Update app.py

app.py

@@ -594,10 +594,10 @@ if __name__ == "__main__":
     bits = 4
     # set bits=4 for T4, bits=8 for A10G (24G), and bits=16 for A100 (40G)
 
-    preload_models(model_path, bits=bits)
+    preload_models(model_path)
 
     controller_proc = start_controller()
-    worker_proc = start_worker(model_path)
+    worker_proc = start_worker(model_path, bits=bits)
 
     # Wait for worker and controller to start
     time.sleep(10)