app.py

import os
import torch
import logging
from telegram import Update
from telegram.ext import ApplicationBuilder, CommandHandler, MessageHandler, ContextTypes
import telegram.ext.filters as filters
from transformers import AutoModelForCausalLM, AutoTokenizer

# Set up logging
logging.basicConfig(format='%(asctime)s - %(name)s - %(levelname)s - %(message)s', level=logging.INFO)
logger = logging.getLogger(__name__)

# Load the model and tokenizer
model_name = "tanusrich/Mental_Health_Chatbot"
tokenizer = AutoTokenizer.from_pretrained(model_name)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = AutoModelForCausalLM.from_pretrained(model_name).to(device)

# Function to read the token from the token.txt file
def read_token():
    with open('my_support_bot_token.txt', 'r') as file:
        return file.read().strip()  # Removes any leading/trailing whitespace

# Function to format the prompt for the model
def format_prompt(prompt, chat_history):
    history = "".join([f"User: {entry['user']}\nAI: {entry['ai']}\n" for entry in chat_history])
    return f"[INST] <<SYS>> You are a virtual AI therapy assistant. Your role is to provide thoughtful and supportive responses. Always ensure that you complete your last sentence with a period.<</SYS>> {history}User: {prompt.strip()} [/INST]"

# Function to clean the output from the model
def clean_output(output_text, input_text):
    output_text = output_text.replace(input_text, "")
    output_text = output_text.replace("[INST]", "").replace("[/INST]", "").replace("(period)", "").replace("(Period)", "")
    return output_text.strip()

async def start(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
    await update.message.reply_text("Hello! I'm here to listen and support you. How can I help today?")

async def handle_message(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
    user_message = update.message.text
    
    # Prepare prompt for model input
    formatted_prompt = format_prompt(user_message, [])
    
    # Tokenize input and generate response
    inputs = tokenizer(formatted_prompt, return_tensors="pt").to(device)
    
    with torch.no_grad():
        output = model.generate(
            **inputs,
            max_new_tokens=200,
            temperature=0.7,
            top_k=50,
            top_p=0.9,
            repetition_penalty=1.2,
            pad_token_id=tokenizer.eos_token_id
        )
    
    response = tokenizer.decode(output[0], skip_special_tokens=True)
    logger.debug("here is the llm response: %s" % response)
    
    # Clean response and send it back to the user
    clean_response = clean_output(response, formatted_prompt)
    await update.message.reply_text(clean_response)
 

def main():
    token = read_token()
    
    # Create an Application instance
    application = ApplicationBuilder().token(token).build()
    
    # Register command handlers
    application.add_handler(CommandHandler("start", start))
    application.add_handler(MessageHandler(filters.TEXT & ~filters.COMMAND, handle_message))

    # Start the Bot
    application.run_polling()

if __name__ == '__main__':
    main()