Update app.py

app.py

@@ -35,12 +35,11 @@ def generate_story(theme):
     story = gpt2_tokenizer.decode(story_ids[0], skip_special_tokens=True)
     return story
 
-
 def generate_response(user_input):
-    # Enhanced empathetic prompt
-    response_prompt = f"User shares: '{user_input}'. Respond with empathy and understanding. Acknowledge their feelings and offer words of comfort and encouragement. Your response should focus on offering reassurance and emotional support, not practical advice."
+    # Empathy-focused prompt to guide the bot
+    response_prompt = f"The user has shared the following: '{user_input}'. Respond with empathy, compassion, and understanding. Acknowledge their sadness and offer comforting, reassuring words. Show that you care and validate their feelings without giving unsolicited advice."
     
-    # Generate the response using the model
+    # Generate the response using the GPT-2 model
     input_ids = gpt2_tokenizer.encode(response_prompt, return_tensors='pt')
     response_ids = gpt2_model.generate(
         input_ids,
@@ -51,18 +50,16 @@ def generate_response(user_input):
         num_return_sequences=1
     )
     
-    # Decode and clean up the response
+    # Decode the response and clean it up by removing the prompt
     response = gpt2_tokenizer.decode(response_ids[0], skip_special_tokens=True)
     
-    # Clean response from repeating the prompt
-    cleaned_response = response.replace(f"User shares: '{user_input}'. Respond with empathy and understanding. Acknowledge their feelings and offer words of comfort and encouragement. Your response should focus on offering reassurance and emotional support, not practical advice.", "").strip()
+    # Strip out the prompt portion to get a clean, empathetic message
+    cleaned_response = response.replace(f"The user has shared the following: '{user_input}'. Respond with empathy, compassion, and understanding. Acknowledge their sadness and offer comforting, reassuring words. Show that you care and validate their feelings without giving unsolicited advice.", "").strip()
     
-    # Return the cleaned response
     return cleaned_response
 
 
 
-
 # Analyze user input for emotional tone
 def get_emotion(user_input):
     emotions = emotion_classifier(user_input)