app.py

from flask import Flask, jsonify, request, render_template
from flask_cors import CORS
from transformers import GPT2Tokenizer, GPT2LMHeadModel
import torch

app = Flask(__name__)
CORS(app)

# Global variables for model and tokenizer
MODEL_PATH = "./models/fine-tuned-gpt2"
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
tokenizer = None
model = None

def load_chatbot_model():
    """Load the chatbot model and tokenizer"""
    global tokenizer, model
    if model is None:
        print(f"Loading chatbot model from {MODEL_PATH}...")
        print(f"Using device: {device}")
        
        tokenizer = GPT2Tokenizer.from_pretrained(MODEL_PATH)
        model = GPT2LMHeadModel.from_pretrained(MODEL_PATH)
        model.to(device)
        
        print("Model loaded successfully!")

# Load model on startup
load_chatbot_model()

@app.route("/")
def index():
    """Serve the chat interface"""
    return render_template('index.html')

@app.route("/api")
def root():
    return jsonify({
        "message": "Chatbot API",
        "status": "running",
        "model": "fine-tuned-gpt2",
        "device": str(device)
    })

@app.route("/health")
def health():
    return jsonify({
        "status": "healthy",
        "model_loaded": model is not None,
        "device": str(device)
    })

@app.route("/chat", methods=["POST"])
def chat():
    """
    Generate a chatbot response based on conversation history
    """
    if model is None or tokenizer is None:
        return jsonify({"error": "Model not loaded"}), 500
    
    try:
        data = request.get_json()
        user_messages = data.get("user", [])
        ai_messages = data.get("ai", [])
        
        # Build conversation history
        combined_prompt = ""
        
        # Limit history to last 7 exchanges
        user_msgs = user_messages[-7:] if len(user_messages) > 7 else user_messages
        ai_msgs = ai_messages[-6:] if len(ai_messages) > 6 else ai_messages
        
        # Add conversation history
        for user_message, ai_message in zip(user_msgs[:-1], ai_msgs):
            combined_prompt += f"<user> {user_message}{tokenizer.eos_token}<AI> {ai_message}{tokenizer.eos_token}"
        
        # Add current message
        if user_msgs:
            combined_prompt += f"<user> {user_msgs[-1]}{tokenizer.eos_token}<AI>"
        
        # Tokenize and generate
        inputs = tokenizer.encode(combined_prompt, return_tensors="pt").to(device)
        attention_mask = torch.ones(inputs.shape, device=device)
        
        outputs = model.generate(
            inputs,
            max_new_tokens=50,
            num_beams=5,
            early_stopping=True,
            no_repeat_ngram_size=2,
            temperature=0.7,
            top_k=50,
            top_p=0.95,
            pad_token_id=tokenizer.eos_token_id,
            attention_mask=attention_mask,
            repetition_penalty=1.2
        )
        
        response = tokenizer.decode(outputs[0], skip_special_tokens=True)
        
        # Extract only the new AI response
        # Split by <AI> and get the last response
        if "<AI>" in response:
            response = response.split("<AI>")[-1].strip()
        
        # Remove any <user> tags if they appear (model might generate them)
        if "<user>" in response:
            response = response.split("<user>")[0].strip()
        
        # Clean up any remaining special tokens
        response = response.replace("<AI>", "").replace("<user>", "").strip()
        
        # If empty response, provide a default
        if not response:
            response = "I'm not sure how to respond to that."
        
        return jsonify({
            "response": response,
            "device": str(device)
        })
    
    except Exception as e:
        return jsonify({"error": str(e)}), 500

if __name__ == "__main__":
    app.run(host="0.0.0.0", port=7860, debug=False)