Update app.py

app.py

@@ -118,12 +118,8 @@ def build_context(messages, user_message, bias_mode):
 
 
 def generate_response(user_message, messages, bias_mode):
-    """
-    - messages: list of message dicts (Chatbot "messages" format)
-    Returns: (cleared textbox, updated messages, last_user, last_bot)
-    """
     if not user_message.strip():
-        return "", messages, "", ""
+        return "", messages, messages, "", ""
 
     prompt_text = build_context(messages, user_message, bias_mode)
 
@@ -131,24 +127,30 @@ def generate_response(user_message, messages, bias_mode):
         prompt_text,
         max_new_tokens=120,
         do_sample=True,
-        top_p=0.95,
-        temperature=0.8,
+        top_p=0.9,
+        temperature=0.7,
         pad_token_id=tokenizer.eos_token_id,
     )
 
     full_text = outputs[0]["generated_text"]
     if "Assistant:" in full_text:
-        bot_reply = full_text.split("Assistant:")[-1].strip()
+        bot_part = full_text.split("Assistant:", 1)[1]
     else:
-        bot_reply = full_text.strip()
+        bot_part = full_text
+
+    # *** NEW: cut off if model starts writing a new "User:" line ***
+    bot_part = bot_part.split("\nUser:")[0].strip()
+
+    bot_reply = bot_part
 
-    # Update message history in "messages" format
     messages = messages + [
         {"role": "user", "content": user_message},
         {"role": "assistant", "content": bot_reply},
     ]
 
-    return "", messages, user_message, bot_reply
+    # return: cleared textbox, chatbot messages, state_messages, last_user, last_bot
+    return "", messages, messages, user_message, bot_reply
+
 
 
 def handle_thumb(thumb_value, last_user, last_bot, bias_mode):