Update app.py

app.py

@@ -50,9 +50,7 @@ def respond(
     ):
         token = message.choices[0].delta.content
         response += token
-
-    # Append the response to the history
-    history.append((message, response))
+        yield response
 
     # Prepare the next question if there are more questions left
     if question_index < len(interview_questions):
@@ -61,9 +59,9 @@ def respond(
     else:
         next_question = "Thank you for answering all the questions. Do you have any questions for me?"
 
-    # Yield the assistant's response followed by the next question
-    full_response = response + "\n\n" + next_question
-    yield full_response
+    # Append the response to the history and add the next question
+    history.append((message, response))
+    yield response + "\n\n" + next_question
 
 """
 For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
@@ -73,7 +71,7 @@ demo = gr.ChatInterface(
     additional_inputs=[
         gr.Textbox(value="You are a manager conducting a job interview. Ask questions related to the candidate's experience, skills, and suitability for the role.", label="System message"),
         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=4.0, value=0.7, step=0.1, label="Temperature"),
         gr.Slider(
             minimum=0.1,
             maximum=1.0,