model.py

# model.py - Custom AI Model untuk Bangdim CS
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
import torch

class BangdimAI:
    def __init__(self):
        print("Loading Bangdim AI Model...")
        
        # Gunakan model dasar yang ringan
        model_name = "microsoft/DialoGPT-medium"  # Bisa ganti dengan model lain
        
        self.tokenizer = AutoTokenizer.from_pretrained(model_name)
        self.model = AutoModelForCausalLM.from_pretrained(model_name)
        
        # Add padding token
        self.tokenizer.pad_token = self.tokenizer.eos_token
        
        print("✅ Model loaded successfully!")
    
    def generate_response(self, user_input, history=[]):
        # Format input dengan history
        prompt = self.format_prompt(user_input, history)
        
        # Encode
        inputs = self.tokenizer.encode(prompt, return_tensors='pt')
        
        # Generate response
        with torch.no_grad():
            outputs = self.model.generate(
                inputs,
                max_length=200,
                temperature=0.8,
                top_p=0.9,
                do_sample=True,
                pad_token_id=self.tokenizer.eos_token_id
            )
        
        # Decode response
        response = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
        
        # Remove prompt from response
        response = response[len(prompt):].strip()
        
        return response if response else "Maaf kak, saya kurang paham. Bisa diulang? 😊"
    
    def format_prompt(self, user_input, history):
        prompt = """Anda adalah Bangdim AI, CS toko top up game yang ramah. Panggil user dengan 'kak'.

"""
        # Add history
        for h in history[-3:]:
            if 'user' in h:
                prompt += f"User: {h['user']}\n"
            if 'bot' in h:
                prompt += f"Assistant: {h['bot']}\n"
        
        prompt += f"User: {user_input}\nAssistant: "
        return prompt

# Initialize model
bangdim_ai = BangdimAI()

# For HuggingFace Spaces
def predict(user_input, history=[]):
    response = bangdim_ai.generate_response(user_input, history)
    return response