Update app.py

app.py

@@ -64,6 +64,15 @@ async def respond(
         print(f"Error: {e}")
         return "Error occurred. Please try again."
 
+
+# Make the Gradio app async
+async def generate_response(message, history, system_message, max_tokens, temperature, top_p):
+    new_history = history + [[message, ""]]
+    response = await respond(message, history, system_message, max_tokens, temperature, top_p)
+    new_history[-1][1] = response
+    return response, new_history
+
+
 def launch_app():
     try:
         demo = gr.Blocks()
@@ -77,12 +86,7 @@ def launch_app():
             top_p = gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-P")
             response = gr.Text(label="Response")
 
-            def generate_response(message, history, system_message, max_tokens, temperature, top_p):
-                new_history = history + [[message, ""]]
-                response = asyncio.run(respond(message, history, system_message, max_tokens, temperature, top_p))
-                new_history[-1][1] = response
-                return response, new_history
-
+            # Use the async version of generate_response
             gr.Button("Generate Response").click(
                 generate_response,
                 inputs=[message, history, system_message, max_tokens, temperature, top_p],