Update app.py

app.py

@@ -1,6 +1,7 @@
-# --- 🔱 हनुमान AI: अजेय एवं आत्मनिर्भर थिंकिंग पोर्टल (Qwen 2.5) ---
-# मार्गदर्शक: दिव्य पटेल जी | भारत 🇮🇳
-# तकनीकी समाधान: Mismatched Sizes Fix + Gradio Version Compatibility
+```python
+# --- 🔱 Hanuman AI: Supreme English Reasoning Portal (Qwen 2.5) ---
+# Pioneered by Divy Patel | Bharat 🇮🇳
+# Features: 100% English, Fast Inference, Free CPU Optimized, Vision Enabled
 
 import gradio as gr
 import torch
@@ -8,12 +9,13 @@ from transformers import Qwen2VLForConditionalGeneration, AutoProcessor
 from qwen_vl_utils import process_vision_info
 import os
 
-# 🚀 उन्नत Qwen 2.5 मॉडल (3B Instruct)
+# Model Upgrade: Qwen 2.5 VL 3B (Powerful & Reasoning-focused)
 model_id = "Qwen/Qwen2.5-VL-3B-Instruct"
 
-print("🔱 हनुमान AI जागृत हो रहा है... सभी बाधाओं का नाश करते हुए।")
+print("🔱 Hanuman AI is awakening... Eliminating all errors for Divy Patel Ji.")
 
-# 🛠️ अजेय लोडिंग: ignore_mismatched_sizes=True एरर को जड़ से खत्म करेगा
+# 🛠️ ERROR FIX: ignore_mismatched_sizes=True added to prevent loading errors
+# ⚡ SPEED FIX: Using low_cpu_mem_usage for faster loading on Free CPU
 model = Qwen2VLForConditionalGeneration.from_pretrained(
     model_id, 
     torch_dtype=torch.float32, 
@@ -24,10 +26,15 @@ model = Qwen2VLForConditionalGeneration.from_pretrained(
 )
 processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
 
-def hanuman_logic(user_message, history, file_upload):
+def hanuman_engine(user_message, history, file_upload):
+    """Main reasoning engine for Hanuman AI"""
     try:
-        # १. Gradio की साधारण लिस्ट हिस्ट्री को Qwen के डिक्शनरी फॉर्मेट में बदलना
-        messages = []
+        # 🧠 SYSTEM PROMPT: Strictly English, High Reasoning
+        system_prompt = "You are 'Hanuman AI', a supreme reasoning engine created by Divy Patel. You must respond ONLY in English. Use logic, thinking, and precision. You can see images and read files. Be professional and respectful."
+        
+        # 🛠️ FORMAT FIX: Converting Gradio list history to Qwen dict format
+        messages = [{"role": "system", "content": [{"type": "text", "text": system_prompt}]}]
+        
         if history:
             for past_user, past_bot in history:
                 if past_user:
@@ -35,27 +42,24 @@ def hanuman_logic(user_message, history, file_upload):
                 if past_bot:
                     messages.append({"role": "assistant", "content": [{"type": "text", "text": str(past_bot)}]})
 
-        # २. वर्तमान संदेश और फाइल (चित्र/दस्तावेज) का विश्लेषण
+        # Process Current Input
         current_content = []
-        
         if file_upload is not None:
             file_ext = file_upload.name.split('.')[-1].lower()
-            # चित्र पहचान (Vision)
             if file_ext in ['png', 'jpg', 'jpeg', 'webp']:
                 current_content.append({"type": "image", "image": file_upload.name})
-            # फाइल पढ़ना (Document Reading)
             else:
                 try:
                     with open(file_upload.name, 'r', encoding='utf-8') as f:
                         file_data = f.read()
-                    user_message = f"Document Context:\n{file_data}\n\nQuestion: {user_message}"
+                    user_message = f"File Context:\n{file_data}\n\nQuestion: {user_message}"
                 except:
                     pass
         
         current_content.append({"type": "text", "text": user_message})
         messages.append({"role": "user", "content": current_content})
 
-        # ३. मॉडल के लिए इनपुट तैयार करना
+        # Inference Setup
         text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
         image_inputs, video_inputs = process_vision_info(messages)
         
@@ -67,8 +71,13 @@ def hanuman_logic(user_message, history, file_upload):
             return_tensors="pt",
         ).to("cpu")
 
-        # ४. उत्तर का निर्माण (8192 टोकन क्षमता के साथ)
-        generated_ids = model.generate(**inputs, max_new_tokens=1024) 
+        # ⚡ SPEED OPTIMIZATION: Reduced max_new_tokens for faster CPU response
+        generated_ids = model.generate(
+            **inputs, 
+            max_new_tokens=512, 
+            do_sample=False # Greedy search for faster and more logical output
+        ) 
+        
         generated_ids_trimmed = [
             out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
         ]
@@ -76,35 +85,33 @@ def hanuman_logic(user_message, history, file_upload):
             generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
         )[0]
 
-        # ५. पुराने Gradio फॉर्मेट (List of Lists) में हिस्ट्री वापस करना
         history.append([user_message, response])
         return "", history, None
 
     except Exception as e:
-        error_msg = f"क्षमा करें दिव्य जी, एक तकनीकी बाधा आई है: {str(e)}"
+        error_msg = f"System error: {str(e)}"
         history.append([user_message, error_msg])
         return "", history, None
 
-# --- दिव्य पोर्टल इंटरफ़ेस (Gradio) ---
-# 🛡️ 'theme' को यहाँ से हटा दिया गया है ताकि कोई Warning न आए
+# --- UI PORTAL ---
 with gr.Blocks() as demo:
-    gr.HTML("<div style='text-align: center; color: #ff5500;'><h1>🔱 हनुमान AI - अजेय थिंकिंग पोर्टल (v2.5)</h1><p><b>Pioneered by Divy Patel | 100% Free | भारत 🇮🇳</b></p></div>")
+    gr.HTML("<div style='text-align: center;'><h1>🔱 Hanuman AI - English Reasoning Portal</h1><p>Pioneered by Divy Patel | 100% Free CPU | Bharat 🇮🇳</p></div>")
     
     with gr.Row():
         with gr.Column(scale=4):
-            # 🛡️ पुराने Gradio के लिए सरल चैटबॉट (बिना 'type' के)
             chatbot = gr.Chatbot(height=550)
             with gr.Row():
-                msg_input = gr.Textbox(placeholder="अपना प्रश्न यहाँ लिखें दिव्य जी...", scale=7)
-                file_input = gr.File(label="फोटो/फाइल", scale=2)
-                submit_btn = gr.Button("पूछें", variant="primary", scale=1)
+                msg_input = gr.Textbox(placeholder="Ask your complex question here...", scale=7)
+                file_input = gr.File(label="Upload Image/File", scale=2)
+                submit_btn = gr.Button("ASK", variant="primary", scale=1)
 
-    # इवेंट हैंडलिंग
     submit_btn.click(
-        fn=hanuman_logic, 
+        fn=hanuman_engine, 
         inputs=[msg_input, chatbot, file_input], 
         outputs=[msg_input, chatbot, file_input]
     )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()
+
+```