From c1af34ba0f37dd972557349be7865c4aa15c0b03 Mon Sep 17 00:00:00 2001 From: Ali Aljishi Date: Sun, 4 Aug 2024 21:16:25 +0100 Subject: [PATCH] ability to summarise a certain number of messages in a channel using openai magic --- cogs/commands/summarise.py | 90 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 90 insertions(+) create mode 100644 cogs/commands/summarise.py diff --git a/cogs/commands/summarise.py b/cogs/commands/summarise.py new file mode 100644 index 0000000..594f79e --- /dev/null +++ b/cogs/commands/summarise.py @@ -0,0 +1,90 @@ +from discord.ext import commands +import logging +from typing import Optional + +from discord.ext.commands import Bot, Context, clean_content +from cogs.commands.openaiadmin import is_author_banned_openai +from discord import AllowedMentions + +from config import CONFIG +from discord import AllowedMentions +from utils.utils import get_name_and_content, split_into_messages + +import openai + +LONG_HELP_TEXT = """ +Too much yapping? Summarise what people have said using the power of the GPT overlords! +""" + +SHORT_HELP_TEXT = """Summarise messages.""" + +mentions = AllowedMentions(everyone=False, users=False, roles=False, replied_user=True) +chat_cmd = CONFIG.PREFIX + "chat" +summarise_cmd = CONFIG.PREFIX + "summarise" + + +def clean(msg, *prefixes): + for pre in prefixes: + msg = msg.strip().removeprefix(pre) + return msg.strip() + +class Summarise(commands.Cog): + def __init__(self, bot: Bot): + self.bot = bot + openai.api_key = CONFIG.OPENAI_API_KEY + self.model = "gpt-3.5-turbo" + self.system_prompt = "People yap too much, I don't want to read all of it. In 200 words or less give me the gist of what is being said. Note that the messages are in reverse chronological order:" + + @commands.hybrid_command(help=LONG_HELP_TEXT, brief=SHORT_HELP_TEXT) + async def tldr(self, ctx: Context, number_of_messages: int=100, gpt4: bool=False ): + number_of_messages = 400 if number_of_messages > 400 else number_of_messages + + # avoid banned users + if not await is_author_banned_openai(ctx): + return + + # get the last "number_of_messages" messages from the current channel and build the prompt + curr_channel = ctx.guild.get_channel(ctx.channel.id) + messages = curr_channel.history(limit=number_of_messages) + messages = await self.create_message(messages) + + # send the prompt to the ai overlords to process + async with ctx.typing(): + response = await self.dispatch_api(messages, gpt4) + if response: + prev = ctx.message + for content in split_into_messages(response): + prev = await prev.reply(content, allowed_mentions=mentions) + + + async def dispatch_api(self, messages, gpt4) -> Optional[str]: + logging.info(f"Making OpenAI request: {messages}") + + # Make request + model = "gpt-4" if gpt4 else self.model + response = await openai.ChatCompletion.acreate(model=model, messages=messages) + logging.info(f"OpenAI Response: {response}") + + # Remove prefix that chatgpt might add + reply = response.choices[0].message.content + if CONFIG.AI_INCLUDE_NAMES: + name = f"{self.bot.user.display_name}: " + reply = clean(reply, "Apollo: ", "apollo: ", name) + return reply + + + async def create_message(self, message_chain): + # get initial prompt + initial = self.system_prompt + "\n" + + # for each message, append it to the prompt as follows --- author : message \n + async for msg in message_chain: + if CONFIG.AI_INCLUDE_NAMES and msg.author != self.bot.user: + initial+= msg.author.name + ":" + msg.content + "\n" + + messages = [dict(role="system", content=initial)] + + return messages + +async def setup(bot: Bot): + await bot.add_cog(Summarise(bot))