MODEL,
message=m,
temperature=0.8,
chat_history=player_message,
prompt_truncation="AUTO",
connectors=[{"id": "web-search"}],
citation_quality="accurate",
)
s = ""
source = ""
start = time.time()
for event in stream:
if event.event_type == "text-generation":
s += event.text.encode("utf-8").decode("utf-8")
if time.time() - start > 1.2:
start = time.time()
bot_reply_markdown(reply_id, who, s, bot, split_text=True)
elif event.event_type == "search-results":
for doc in event.documents:
source += f"\n[{doc['title']}]({doc['url']})"
elif event.event_type == "stream-end":
break
s += "\n" + source + "\n"
if not bot_reply_markdown(reply_id, who, s, bot):
# maybe not complete
# maybe the same message
player_message.clear()
return
player_message.append(
{
"role": "Chatbot",
"message": convert(s),
}
)
except Exception as e:
print(e)
bot_reply_markdown(reply_id, who, "Answer wrong", bot)
player_message.clear()
return
def cohere_handler(message: Message, bot: TeleBot) -> None:
"""cohere : /cohere <question>"""
m = message.text.strip()
player_message = []
if str(message.from_user.id) not in cohere_player_dict:
cohere_player_dict[str(message.from_user.id)] = player_message
else:
player_message = cohere_player_dict[str(message.from_user.id)]
if m.strip() == "clear":
bot.reply_to(
message,
"Just cleared your Cohere messages history",
)
player_message.clear()
return
if m[:4].lower() == "new ":
m = m[4:].strip()
player_message.clear()
m = enrich_text_with_urls(m)
who = "Command R Plus"
reply_id = bot_reply_first(message, who, bot)
player_message.append({"role": "User", "message": m})
# keep the last 5, every has two ask and answer.
if len(player_message) > 10:
player_message = player_message[2:]
try:
stream = co.chat_stream(
model=COHERE_MODEL,
message=m,
temperature=0.8,
chat_history=player_message,
prompt_truncation="AUTO",
connectors=[{"id": "web-search"}],
citation_quality="accurate",
)
s = ""
source = ""
start = time.time()
for event in stream:
if event.event_type == "text-generation":
s += event.text.encode("utf-8").decode("utf-8")
if time.time() - start > 1.2:
start = time.time()
bot_reply_markdown(reply_id, who, s, bot, split_text=False)
elif event.event_type == "search-results":
for doc in event.documents:
source += f"\n{doc['title']}\n{doc['url']}\n"
elif event.event_type == "stream-end":
break
content = s + "\n------\n------\n" + source
ph_s = ph.create_page_md(title="Cohere", markdown_text=content)
s += f"\n\n[View]({ph_s})"
if not bot_reply_markdown(reply_id, who, s, bot):
# maybe not complete
# maybe the same message
player_message.clear()
return
player_message.append(
{
"role": "Chatbot",
"message": convert(s),
}
)
except Exception as e:
print(e)
bot_reply_markdown(reply_id, who, "Answer wrong", bot)
player_message.clear()
return
if COHERE_API_KEY:
if not TELEGRA_PH_TOKEN:
def register(bot: TeleBot) -> None:
bot.register_message_handler(
cohere_handler_direct, commands=["cohere"], pass_bot=True
)
bot.register_message_handler(
cohere_handler_direct, regexp="^cohere:", pass_bot=True
)
else:
def register(bot: TeleBot) -> None:
bot.register_message_handler(
cohere_handler_direct, commands=["cohere"], pass_bot=True
)
bot.register_message_handler(
cohere_handler_direct, regexp="^cohere:", pass_bot=True
)

View File

@ -18,6 +18,14 @@ from . import *
from telegramify_markdown.customize import markdown_symbol
import cohere
COHERE_API_KEY = environ.get("COHERE_API_KEY")
TELEGRA_PH_TOKEN = environ.get("TELEGRA_PH_TOKEN")
co = cohere.Client(api_key=COHERE_API_KEY)
ph = TelegraphAPI(TELEGRA_PH_TOKEN)
COHERE_MODEL = "command-r-plus"
chat_message_dict = ExpiringDict(max_len=100, max_age_seconds=120)
chat_user_dict = ExpiringDict(max_len=100, max_age_seconds=20)
@ -91,6 +99,7 @@ def latest_handle_messages(message: Message, bot: TeleBot):
"sd",
"map",
"yi",
"cohere",
)
):
return
@ -120,6 +129,7 @@ def answer_it_handler(message: Message, bot: TeleBot):
latest_message = chat_message_dict.get(chat_id)
m = latest_message.text.strip()
m = enrich_text_with_urls(m)
full = ""
##### Gemini #####
who = "Gemini Pro"
# show something, make it more responsible
@ -141,6 +151,8 @@ def answer_it_handler(message: Message, bot: TeleBot):
convo.history.clear()
bot_reply_markdown(reply_id, who, "Error", bot)
full += f"{who}:\n{s}"
chat_id_list = [reply_id.message_id]
##### ChatGPT #####
who = "ChatGPT Pro"
reply_id = bot_reply_first(latest_message, who, bot)
@ -173,6 +185,83 @@ def answer_it_handler(message: Message, bot: TeleBot):
print(e)
bot_reply_markdown(reply_id, who, "answer wrong", bot)
full += f"\n---\n{who}:\n{s}"
chat_id_list.append(reply_id.message_id)
##### Cohere #####
if COHERE_API_KEY:
full, chat_id = cohere_answer(latest_message, bot, full, m)
chat_id_list.append(chat_id)
else:
pass
##### Answer #####
if TELEGRA_PH_TOKEN:
final_answer(latest_message, bot, full, chat_id_list)
else:
pass
def cohere_answer(latest_message: Message, bot: TeleBot, full, m):
"""cohere answer"""
who = "Command R Plus"
reply_id = bot_reply_first(latest_message, who, bot)
player_message = [{"role": "User", "message": m}]
try:
r = co.chat_stream(
model=COHERE_MODEL,
message=m,
temperature=0.4,
chat_history=player_message,
prompt_truncation="AUTO",
connectors=[{"id": "web-search"}],
citation_quality="fast",
)
s = ""
source = ""
start = time.time()
for event in r:
if event.event_type == "text-generation":
s += event.text.encode("utf-8").decode("utf-8")
if time.time() - start > 1.2:
start = time.time()
bot_reply_markdown(reply_id, who, s, bot, split_text=False)
elif event.event_type == "search-results":
for doc in event.documents:
source += f"\n[{doc['title']}]({doc['url']})"
elif event.event_type == "stream-end":
break
# maybe not complete
# maybe the same message
try:
bot_reply_markdown(reply_id, who, s, bot)
except:
pass
except Exception as e:
print(e)
bot_reply_markdown(reply_id, who, "Answer wrong", bot)
content = s + "\n------\n" + source
full += f"\n---\n{who}:\n{content}"
chat_id = reply_id.chat.id
return full, chat_id
def final_answer(latest_message: Message, bot: TeleBot, full, list):
"""final answer"""
who = "Answer"
reply_id = bot_reply_first(latest_message, who, bot)
ph_s = ph.create_page_md(title="Answer it", markdown_text=full)
bot_reply_markdown(reply_id, who, f"[View]({ph_s})", bot)
# delete the chat message, only leave a telegra.ph link
# for i in list:
# bot.delete_message(chat_id=chat_id, message_id=i)
if GOOGLE_GEMINI_KEY and CHATGPT_API_KEY: