Update app.py

app.py

@@ -14,12 +14,17 @@ class GradioBackend:
         'Conversation': 'video_qa'
     }
 
-    def __init__(self, infer, audio_pipeline):
-        self.infer = infer
-        self.audio_pipeline = audio_pipeline
+    def __init__(self, ):
+        self.infer = LiveCCDemoInfer(model_path, device='cpu')
+        self.audio_pipeline = KPipeline(lang_code='a')
 
     def __call__(self, query: str = None, state: dict = {}, mode: str = 'Real-Time Commentary', **kwargs):
         return getattr(self.infer, self.mode2api[mode])(query=query, state=state, **kwargs)
+
+    def to(self, device):
+        self.infer.model.to(device)
+
+gradio_backend = GradioBackend()
             
 with gr.Blocks() as demo:
 
@@ -60,10 +65,8 @@ with gr.Blocks() as demo:
 
             @spaces.GPU
             def gr_chatinterface_fn(message, history, state, video_path, mode):
+                gradio_backend.to('cuda')
                 state['video_path'] = video_path
-                infer = LiveCCDemoInfer(model_path)
-                audio_pipeline = KPipeline(lang_code='a')
-                gradio_backend = GradioBackend(infer, audio_pipeline)
                 response, state = gradio_backend(query=message, state=state, mode=mode)
                 return response, state
             def gr_chatinterface_chatbot_clear_fn():