fix: minor QA pipeline fix

qa/utils.py

@@ -4,6 +4,7 @@ from langchain import OpenAI
 from langchain.chains.qa_with_sources import load_qa_with_sources_chain
 from langchain.embeddings.openai import OpenAIEmbeddings
 from langchain.llms import OpenAI
+from langchain.chat_models import ChatOpenAI
 from langchain.docstore.document import Document
 from langchain.vectorstores import FAISS, VectorStore
 import docx2txt
@@ -13,7 +14,7 @@ from io import BytesIO
 import streamlit as st
 from .prompts import STUFF_PROMPT
 from pypdf import PdfReader
-from openai.error import AuthenticationError
+from langchain.memory import ConversationBufferWindowMemory
 
 class PDFFile:
     """A PDF file class for typing purposes."""
@@ -140,6 +141,7 @@ def embed_docs(_docs: Tuple[Document]) -> VectorStore:
 
     return index
 
+
 @st.cache_data
 def search_docs(_index: VectorStore, query: str, k:int=5) -> List[Document]:
     """Searches a FAISS index for similar chunks to the query
@@ -153,13 +155,15 @@ def search_docs(_index: VectorStore, query: str, k:int=5) -> List[Document]:
 @st.cache_data
 def get_answer(_docs: List[Document], query: str) -> Dict[str, Any]:
     """Gets an answer to a question from a list of Documents."""
-    # Get the answer
+    memory = ConversationBufferWindowMemory(k=5, input_key="question")  # only considering the last 5 messages
+    
+    # Create the chain to be used in this specific setting
     chain = load_qa_with_sources_chain(
-        OpenAI(temperature=0, 
-               openai_api_key=st.session_state.get("OPENAI_API_KEY")), 
-               chain_type="stuff",
-               prompt=STUFF_PROMPT
-            )
+        ChatOpenAI(temperature=0, openai_api_key=st.session_state.get("OPENAI_API_KEY"), model="gpt-4", streaming=True), 
+        chain_type="stuff",
+        prompt=STUFF_PROMPT,
+        # memory=memory NOTE: As of Aug 2023, memory is not supported in the QA chain - uncomment this line when it is supported
+        )
     # also returnig the text of the source used to form the answer
     answer = chain(
         {"input_documents": _docs, "question": query}