Spaces:
Sleeping
Sleeping
import discord | |
import logging | |
import os | |
import json | |
from huggingface_hub import InferenceClient | |
import asyncio | |
import subprocess | |
from sentence_transformers import SentenceTransformer, util | |
import torch | |
# λ‘κΉ μ€μ | |
logging.basicConfig(level=logging.DEBUG, format='%(asctime)s:%(levelname)s:%(name)s: %(message)s', handlers=[logging.StreamHandler()]) | |
# μΈν νΈ μ€μ | |
intents = discord.Intents.default() | |
intents.message_content = True | |
intents.messages = True | |
intents.guilds = True | |
intents.guild_messages = True | |
# μΆλ‘ API ν΄λΌμ΄μΈνΈ μ€μ | |
hf_client = InferenceClient("CohereForAI/c4ai-command-r-plus", token=os.getenv("HF_TOKEN")) | |
# νΉμ μ±λ ID | |
SPECIFIC_CHANNEL_ID = int(os.getenv("DISCORD_CHANNEL_ID")) | |
# λν νμ€ν 리λ₯Ό μ μ₯ν μ μ λ³μ | |
conversation_history = [] | |
# JSON λ°μ΄ν°μ λ‘λ | |
with open("jangtest.json", "r", encoding="utf-8") as f: | |
dataset = json.load(f) | |
# λ¬Έμ₯ μλ² λ© λͺ¨λΈ λ‘λ | |
model = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2') | |
# λ°μ΄ν°μ μ μλ² λ©μ 미리 κ³μ° | |
dataset_texts = [json.dumps(item, ensure_ascii=False) for item in dataset] | |
dataset_embeddings = model.encode(dataset_texts, convert_to_tensor=True) | |
class MyClient(discord.Client): | |
def __init__(self, *args, **kwargs): | |
super().__init__(*args, **kwargs) | |
self.is_processing = False | |
async def on_ready(self): | |
logging.info(f'{self.user}λ‘ λ‘κ·ΈμΈλμμ΅λλ€!') | |
subprocess.Popen(["python", "web.py"]) | |
logging.info("Web.py server has been started.") | |
async def on_message(self, message): | |
if message.author == self.user: | |
return | |
if not self.is_message_in_specific_channel(message): | |
return | |
if self.is_processing: | |
return | |
self.is_processing = True | |
try: | |
response = await generate_response(message) | |
await message.channel.send(response) | |
finally: | |
self.is_processing = False | |
def is_message_in_specific_channel(self, message): | |
return message.channel.id == SPECIFIC_CHANNEL_ID or ( | |
isinstance(message.channel, discord.Thread) and message.channel.parent_id == SPECIFIC_CHANNEL_ID | |
) | |
async def generate_response(message): | |
global conversation_history | |
user_input = message.content | |
user_mention = message.author.mention | |
# μ μ¬ν λ°μ΄ν° μ°ΎκΈ° | |
most_similar_data = find_most_similar_data(user_input) | |
system_message = f""" | |
λΉμ μ 'kAI'λΌλ μ΄λ¦μ νκ΅ λ³΄ν μνμ λν AI μ‘°μΈμμ λλ€. | |
λ°λμ μ 곡λ λ°μ΄ν°μ μ μ 보λ§μ μ¬μ©νμ¬ λ΅λ³ν΄μΌ ν©λλ€. | |
μ 곡λ λ°μ΄ν°μ μλ μ 보μ λν΄μλ "μ£μ‘ν©λλ€. ν΄λΉ μ 보λ μ κ° κ°μ§ λ°μ΄ν°μ μμ΅λλ€."λΌκ³ λ΅λ³νμμμ€. | |
λͺ¨λ λ΅λ³μ νκΈλ‘ νκ³ , markdown νμμΌλ‘ μΆλ ₯νμΈμ. | |
""" | |
conversation_history.append({"role": "user", "content": user_input}) | |
messages = [{"role": "system", "content": system_message}] + conversation_history | |
if most_similar_data: | |
messages.append({"role": "system", "content": f"κ΄λ ¨ μ 보: {most_similar_data}"}) | |
else: | |
return f"{user_mention}, μ£μ‘ν©λλ€. κ·νμ μ§λ¬Έκ³Ό κ΄λ ¨λ μ 보λ₯Ό μ°Ύμ μ μμ΅λλ€." | |
logging.debug(f'Messages to be sent to the model: {messages}') | |
loop = asyncio.get_event_loop() | |
response = await loop.run_in_executor(None, lambda: hf_client.chat_completion( | |
messages, max_tokens=1000, stream=True, temperature=0.7, top_p=0.85)) | |
full_response = [] | |
for part in response: | |
logging.debug(f'Part received from stream: {part}') | |
if part.choices and part.choices[0].delta and part.choices[0].delta.content: | |
full_response.append(part.choices[0].delta.content) | |
full_response_text = ''.join(full_response) | |
logging.debug(f'Full model response: {full_response_text}') | |
conversation_history.append({"role": "assistant", "content": full_response_text}) | |
return f"{user_mention}, {full_response_text}" | |
def find_most_similar_data(query): | |
query_embedding = model.encode(query, convert_to_tensor=True) | |
# μ½μ¬μΈ μ μ¬λ κ³μ° | |
cos_scores = util.pytorch_cos_sim(query_embedding, dataset_embeddings)[0] | |
top_result = torch.topk(cos_scores, k=1) | |
if top_result.values[0] > 0.5: # μκ³κ° μ€μ | |
return json.dumps(dataset[top_result.indices[0]], ensure_ascii=False, indent=2) | |
else: | |
return None | |
if __name__ == "__main__": | |
discord_client = MyClient(intents=intents) | |
discord_client.run(os.getenv('DISCORD_TOKEN')) |