Update app.py

app.py

@@ -360,8 +360,14 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
         prompt_template = QA_CHAIN_PROMPT_1 if choice == "Details" else QA_CHAIN_PROMPT_2
 
         if retrieval_mode == "VDB":
+            # Retrieve context from the vector database
+            context = retriever.get_relevant_documents(message)
+
+            # Format the prompt
+            prompt = prompt_template.format(context=context, question=message)
+
             if selected_model == chat_model:
-                # Use Langchain with GPT-4o
+                # Use GPT-4o with Langchain
                 qa_chain = RetrievalQA.from_chain_type(
                     llm=chat_model,
                     chain_type="stuff",
@@ -370,15 +376,17 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
                 )
                 response = qa_chain({"query": message})
                 return response['result'], extract_addresses(response['result'])
+
             elif selected_model == phi_pipe:
-                # Directly use the Phi-3.5 model for text generation
-                response = selected_model(message, **{
-                    "max_new_tokens": 500,
+                # Use Phi-3.5 directly with the formatted prompt
+                response = selected_model(prompt, **{
+                    "max_new_tokens": 300,  # Limit the tokens for faster generation
                     "return_full_text": False,
-                    "temperature": 0.0,
-                    "do_sample": False,
+                    "temperature": 0.5,  # Adjust temperature for more consistent answers
+                    "do_sample": True,
                 })[0]['generated_text']
                 return response, extract_addresses(response)
+
         elif retrieval_mode == "KGF":
             response = chain_neo4j.invoke({"question": message})
             return response, extract_addresses(response)