Update app.py

app.py

@@ -376,14 +376,23 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
             elif selected_model == phi_pipe:
                 # Use Phi-3.5 with its vector store and always use template2
                 retriever = phi_retriever
-                prompt_template = QA_CHAIN_PROMPT_2
+                context = retriever.get_relevant_documents(message)
+                # Construct a simple, direct prompt for Phi-3.5
+                prompt = f"""
+                Based on the following information, provide a concise and well-formatted response without including questions or 'Helpful Answer' sections:
 
-            # Retrieve context and construct the prompt
-            context = retriever.get_relevant_documents(message)
-            prompt = prompt_template.format(context=context, question=message)
+                {context}
+
+                Information:
+                {message}
+                """
 
             if selected_model == chat_model:
                 # Use GPT-4o with Langchain
+                prompt_template = QA_CHAIN_PROMPT_2  # Always using template2 for simplicity
+                context = retriever.get_relevant_documents(message)
+                prompt = prompt_template.format(context=context, question=message)
+
                 qa_chain = RetrievalQA.from_chain_type(
                     llm=chat_model,
                     chain_type="stuff",
@@ -394,7 +403,7 @@ 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 directly with the formatted prompt and specific vector store
+                # Use Phi-3.5 directly with the simplified prompt
                 response = selected_model(prompt, **{
                     "max_new_tokens": 300,
                     "return_full_text": False,