Fix: Updated app.py with better error handling and compatibility

app.py

@@ -6,50 +6,62 @@ from peft import PeftModel
 import re
 
 # Model configuration
-BASE_MODEL = "deepseek-ai/deepseek-math-7b-instruct"
+BASE_MODEL = "deepseek-ai/deepseek-math-7b-instruct" 
 REPO_ID = "danxh/math-mcq-generator-v1"
 
-# Load model and tokenizer
-@torch.no_grad()
+# Global variables for model and tokenizer
+model = None
+tokenizer = None
+
 def load_model():
-    """Load the fine-tuned model"""
-    
-    bnb_config = BitsAndBytesConfig(
-        load_in_4bit=True,
-        bnb_4bit_use_double_quant=True,
-        bnb_4bit_quant_type="nf4",
-        bnb_4bit_compute_dtype="bfloat16"
-    )
-    
-    # Load base model
-    base_model = AutoModelForCausalLM.from_pretrained(
-        BASE_MODEL,
-        quantization_config=bnb_config,
-        device_map="auto",
-        torch_dtype=torch.bfloat16
-    )
+    """Load the fine-tuned model with error handling"""
+    global model, tokenizer
     
-    # Load LoRA adapter
-    model = PeftModel.from_pretrained(base_model, REPO_ID)
-    
-    # Load tokenizer
-    tokenizer = AutoTokenizer.from_pretrained(REPO_ID)
-    if tokenizer.pad_token is None:
-        tokenizer.pad_token = tokenizer.eos_token
-    
-    return model, tokenizer
-
-# Initialize model
-print("🔄 Loading model...")
-model, tokenizer = load_model()
-print("✅ Model loaded successfully!")
+    try:
+        print("🔄 Loading model and tokenizer...")
+        
+        # Simplified loading for Hugging Face Spaces
+        bnb_config = BitsAndBytesConfig(
+            load_in_4bit=True,
+            bnb_4bit_use_double_quant=True,
+            bnb_4bit_quant_type="nf4",
+            bnb_4bit_compute_dtype=torch.float16  # Changed to float16 for better compatibility
+        )
+        
+        # Load base model
+        base_model = AutoModelForCausalLM.from_pretrained(
+            BASE_MODEL,
+            quantization_config=bnb_config,
+            device_map="auto",
+            torch_dtype=torch.float16,
+            trust_remote_code=True
+        )
+        
+        # Load LoRA adapter
+        model = PeftModel.from_pretrained(base_model, REPO_ID)
+        
+        # Load tokenizer
+        tokenizer = AutoTokenizer.from_pretrained(REPO_ID)
+        if tokenizer.pad_token is None:
+            tokenizer.pad_token = tokenizer.eos_token
+        
+        print("✅ Model loaded successfully!")
+        return True
+        
+    except Exception as e:
+        print(f"❌ Error loading model: {str(e)}")
+        return False
 
 def generate_mcq(chapter, topics, difficulty="medium", cognitive_skill="direct_application"):
     """Generate MCQ using the fine-tuned model"""
     
-    input_text = f"chapter: {chapter}\ntopics: {topics}\nDifficulty: {difficulty}\nCognitive Skill: {cognitive_skill}"
+    if model is None or tokenizer is None:
+        return "❌ Model not loaded. Please wait for initialization."
     
-    prompt = f"""### Instruction:
+    try:
+        input_text = f"chapter: {chapter}\ntopics: {topics}\nDifficulty: {difficulty}\nCognitive Skill: {cognitive_skill}"
+        
+        prompt = f"""### Instruction:
 Generate a math MCQ similar in style to the provided examples.
 
