Allowed to max 256 tokens.

app.py

@@ -85,7 +85,7 @@ def evaluate_stream(msg, history, temperature, top_p):
     context = context.replace(r'<br>', '')
 
     # TODO: Avoid the tokens are too long.
-    CUTOFF = 192 
+    CUTOFF = 256 
     while len(tokenizer.encode(context)) > CUTOFF:
         # save 15 token size for the answer
         context = context[15:]
@@ -112,7 +112,7 @@ with gr.Blocks() as demo:
     state = gr.State()
     with gr.Row():
         with gr.Column(scale=2):
-            temp = gr.components.Slider(minimum=0, maximum=1.1, value=1.0, label="Temperature",
+            temp = gr.components.Slider(minimum=0, maximum=1.1, value=0.95, label="Temperature",
                 info="温度参数，越高的温度生成的内容越丰富，但是有可能出现语法问题。")
             top_p = gr.components.Slider(minimum=0.5, maximum=1.0, value=0.975, label="Top-p",
                 info="top-p参数，只输出前p>top-p的文字，越大生成的内容越丰富，但也可能出现语法问题。数字越小似乎上下文的衔接性越好。")
@@ -131,4 +131,4 @@ with gr.Blocks() as demo:
     gr.HTML(footer)
 
 demo.queue()
-demo.launch(debug=False)
+demo.launch(debug=False)