Update app.py

app.py

@@ -50,8 +50,10 @@ def respond(message, history: list[tuple[str, str]], system_message, max_tokens,
     generation_thread.start()
 
     # Stream the tokens as they are generated
+    text_buffer = ""
     for new_text in streamer:
-        yield new_text
+        text_buffer+=new_text
+        yield text_buffer
 
 
 # Create the Gradio interface
@@ -62,8 +64,7 @@ demo = gr.ChatInterface(
         gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
         gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
         gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
-    ],
-    live=True
+    ]
 )
 
 # Launch the Gradio app