Catch the failure to load the model too.

app.py

@@ -44,7 +44,7 @@ def fetch_model() -> object:
         if hasattr(_MODEL, 'name_or_path'):
             gradio.Info(title='Info', message='Switched to `{}`.'.format(getattr(_MODEL, 'name_or_path', 'None')), duration=2)
         else:
-            gradio.Warning(title='Warning', message='The GPU time slot expired before the model could be loaded.', duration=4)
+            gradio.Warning(title='Warning', message='Aborted: the GPU slot expired.', duration=4)
     # model object or `None`
     return _MODEL
 
@@ -117,16 +117,17 @@ def compute_logits(
     export_str: str,
 ) -> object:
     __logits = None
-    # load the model inside the GPU wrapper (not before)
-    __model = fetch_model()
     # the allocation might expire before the calculations are finished
     try:
+        # load the model inside the GPU wrapper (not before)
+        __model = fetch_model()
+        # compute the raw logits
         __logits = _ux.update_logits_state(
             indices_arr=indices_arr,
             export_str=export_str,
             model_obj=__model)
     except:
-        gradio.Warning(title='Warning', message='Aborted because the GPU slot expired.', duration=4)
+        gradio.Warning(title='Warning', message='Aborted: the GPU slot expired.', duration=4)
     # tensor or None
     return __logits