Update app.py

app.py

@@ -64,6 +64,15 @@ def chat(history, system_message, max_tokens, temperature, top_p, top_k, repetit
     updated_history = call_nvidia_api(history, max_tokens, temperature, top_p)
     return updated_history, ""
 
+def update_chatbot(message, chat_history, system_message, max_tokens, temperature, top_p):
+    print("Updating chatbot...")
+    if not chat_history or (chat_history and chat_history[-1]["role"] != "user"):
+        chat_history = user(message, chat_history, system_message if not chat_history else None)
+    else:
+        chat_history = user(message, chat_history)
+    chat_history, _ = call_nvidia_api(chat_history, max_tokens, temperature, top_p)
+    return chat_history
+    
 # Gradio interface setup
 with gr.Blocks() as demo:
     with gr.Row():
@@ -87,41 +96,23 @@ with gr.Blocks() as demo:
 """
     gr.Markdown(description)
     chatbox = gr.Textbox(label="What do you want to chat about?", placeholder="Ask me anything.", lines=3)
-    chatbot = gr.ChatInterface(
-        fn=chat,
-        title="LLAMA 2 70B Chatbot",
-        submit_btn="Submit",
-        stop_btn="Stop",
-        retry_btn="🔄 Retry",
-        undo_btn="↩️ Undo",
-        clear_btn="🗑️ Clear",
-    )
     system_msg = gr.Textbox(BASE_SYSTEM_MESSAGE, label="System Message", placeholder="System prompt.", lines=5)
     max_tokens = gr.Slider(20, 1024, label="Max Tokens", step=20, value=1024, interactive=True)
     temperature = gr.Slider(0.0, 1.0, label="Temperature", step=0.1, value=0.2, interactive=True)
     top_p = gr.Slider(0.0, 1.0, label="Top P", step=0.05, value=0.7, interactive=True)
     chat_history_state = gr.State([])
 
-    with gr.Accordion("Additional Inputs"):
-        gr.Markdown("Customize the model's behavior using the inputs below.")
-        system_msg_container = gr.Container(system_msg, label="System Message")
-        max_tokens_container = gr.Container(max_tokens, label="Max Tokens")
-        temperature_container = gr.Container(temperature, label="Temperature")
-        top_p_container = gr.Container(top_p, label="Top P")
-
-    def update_chatbot(message, chat_history, system_message, max_tokens, temperature, top_p):
-        print("Updating chatbot...")
-        if not chat_history or (chat_history and chat_history[-1]["role"] != "user"):
-            chat_history = user(message, chat_history, system_message if not chat_history else None)
-        else:
-            chat_history = user(message, chat_history)
-        chat_history, _ = chat(chat_history, system_message, max_tokens, temperature, top_p, 40, 1.1)
-        return chat_history
-
-    chatbot.submit(
+    additional_inputs = [system_msg, max_tokens, temperature, top_p]
+
+    chatbot = gr.ChatInterface(
         fn=update_chatbot,
-        inputs=[chatbox, chat_history_state, system_msg, max_tokens, temperature, top_p],
-        outputs=chat_history_state
+        inputs=[chatbox, chat_history_state],
+        outputs=chat_history_state,
+        additional_inputs=additional_inputs,
+        title="LLAMA 2 70B Chatbot",
+        submit_btn="Submit",
+        clear_btn="🗑️ Clear",
+        textbox=chatbox,
     )
 
     chatbot.clear(