added timeout and concurrency

app.py

@@ -30,12 +30,13 @@ async def predict(input, history):
 """
 Gradio Blocks low-level API that allows to create custom web applications (here our chat app)
 """
-with gr.Blocks() as demo:
-    logger.info("Starting Demo...")
+with gr.Blocks() as app:
+    logger.info("Starting app...")
     chatbot = gr.Chatbot(label="Allofresh Assistant")
     state = gr.State([])
     with gr.Row():
         txt = gr.Textbox(show_label=False, placeholder="Enter text, then press enter").style(container=False)
     txt.submit(predict, [txt, state], [chatbot, state])
 
-demo.launch()
+app.queue(concurrency_count=4)
+app.launch()

utils.py

@@ -7,7 +7,10 @@ load_dotenv()
 ALLOFRESH_SEARCH_API_BASE = os.getenv("ALLOFRESH_SEARCH_API_BASE")
 
 def search_allo_api(query, limit=3):
-    response = requests.get(f'{ALLOFRESH_SEARCH_API_BASE}?keyword={query}&limit={limit}&p=1')
+    response = requests.get(
+        f'{ALLOFRESH_SEARCH_API_BASE}?keyword={query}&limit={limit}&p=1',
+        timeout=120
+    )
     return json.loads(response.text)
 
 def lctool_search_allo_api(queries):