Update app.py

app.py

@@ -70,22 +70,25 @@ def predict_action(user_text):
 def generate_follow_up(user_text):
     """
     Generates a concise and conversational follow-up question related to the user's input.
-    This version is designed to prompt a friendly follow-up question that is relevant to the topic.
+    This version uses a clear prompt and provides a fallback if the generated output is empty.
     """
     prompt_text = (
-        f"Given the user's question: '{user_text}', generate a friendly and varied follow-up question. "
-        "For example, if the question is about quarks, you might ask, "
-        "'Would you like to know more about the six types of quarks, or is there another topic you'd like to explore?' "
-        "Keep it concise and conversational."
+        f"Generate a friendly, concise follow-up question based on the user's question: '{user_text}'. "
+        "The follow-up should invite further discussion. For example, if the user asked about quarks, you might ask, "
+        "'Would you like to learn more about the six types of quarks, or is there another topic you're curious about?' "
+        "Always return a follow-up question."
     )
-    hf = get_llm_hf_inference(max_new_tokens=64, temperature=0.7)
-    return hf.invoke(input=prompt_text).strip()
+    hf = get_llm_hf_inference(max_new_tokens=64, temperature=0.8)
+    follow_up = hf.invoke(input=prompt_text).strip()
+    if not follow_up:
+        follow_up = "Would you like to explore this topic further?"
+    return follow_up
 
 def get_response(system_message, chat_history, user_text, max_new_tokens=256):
     """
     Generates HAL's response in a friendly, conversational manner.
-    Uses sentiment analysis to adjust tone when appropriate, but only overrides for negative sentiment if
-    the user's input is not a direct question.
+    Uses sentiment analysis to adjust tone when appropriate.
+    Always generates a follow-up question that is appended to the chat history.
     """
     sentiment = analyze_sentiment(user_text)
     action = predict_action(user_text)
@@ -113,7 +116,7 @@ def get_response(system_message, chat_history, user_text, max_new_tokens=256):
         (
             "[INST] {system_message}\n\nCurrent Conversation:\n{chat_history}\n\n"
             "User: {user_text}.\n [/INST]\n"
-            "AI: Please answer the user's question without repeating any previous greetings. "
+            "AI: Please answer the user's question without repeating previous greetings. "
             "Keep your response friendly and conversational, starting with a phrase like "
             "'Certainly!', 'Of course!', or 'Great question!'.\nHAL:"
         )
@@ -126,10 +129,9 @@ def get_response(system_message, chat_history, user_text, max_new_tokens=256):
     chat_history.append({'role': 'user', 'content': user_text})
     chat_history.append({'role': 'assistant', 'content': response})
 
-    # Only override for negative sentiment if the user input is not a direct question.
+    # Only override with an empathetic response for negative sentiment if appropriate.
     if sentiment == "NEGATIVE" and not user_text.strip().endswith("?"):
         response = "I'm sorry you're feeling this way. I'm here to help. What can I do to assist you further?"
-        # Update the last assistant message in chat_history with the empathetic response.
         chat_history[-1]['content'] = response
 
     follow_up = generate_follow_up(user_text)