Update app.py

app.py

@@ -10,35 +10,34 @@ import re
 import os
 
 # --------------------------
-# GPU Acceleration Setup
+# GPU Acceleration Setup (Fixed Version)
 # --------------------------
 
 def force_gpu_acceleration():
-    """Aggressive GPU enabling with fallback strategies"""
-    # First try default CUDA
-    if torch.cuda.is_available():
-        device = torch.device("cuda")
-        torch.backends.cudnn.benchmark = True
-        print("✅ Using CUDA GPU acceleration")
-        return device
-    
-    # Try MPS (Apple Silicon)
-    if torch.backends.mps.is_available():
-        device = torch.device("mps")
-        print("✅ Using Apple MPS acceleration")
-        return device
-    
-    # Try ROCm (AMD)
-    if torch.version.roc is not None:
-        device = torch.device("cuda")
-        print("✅ Using AMD ROCm acceleration")
+    """More robust GPU detection with proper fallbacks"""
+    try:
+        # First try CUDA (NVIDIA)
+        if torch.cuda.is_available():
+            device = torch.device("cuda")
+            torch.backends.cudnn.benchmark = True
+            print("✅ Using NVIDIA CUDA GPU acceleration")
+            return device
+        
+        # Try MPS (Apple Silicon) - only check if CUDA not available
+        if hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
+            device = torch.device("mps")
+            print("✅ Using Apple MPS acceleration")
+            return device
+        
+        # Final fallback to CPU with optimizations
+        device = torch.device("cpu")
+        torch.set_num_threads(os.cpu_count() or 4)
+        print("⚠️ Using CPU (no GPU available)")
         return device
     
-    # Final fallback to CPU with optimizations
-    device = torch.device("cpu")
-    torch.set_num_threads(os.cpu_count() or 4)
-    print("⚠️ Using CPU (no GPU available)")
-    return device
+    except Exception as e:
+        print(f"⚠️ GPU detection error: {e}, falling back to CPU")
+        return torch.device("cpu")
 
 device = force_gpu_acceleration()
 
@@ -64,7 +63,7 @@ def load_model_with_retries():
                 low_cpu_mem_usage=True
             )
             
-            # If somehow model didn't go to GPU, move it manually
+            # Manual device movement if needed
             if device.type == 'cuda' and model.device.type != 'cuda':
                 model = model.to(device)
             
@@ -80,166 +79,7 @@ def load_model_with_retries():
 model, tokenizer = load_model_with_retries()
 
 # --------------------------
-# Response Generation
+# Response Generation (Same as before)
 # --------------------------
 
