Update app.py

app.py

@@ -24,11 +24,11 @@ class LLMManager:
         self.models_config = models_config
         self.executor = ThreadPoolExecutor(max_workers=2)
         self.generation_lock = threading.Lock()
-        self.load_all_models()
         self.session = requests.Session()
         adapter = requests.adapters.HTTPAdapter(pool_connections=2, pool_maxsize=2)
         self.session.mount('http://', adapter)
         self.session.mount('https://', adapter)
+        self.load_all_models()
 
     def load_all_models(self):
         """Cargar todos los modelos en RAM"""
@@ -102,19 +102,19 @@ class LLMManager:
         """Generar respuesta con modelo específico"""
         if not self.generation_lock.acquire(blocking=False):
             return {"error": "Servidor ocupado - Generación en progreso"}
-    
+        
         try:
             model_data = self.get_model(model_name)
-        
+            
             if not model_data or not model_data["loaded"]:
                 error_msg = f"Modelo {model_name} no cargado"
                 if model_data and "error" in model_data:
                     error_msg += f": {model_data['error']}"
                 return {"error": error_msg}
-        
+            
             result = [None]
             exception = [None]
-        
+            
             def generate():
                 try:
                     result[0] = model_data["instance"].create_chat_completion(
@@ -123,21 +123,21 @@ class LLMManager:
                     )
                 except Exception as e:
                     exception[0] = e
-        
+            
             gen_thread = threading.Thread(target=generate, daemon=True)
             gen_thread.start()
             gen_thread.join(timeout=120)
-        
+            
             if gen_thread.is_alive():
                 return {"error": "Timeout en generación (120 segundos)"}
-        
+            
             if exception[0]:
                 raise exception[0]
-        
+            
             result[0]["provider"] = "telechars-ai"
             result[0]["model"] = model_name
             return result[0]
-        
+            
         finally:
             self.generation_lock.release()