question_qualifer.py

from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
import gradio as gr
import re
import torch

# Use a much smaller, CPU-friendly model
MODEL_NAME = "microsoft/DialoGPT-small"  # Much lighter alternative

print("Loading lightweight model for CPU...")
print("This should load much faster...")

try:
    # Create pipeline directly - simpler approach
    generator = pipeline(
        "text-generation",
        model=MODEL_NAME,
        device=-1,  # Force CPU
        do_sample=True,
        temperature=0.7
    )
    print("✅ Model loaded successfully!")
    
except Exception as e:
    print(f"❌ Error: {e}")
    print("Falling back to rule-based classification...")
    generator = None

def rule_based_classify(question):
    """Simple rule-based classification as fallback"""
    question_lower = question.lower()
    
    # Math patterns
    math_patterns = [
        r'\d+\s*[\+\-\*\/\^]\s*\d+',
        r'what\s+is\s+\d+',
        r'calculate',
        r'solve',
        r'equation',
        r'math',
        r'plus|minus|times|divided|multiply'
    ]
    
    # Opinion patterns  
    opinion_patterns = [
        r'favorite|prefer|best|worst|opinion',
        r'what.*think|feel|believe',
        r'should\s+i|recommend',
        r'better|worse'
    ]
    
    # Check for math
    for pattern in math_patterns:
        if re.search(pattern, question_lower):
            return "math", "This appears to be a math question. Please solve it step by step."
    
    # Check for opinion
    for pattern in opinion_patterns:
        if re.search(pattern, question_lower):
            return "opinion", "This is asking for an opinion or preference. Answers may vary by person."
    
    # Default to factual
    return "factual", "This appears to be asking for factual information."

def classify_and_answer(question):
    """Classify and answer questions with fallback options"""
    if not question or question.strip() == "":
        return "Please enter a valid question."
    
    # If model failed to load, use rule-based approach
    if generator is None:
        category, answer = rule_based_classify(question)
        return f"**Category:** {category.title()}\n**Answer:** {answer}"
    
    # Simple prompt for smaller model
    prompt = f"Question: {question}\nThis question is asking about:"
    
    try:
        # Generate with small model
        output = generator(
            prompt,
            max_length=len(prompt) + 50,
            num_return_sequences=1,
            pad_token_id=generator.tokenizer.eos_token_id
        )
        
        generated_text = output[0]['generated_text']
        response = generated_text[len(prompt):].strip()
        
        # Simple classification based on keywords in response
        response_lower = response.lower()
        if any(word in response_lower for word in ['calculate', 'math', 'number', 'equation']):
            category = "math"
        elif any(word in response_lower for word in ['opinion', 'prefer', 'think', 'feel']):
            category = "opinion"
        else:
            category = "factual"
            
        return f"Category: {category.title()}\nAnswer: {response}"
        
    except Exception as e:
        print(f"Generation error: {e}")
        # Fallback to rule-based
        category, answer = rule_based_classify(question)
        return f"**Category:** {category.title()}\n**Answer:** {answer}"

# Build simple Gradio UI
with gr.Blocks(title="Quick Question Classifier") as iface:
    gr.Markdown("# 🚀 Quick Question Classifier")
    gr.Markdown("*Lightweight CPU-friendly version*")
    
    question_input = gr.Textbox(
        lines=2,
        placeholder="Enter your question here...",
        label="Your Question"
    )
    
    submit_btn = gr.Button("Classify Question", variant="primary")
    
    answer_output = gr.Textbox(
        lines=3,
        label="Classification Result",
        interactive=False
    )
    
    submit_btn.click(classify_and_answer, question_input, answer_output)
    question_input.submit(classify_and_answer, question_input, answer_output)
    
    gr.Examples(
        examples=[
            ["What is the capital of France?"],
            ["What is 25 + 17?"],
            ["What's your favorite color?"],
            ["How many planets are in our solar system?"],
            ["What is 144 divided by 12?"]
        ],
        inputs=question_input
    )

if __name__ == "__main__":
    print("Starting Gradio interface...")
    iface.launch(share=False, server_port=7860)