Update app.py

app.py

@@ -45,12 +45,16 @@ def predict(message, history, temperature, max_new_tokens, min_new_tokens):
             yield partial_message
 
 
-iface = gr.ChatInterface(
+iface = gr.Interface(
     fn=predict,
-    inputs=["text", "text", gr.Slider(minimum=0.1, maximum=2.0, value=1.0, label="Temperature"),
-            gr.Slider(minimum=1, maximum=2048, value=1024, label="Max Tokens"),
-            gr.Slider(minimum=1, maximum=1024, value=1, label="Min Tokens")],
-    outputs="text"
+    inputs=["text", "text", gr.Slider(minimum=0.1, maximum=2.0, default=1.0, label="Temperature"),
+            gr.Slider(minimum=1, maximum=2048, default=1024, label="Max Tokens"),
+            gr.Slider(minimum=1, maximum=1024, default=1, label="Min Tokens")],
+    outputs="text",
+    live=True,
+    capture_session=True,
+    layout="vertical",
+    chat=True
 )
 
-iface.queue().launch()
+iface.launch()