fix: add zero-gpu

app.py

@@ -6,13 +6,14 @@ import gradio as gr
 from src.model.model_llava import inference
 from src.serve.gradio_block_arena_vision_named import build_side_by_side_vision_ui_named
 
-
-if __name__ == "__main__":
-    logger.info("inference called!", inference())
+@spaces.GPU(duration=120)
+def main():
     with gr.Blocks() as demo:
-        
         states = build_side_by_side_vision_ui_named(
             models=["llava-fire", "llava-original"]
         )
         
-        demo.launch()
+        demo.launch()
+
+if __name__ == "__main__":
+    main()