 ### Input:
@@ -57,26 +69,29 @@ Generate a math MCQ similar in style to the provided examples.
 
 ### Response:
 """
-    
-    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=512)
-    inputs = {k: v.to(model.device) for k, v in inputs.items()}
-    
-    with torch.no_grad():
-        outputs = model.generate(
-            **inputs,
-            max_new_tokens=300,
-            temperature=0.7,
-            do_sample=True,
-            pad_token_id=tokenizer.eos_token_id,
-            eos_token_id=tokenizer.eos_token_id,
-            repetition_penalty=1.1
-        )
-    
-    generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
-    response_start = generated_text.find("### Response:") + len("### Response:")
-    response = generated_text[response_start:].strip()
-    
-    return response
+        
+        inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=512)
+        inputs = {k: v.to(model.device) for k, v in inputs.items()}
+        
+        with torch.no_grad():
+            outputs = model.generate(
+                **inputs,
+                max_new_tokens=300,
+                temperature=0.7,
+                do_sample=True,
+                pad_token_id=tokenizer.eos_token_id,
+                eos_token_id=tokenizer.eos_token_id,
+                repetition_penalty=1.1
+            )
+        
+        generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
+        response_start = generated_text.find("### Response:") + len("### Response:")
+        response = generated_text[response_start:].strip()
+        
+        return response
+        
+    except Exception as e:
+        return f"❌ Error generating MCQ: {str(e)}"
 
 def parse_mcq_response(response):
     """Parse the model response"""
@@ -112,13 +127,159 @@ def parse_mcq_response(response):
             "error": str(e)
         }
 
