Update app.py

app.py

@@ -418,20 +418,24 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
     logging.debug(f"generate_answer called with choice: {choice} and retrieval_mode: {retrieval_mode}")
 
     try:
+        # Handle hotel-related queries
         if "hotel" in message.lower() or "hotels" in message.lower() and "birmingham" in message.lower():
             response = fetch_google_hotels()
             return response, extract_addresses(response)
         
+        # Handle restaurant-related queries
         if "restaurant" in message.lower() or "restaurants" in message.lower() and "birmingham" in message.lower():
             response = fetch_yelp_restaurants()
             return response, extract_addresses(response)
 
+        # Handle flight-related queries
         if "flight" in message.lower() or "flights" in message.lower() and "birmingham" in message.lower():
             response = fetch_google_flights()
             return response, extract_addresses(response)
 
         if retrieval_mode == "VDB":
             if selected_model == chat_model:
+                # Use GPT-4o with its vector store and template
                 retriever = gpt_retriever
                 prompt_template = QA_CHAIN_PROMPT_1 if choice == "Details" else QA_CHAIN_PROMPT_2
                 context = retriever.get_relevant_documents(message)
@@ -447,25 +451,28 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
                 return response['result'], extract_addresses(response['result'])
 
             elif selected_model == phi_pipe:
+                # Use Phi-3.5 with its vector store and a simplified prompt
                 retriever = phi_retriever
                 context = retriever.get_relevant_documents(message)
-                
-                # Format the Phi-3.5 prompt according to the required format
-                phi_prompt = f"<|system|>\nYou are a helpful assistant.<|end|>\n<|user|>\n{message}\n<|end|>\n<|assistant|>\n"
-                logging.debug(f"Phi-3.5 Prompt: {phi_prompt}")
+                prompt = f"""
+                Here is the information :
+                {context}
+                {message}
+                """
 
-                response = selected_model(phi_prompt, **{
-                    "max_new_tokens": 128,  # Increased to handle longer responses
+                logging.debug(f"Phi-3.5 Prompt: {prompt}")
+
+                response = selected_model(prompt, **{
+                    "max_new_tokens": 128,
                     "return_full_text": False,
-                    "temperature": 0.7,  # Adjusted to avoid cutting off
-                    "do_sample": True,   # Allow sampling to increase response diversity
+                    "temperature": 0.5,
+                    "do_sample": False,
                 })
 
                 if response:
                     generated_text = response[0]['generated_text']
                     logging.debug(f"Phi-3.5 Response: {generated_text}")
-                    cleaned_response = clean_response(generated_text)
-                    return cleaned_response, extract_addresses(cleaned_response)
+                    return generated_text, extract_addresses(generated_text)
                 else:
                     logging.error("Phi-3.5 did not return any response.")
                     return "No response generated.", []
@@ -486,7 +493,6 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
 
 
 
-
 def bot(history, choice, tts_choice, retrieval_mode, model_choice):
     if not history:
         return history