Update app.py

app.py

@@ -3,7 +3,7 @@ import json
 import requests, threading
 import sseclient
 import gradio as gr
-from server import setup_mixinputs, launch_vllm_server
+import server
 
 API_URL = "http://localhost:8000/v1/chat/completions"
 
@@ -80,6 +80,4 @@ with gr.Blocks(title="🎨 Mixture of Inputs (MoI) Demo") as demo:
     clear_btn.click(lambda: None, None, chatbot, queue=False)
 
 if __name__ == "__main__":
-    setup_mixinputs()
-    threading.Thread(target=launch_vllm_server, daemon=True).start()
     demo.launch()