Update app.py

app.py

@@ -108,7 +108,7 @@ def generate(
         max_new_tokens=max_new_tokens,
         top_p=top_p,
         top_k=40,
-        # repetition_penalty=repetition_penalty,
+        repetition_penalty=repetition_penalty,
         do_sample=True,
         truncate=1024,
         # seed=42,
@@ -116,7 +116,7 @@ def generate(
         stop_sequences=[DEFAULT_SEPARATOR]
     )
 
-    print(prompt)
+    # print(prompt)
     stream = CURRENT_CLIENT.generate_stream(
         prompt,
         **generate_kwargs,
@@ -175,7 +175,7 @@ with gr.Blocks(analytics_enabled=False, css=custom_css) as demo:
             )
 
     with gr.Row():
-        with gr.Box():
+        with gr.Group():
             output = gr.Markdown()
             chatbot = gr.Chatbot(elem_id="chat-message", label="Chat")
 
@@ -291,4 +291,4 @@ with gr.Blocks(analytics_enabled=False, css=custom_css) as demo:
 
     clear_chat_button.click(clear_chat, outputs=[chatbot, history])
 
-demo.queue(concurrency_count=16).launch(server_port=8001)
+demo.queue().launch()