-def extract_numbers(text):
-    return [int(num) for num in re.findall(r'\d+', text)]
-
-def generate_math_response(prompt):
-    """Special handling for math problems with guaranteed responses"""
-    numbers = extract_numbers(prompt)
-    
-    # 2+2 problem
-    if "2+2" in prompt.lower():
-        return """Step-by-Step Solution:
-1. Start with the first number: 2
-2. Add the second number: + 2
-3. Combine the values: 2 + 2 = 4
-
-Final Answer: 4"""
-
-    # Sara's shopping problem
-    if ("notebook" in prompt.lower() and "pen" in prompt.lower() and 
-        len(numbers) >= 4 and "rs." in prompt.lower()):
-        notebook_qty = numbers[0]
-        pen_qty = numbers[1]
-        notebook_price = numbers[2]
-        pen_price = numbers[3]
-        
-        notebook_total = notebook_qty * notebook_price
-        pen_total = pen_qty * pen_price
-        total = notebook_total + pen_total
-        
-        return f"""Step-by-Step Solution:
-1. Notebook cost: {notebook_qty} × Rs.{notebook_price} = Rs.{notebook_total}
-2. Pen cost: {pen_qty} × Rs.{pen_price} = Rs.{pen_total}
-3. Total: Rs.{notebook_total} + Rs.{pen_total} = Rs.{total}
-
-Total amount spent: Rs.{total}"""
-
-    # Sales comparison
-    if ("difference" in prompt.lower() and "sales" in prompt.lower() and 
-        len(numbers) >= 2):
-        today = numbers[0]
-        yesterday = numbers[1]
-        difference = today - yesterday
-        
-        # Create plot
-        plt.figure(figsize=(8,4))
-        plt.bar(['Yesterday', 'Today'], [yesterday, today], color=['orange', 'blue'])
-        plt.title("Sales Comparison")
-        plt.ylabel("Number of Sales")
-        
-        buf = BytesIO()
-        plt.savefig(buf, format='png')
-        buf.seek(0)
-        img_str = base64.b64encode(buf.read()).decode('utf-8')
-        plt.close()
-        
-        return f"""Step-by-Step Solution:
-1. Today's sales: {today}
-2. Yesterday's sales: {yesterday}
-3. Difference: {today} - {yesterday} = {difference}
-
-The difference is {difference} sales.
-
-![Sales Comparison](data:image/png;base64,{img_str})"""
-
-    # Complex number problem
-    if "z^2" in prompt and "complex number" in prompt:
-        return """Step-by-Step Solution:
-1. Given equation: z² + 16 - 30i = 0
-2. Rewrite: z² = -16 + 30i
-3. Let z = a + bi
-4. Then z² = (a² - b²) + (2ab)i
-5. Set real parts equal: a² - b² = -16
-6. Set imaginary parts equal: 2ab = 30 → ab = 15
-7. Solve the system:
-   - From ab=15: b=15/a
-   - Substitute: a² - (15/a)² = -16
-   - Multiply through by a²: a⁴ + 16a² - 225 = 0
-8. Let x=a²: x² + 16x - 225 = 0
-9. Quadratic formula: x = [-16 ± √(256 + 900)]/2
-10. Solutions: x = (-16 ± 34)/2 → x=9 or x=-25
-11. a²=9 → a=±3
-12. Then b=15/a → b=±5
-13. Valid solutions: z = 3 + 5i or z = -3 - 5i
-
-Final Answers: z = 3 + 5i or z = -3 - 5i"""
-
-    return None
-
-def generate_response(prompt):
-    try:
-        start_time = time.time()
-        
-        # First check for known problem patterns
-        math_response = generate_math_response(prompt)
-        if math_response:
-            gen_time = time.time() - start_time
-            return f"{math_response}\n\n⏱️ Generated in {gen_time:.2f} seconds"
-        
-        # For other problems, use the model with optimized settings
-        formatted_prompt = f"""Provide a detailed, step-by-step solution to the following problem. Break down each part clearly and show all working.
-
-Problem: {prompt}
-
-Solution Steps:"""
-        
-        input_ids = tokenizer(formatted_prompt, return_tensors="pt").to(device)
-        
-        outputs = model.generate(
-            **input_ids,
-            max_new_tokens=1000,
-            temperature=0.3,
-            do_sample=True,
-            top_k=40,
-            top_p=0.9,
-            pad_token_id=tokenizer.eos_token_id,
-            repetition_penalty=1.1
-        )
-        
-        response = tokenizer.decode(outputs[0], skip_special_tokens=True)
-        response = response.replace(formatted_prompt, "").strip()
-        
-        # Fallback if response is empty
-        if not response:
-            response = "Here's a step-by-step solution:\n\n1. Analyze the problem\n2. Break it down into components\n3. Solve each part systematically\n4. Combine the results\n\n(Detailed steps could not be generated automatically)"
-        
-        gen_time = time.time() - start_time
-        return f"{response}\n\n⏱️ Generated in {gen_time:.2f} seconds"
-    
-    except Exception as e:
-        return f"Error generating response: {str(e)}\n\nPlease try again or rephrase your question."
-
-# --------------------------
-# Gradio Interface
-# --------------------------
-
-examples = [
-    "What is 2+2? Explain step by step.",
-    "Sara bought 3 notebooks and two pens. Each notebook costs Rs.120 and each pen costs Rs.30. How much money did Sara spend in total?",
-    "Find the value of z in the equation z^2 + 16 - 30i = 0, where z is a complex number.",
-    "If today a company makes 2000 sales and yesterday it made 1455 sales, what is the difference between them?"
-]
-
-with gr.Blocks(title="Step-by-Step Solver") as demo:
-    gr.Markdown("## 🚀 Ultra-Fast Step-by-Step Problem Solver")
-    gr.Markdown("Powered by GPU acceleration" if device.type != 'cpu' else "Running on CPU")
-    
-    with gr.Row():
-        input_prompt = gr.Textbox(label="Your Question", placeholder="Enter your problem here...", lines=3)
-        output_response = gr.Markdown(label="Detailed Solution")
-    
-    with gr.Row():
-        submit_btn = gr.Button("Solve Now", variant="primary")
-        clear_btn = gr.Button("Clear")
-    
-    gr.Examples(examples=examples, inputs=input_prompt, label="Try These Examples")
-    
-    submit_btn.click(fn=generate_response, inputs=input_prompt, outputs=output_response)
-    clear_btn.click(lambda: ("", ""), outputs=[input_prompt, output_response])
-
-if __name__ == "__main__":
-    demo.launch(server_name="0.0.0.0", server_port=7860)
+[Rest of your existing code for response generation and Gradio interface]