Prevent preallocation of gradio interface, do it at runtime

app.py

@@ -95,8 +95,8 @@ SUPPORTED_MODEL_NAMES = sorted(list(SUPPORTED_MODEL_NAMES))
 model_dict = {}
 for model_name in SUPPORTED_MODEL_NAMES:
     try:
-        iface = gr.Interface.load(f'models/{model_name}')
-        model_dict[model_name] = iface
+        # iface = gr.Interface.load(f'models/{model_name}')
+        model_dict[model_name] = None
     except:
         pass
 
@@ -370,6 +370,17 @@ def infer_audio(model_name: str, audio_file: str) -> str:
         # Obtain Gradio Model function from cache of models
         if model_name in model_dict:
             model = model_dict[model_name]
+
+            if model is None:
+                # Load the gradio interface
+                try:
+                    iface = gr.Interface.load(f'models/{model_name}')
+                except:
+                    iface = None
+
+                if iface is not None:
+                    # Update model cache
+                    model_dict[model_name] = iface
         else:
             model = None