app.py

import os
import sys
import subprocess
import socket

# Ensure all required packages are installed
def install_packages():
    """
    Install required packages if they are not already installed.
    """
    packages = [
        "gradio==4.44.1", 
        "transformers", 
        "torch", 
        "huggingface_hub"
    ]
    for package in packages:
        try:
            __import__(package.split('==')[0])
        except ImportError:
            print(f"Installing {package}...")
            subprocess.check_call([sys.executable, "-m", "pip", "install", package])

# Install packages before importing
install_packages()

# Now import required libraries
import gradio as gr
import torch
from transformers import pipeline

class HallucinationMitigator:
    def __init__(self, model_name="distilgpt2", penalty_weight=0.5, threshold=0.8):
        """
        Initialize the hallucination mitigation text generator.
        Using distilgpt2 for lighter weight and better performance on CPU
        """
        self.model_name = model_name
        self.penalty_weight = penalty_weight
        self.threshold = threshold
        
        # Explicitly handle device selection
        try:
            # Try to use CUDA if available
            device = 0 if torch.cuda.is_available() else -1
            print(f"Using device: {'CUDA' if device == 0 else 'CPU'}")
        except:
            # Fallback to CPU
            device = -1
            print("Defaulting to CPU")
        
        # Initialize the text generation pipeline
        self.generator = pipeline(
            "text-generation", 
            model=self.model_name, 
            device=device
        )
    
    def detect_hallucination(self, text):
        """
        Simple hallucination detection mechanism.
        """
        try:
            # More robust hallucination detection
            word_count = len(text.split())
            unique_words = len(set(text.split()))
            
            is_potential_hallucination = (
                word_count < 5 or  # Too few words
                text.count('.') < 1 or  # Lacks proper sentence structure
                (unique_words / word_count) < self.threshold  # Low word diversity
            )
            return is_potential_hallucination
        except Exception as e:
            print(f"Hallucination detection error: {e}")
            return False
    
    def generate_text(self, input_text):
        """
        Generate text with hallucination mitigation.
        """
        try:
            # Generate text
            generated = self.generator(
                input_text, 
                max_length=100, 
                num_return_sequences=1,
                do_sample=True,  # Add some randomness
                temperature=0.7  # Control creativity
            )
            
            # Extract generated text
            output = generated[0]['generated_text']
            
            # Check for potential hallucination
            if self.detect_hallucination(output):
                output = f"⚠️ Generated text might contain hallucinations:\n{output}"
            
            return output
        
        except Exception as e:
            return f"Error generating text: {str(e)}"

def create_gradio_interface():
    """
    Create Gradio interface for text generation.
    """
    # Initialize hallucination mitigator
    mitigator = HallucinationMitigator()
    
    # Create Gradio interface with basic theme
    iface = gr.Interface(
        fn=mitigator.generate_text,
        inputs=gr.Textbox(label="Enter your prompt"),
        outputs=gr.Textbox(label="Generated Text"),
        title="OPERA - Hallucination Mitigation",
        description="Text generation with Over-Trust Penalty (OTP) hallucination detection",
        theme="default"
    )
    
    return iface

# Main execution
def main():
    # Create the interface
    interface = create_gradio_interface()
    
    # Launch Gradio app (no manual port or server specified)
    interface.launch(server_name='0.0.0.0')

if __name__ == "__main__":
    main()