trying to avoid second load

app.py

@@ -127,7 +127,6 @@ class ModelManager:
             self.is_loaded = True
             self.is_loading = False
             print("Model loaded successfully!")
-            return self.model
 
         except Exception as e:
             print(f"Error loading model: {e}")
@@ -179,9 +178,9 @@ def prompt_lm(
         list[str]: List of generated responses for each audio-query pair
     """
     if model_manager.model is None:
-        model = model_manager.load_model()
+        model_manager.load_model()
 
-    if model is None:
+    if model_manager.model is None:
         if model_manager.is_loading:
             return "🔄 Loading model for the first query. This takes 20-30 seconds..👷🏽‍♂️🔨🪚"
             # while True:
@@ -196,7 +195,7 @@ def prompt_lm(
             return "Demo mode: Model not loaded. Please check the model configuration."
 
     with torch.amp.autocast(device_type="cuda", dtype=torch.float16):
-        results: list[list[dict]] = model(
+        results: list[list[dict]] = model_manager.model(
             audios,
             queries,
             window_length_seconds=window_length_seconds,