Fix model loading - load on startup not during chat

- Move model loading to application startup instead of inside @spaces.GPU decorator
- Prevents timeout issues during chat by pre-loading the 20B model
- Model loads once at startup, not on every message
- Fixes 'processing' timeout error shown in interface

app.py

@@ -132,10 +132,9 @@ def chat(message, history):
     """Main chat function with improved error handling and conversation management"""
     global model, tokenizer
 
-    # Load model if not already loaded
+    # Check if model is loaded
     if model is None or tokenizer is None:
-        if not load_model():
-            return "❌ عذراً، حدث خطأ في تحميل النموذج. يرجى المحاولة مرة أخرى."
+        return "❌ عذراً، النموذج لم يتم تحميله بعد. يرجى الانتظار قليلاً والمحاولة مرة أخرى."
 
     try:
         # ======================================================
@@ -313,6 +312,12 @@ def create_interface():
     
     return demo
 
+# ======================================================
+# Load model on startup (before creating interface)
+# ======================================================
+logger.info("🚀 Starting application - loading model...")
+load_model()
+
 # Create the interface
 demo = create_interface()