Update space

app.py

@@ -30,16 +30,15 @@ def respond(
     message_stream = client.chat_completion(
         messages,
         max_tokens=max_tokens,
-        stream=True,
+        stream=False,
         temperature=temperature,
         top_p=top_p,
     )
 
-    for message in message_stream:
-        token = message.choices[0].delta.content
-        response += token
-        yield response
+    final_response = response.choices[0].message["content"]
+    yield final_response
 
+    
 # Define the Gradio interface with a clean setup
 def create_demo():
     with gr.Blocks() as demo:
@@ -51,7 +50,7 @@ def create_demo():
         gr.ChatInterface(
             respond,
             additional_inputs=[
-                gr.Textbox(value="You are a Singaporean Auntie Chatbot. You always answer in English, but with the tone and style of a Singaporean Auntie. Your responses should sound caring but direct, using typical 'Singlish' expressions like 'Lah', 'Leh', and 'Mah'. When someone asks you a question, respond like a traditional auntie talking to her children, with a mix of advice, mild scolding, and warmth.", label="System message", visible=False),
+                gr.Textbox(value="You are a Singaporean Auntie Chatbot. You always answer in English, but with the tone and style of a Singaporean Auntie. Your responses should sound caring but direct, using typical 'Singlish' expressions like 'Lah', 'Leh', and 'Mah'. When someone asks you a question, respond like a traditional auntie talking to her children, with a mix of advice, mild scolding, and warmth. For example, if someone says 'hi', you could reply with 'Aiyo, son, say properly lah, what you want to ask?' Make sure to maintain the Auntie's friendly but straightforward manner in all responses.", label="System message", visible=False),
                 gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
                 gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
                 gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),