-# [Include the web interface function here - copy from above]
 def generate_mcq_web(chapter, topics_text, difficulty, cognitive_skill, num_questions=1):
-    # [Copy the function implementation from above]
-    pass
+    """Web interface wrapper for MCQ generation"""
+    
+    if model is None or tokenizer is None:
+        return """
+        <div style="border: 2px solid #ffc107; border-radius: 10px; padding: 20px; margin: 10px 0; background: #fff3cd;">
+            <h3 style="color: #856404;">⏳ Model Loading</h3>
+            <p>The model is still loading. Please wait a moment and try again.</p>
+        </div>
+        """
+    
+    try:
+        # Parse topics
+        topics_list = [t.strip() for t in topics_text.split(',') if t.strip()]
+        if not topics_list:
+            topics_list = ["General"]
+        
+        results = []
+        
+        for i in range(min(num_questions, 3)):  # Limit to 3 questions max
+            # Generate MCQ
+            raw_response = generate_mcq(chapter, topics_list, difficulty, cognitive_skill)
+            parsed = parse_mcq_response(raw_response)
+            
+            if "error" not in parsed:
+                # Format for web display
+                question_html = f"""
+                <div style="border: 2px solid #e1e5e9; border-radius: 10px; padding: 20px; margin: 10px 0; background: #f8f9fa;">
+                    <h3 style="color: #2c3e50; margin-top: 0;">📚 Question {i+1}</h3>
+                    <p style="font-size: 16px; line-height: 1.6; margin: 15px 0;"><strong>{parsed['question']}</strong></p>
+                    
+                    <div style="margin: 15px 0;">
+                        <h4 style="color: #34495e;">Options:</h4>
+                        <ul style="list-style: none; padding: 0;">
+                            <li style="margin: 8px 0; padding: 8px; background: #ecf0f1; border-radius: 5px;">
+                                <strong>(A)</strong> {parsed['options'][0] if len(parsed['options']) > 0 else 'N/A'}
+                            </li>
+                            <li style="margin: 8px 0; padding: 8px; background: #ecf0f1; border-radius: 5px;">
+                                <strong>(B)</strong> {parsed['options'][1] if len(parsed['options']) > 1 else 'N/A'}
+                            </li>
+                            <li style="margin: 8px 0; padding: 8px; background: #ecf0f1; border-radius: 5px;">
+                                <strong>(C)</strong> {parsed['options'][2] if len(parsed['options']) > 2 else 'N/A'}
+                            </li>
+                            <li style="margin: 8px 0; padding: 8px; background: #ecf0f1; border-radius: 5px;">
+                                <strong>(D)</strong> {parsed['options'][3] if len(parsed['options']) > 3 else 'N/A'}
+                            </li>
+                        </ul>
+                    </div>
+                    
+                    <div style="margin-top: 15px; padding: 10px; background: #d5edda; border-radius: 5px; border-left: 4px solid #28a745;">
+                        <strong>✅ Correct Answer: {parsed['correct_answer']}</strong>
+                    </div>
+                </div>
+                """
+                results.append(question_html)
+            else:
+                error_html = f"""
+                <div style="border: 2px solid #dc3545; border-radius: 10px; padding: 20px; margin: 10px 0; background: #f8d7da;">
+                    <h3 style="color: #721c24;">❌ Error generating question {i+1}</h3>
+                    <p>{parsed.get('error', 'Unknown error occurred')}</p>
+                </div>
+                """
+                results.append(error_html)
+        
+        return "".join(results)
+        
+    except Exception as e:
+        return f"""
+        <div style="border: 2px solid #dc3545; border-radius: 10px; padding: 20px; margin: 10px 0; background: #f8d7da;">
+            <h3 style="color: #721c24;">❌ System Error</h3>
+            <p>Error: {str(e)}</p>
+        </div>
+        """
+
+# Create the interface
+interface = gr.Interface(
+    fn=generate_mcq_web,
+    inputs=[
+        gr.Textbox(
+            label="📚 Chapter",
+            placeholder="e.g., Applications of Trigonometry, Conic Sections",
+            value="Applications of Trigonometry",
+            info="Enter the mathematics chapter or topic area"
+        ),
+        gr.Textbox(
+            label="📝 Topics (comma-separated)",
+            placeholder="e.g., Heights and Distances, Circle, Tangents",
+            value="Heights and Distances",
+            info="Enter specific topics within the chapter, separated by commas"
+        ),
+        gr.Dropdown(
+            choices=["easy", "medium", "hard"],
+            label="⚡ Difficulty Level",
+            value="medium",
+            info="Select the difficulty level for the questions"
+        ),
+        gr.Dropdown(
+            choices=["recall", "direct_application", "pattern_recognition", "strategic_reasoning", "trap_aware"],
+            label="🧠 Cognitive Skill",
+            value="direct_application",
+            info="Select the type of thinking skill required"
+        ),
+        gr.Slider(
+            minimum=1,
+            maximum=3,
+            step=1,
+            label="🔢 Number of Questions",
+            value=1,
+            info="How many questions to generate (max 3)"
+        )
+    ],
+    outputs=gr.HTML(label="Generated MCQ(s)"),
+    
+    title="🧮 Mathematics MCQ Generator",
+    description="""
+    Generate high-quality mathematics multiple choice questions using AI. 
+    This model has been fine-tuned specifically for educational content creation.
+    
+    **Note**: Model loading may take a few minutes on first startup.
+    """,
+    
+    article="""
+    ### 🔬 About This Model
+    
+    This MCQ generator is powered by a fine-tuned version of DeepSeek-Math-7B, specifically adapted for mathematics education.
+    
+    ### 💡 Tips for Best Results:
+    - Be specific with chapter and topic names
+    - Try different cognitive skill levels for variety
+    - Start with 1 question to test, then generate more
+    
+    ### 🤝 Collaboration
+    This is part of a collaborative project to create specialized educational AI tools.
+    """,
+    
+    theme=gr.themes.Soft(primary_hue="blue", secondary_hue="purple"),
+    
+    examples=[
+        ["Applications of Trigonometry", "Heights and Distances", "easy", "recall", 1],
+        ["Conic Sections", "Circle", "medium", "pattern_recognition", 1],
+        ["Applications of Trigonometry", "Angle of Elevation and Depression", "hard", "strategic_reasoning", 1]
+    ]
+)
+
+# Initialize model loading
+print("🚀 Starting model loading...")
+model_loaded = load_model()
 
-# Create and launch interface
-interface = create_gradio_interface()
+if model_loaded:
+    print("✅ Ready to generate MCQs!")
+else:
+    print("❌ Model loading failed. The interface may not work properly.")
 
+# Launch the interface
 if __name__ == "__main__":
     interface.launch()