Update app.py

app.py

@@ -357,15 +357,16 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
             response = fetch_google_flights()
             return response, extract_addresses(response)
 
-        prompt_template = QA_CHAIN_PROMPT_1 if choice == "Details" else QA_CHAIN_PROMPT_2
-
-        if retrieval_mode == "VDB":
-            # Retrieve context from the vector database
+        # Use a simple, direct prompt for Phi-3.5
+        if selected_model == phi_pipe:
+            prompt = f"Here is the information : , {message}"
+        else:
+            # Use the existing prompt templates for GPT-4o
+            prompt_template = QA_CHAIN_PROMPT_1 if choice == "Details" else QA_CHAIN_PROMPT_2
             context = retriever.get_relevant_documents(message)
-
-            # Format the prompt
             prompt = prompt_template.format(context=context, question=message)
 
+        if retrieval_mode == "VDB":
             if selected_model == chat_model:
                 # Use GPT-4o with Langchain
                 qa_chain = RetrievalQA.from_chain_type(
@@ -380,10 +381,10 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
             elif selected_model == phi_pipe:
                 # Use Phi-3.5 directly with the formatted prompt
                 response = selected_model(prompt, **{
-                    "max_new_tokens": 300,  # Limit the tokens for faster generation
+                    "max_new_tokens": 300,
                     "return_full_text": False,
-                    "temperature": 0.5,  # Adjust temperature for more consistent answers
-                    "do_sample": True,
+                    "temperature": 0.5,
+                    "do_sample": False,
                 })[0]['generated_text']
                 return response, extract_addresses(response)