revise default setting

app.py

@@ -46,7 +46,7 @@ if __name__ == '__main__':
     parser.add_argument('--top_p', type=float, default=1.0, help='top_p for the llm model')
     parser.add_argument('--top_k', type=int, default=40, help='top_k for the llm model')
     parser.add_argument('--num_beams', type=int, default=4, help='num_beams for the llm model')
-    parser.add_argument('--keep_last_n_paragraphs', type=int, default=1, help='keep last n paragraphs in the memory')
+    parser.add_argument('--keep_last_n_paragraphs', type=int, default=0, help='keep last n paragraphs in the memory')
     args = parser.parse_args()
 
     examples = [
@@ -81,10 +81,10 @@ if __name__ == '__main__':
                         interactive=True,
                         label="Remember Last N Paragraphs")
                     max_new_token = gr.Slider(
-                        minimum=128,
-                        maximum=1024,
+                        minimum=64,
+                        maximum=512,
                         value=args.max_new_tokens,
-                        step=64,
+                        step=1,
                         interactive=True,
                         label="Max New Tokens")
                     temperature = gr.Slider(