Update app.py

app.py

@@ -40,13 +40,13 @@ css = """
 .thoughts {
     border: 1px solid #ccc;
     padding: 10px;
-    background-color: #f9f9f9;
+    background-color: #000000;
     border-radius: 5px;
 }
 details summary {
     cursor: pointer;
     padding: 5px;
-    background-color: #e0e0e0;
+    background-color: #000000;
     border-radius: 5px;
     font-weight: bold;
 }
@@ -62,13 +62,13 @@ details[open] summary:after {
 """
 
 with gr.Blocks(css=css) as demo:
-    gr.Markdown("## Chat with Superthoughts")
-    gr.Markdown("**Warning:** The first output from the AI may take a few moments. After the first message, it should work quickly.")
+    gr.Markdown("## Chat with Superthoughts lite! (1.7B)")
+    gr.Markdown("**Warning:** The first output from the AI may take a few moments. After the first message, it should work at a decent speed.")
 
     chatbot = gr.Chatbot()
     msg = gr.Textbox()
     system_message = gr.Textbox(value="You must always include <think> ... </think> <output> </output> tokens.", label="System message")
-    max_tokens = gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens")
+    max_tokens = gr.Slider(minimum=1, maximum=4096, value=512, step=1, label="Max new tokens")
     temperature = gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature")
     top_p = gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)")
 
@@ -84,7 +84,7 @@ with gr.Blocks(css=css) as demo:
         history[-1][1] = formatted_response
         return history
 
-    msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then(
+    msg.submit(user, [msg, chatbot], [msg, chatbot], queue=True).then(
         bot, [chatbot, system_message, max_tokens, temperature, top_p], chatbot
     )