Error when Invoke RetrievalQA multiple output keys

[krisnagita]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

import os
import re

import json
import tiktoken
import requests
import nest_asyncio
import numpy as np
import pandas as pd
# import functions_framework
from datetime import datetime

from flask import jsonify

from typing import Any, Callable, Dict, List, Optional, Tuple, Type, Union

from google.cloud import bigquery, storage

from langchain_openai import AzureChatOpenAI, AzureOpenAIEmbeddings
from langchain_google_vertexai import ChatVertexAI, VertexAI, HarmBlockThreshold, HarmCategory

# from langchain.memory import ConversationBufferWindowMemory
from langchain.prompts import PromptTemplate, ChatPromptTemplate, MessagesPlaceholder, HumanMessagePromptTemplate
from langchain.text_splitter import CharacterTextSplitter, RecursiveCharacterTextSplitter
from langchain.docstore.document import Document

from langchain.chains import StuffDocumentsChain, LLMChain, RetrievalQA, create_retrieval_chain
from langchain.chains.combine_documents import create_stuff_documents_chain

from langchain_core.messages import SystemMessage, BaseMessage, HumanMessage, AIMessage, trim_messages
from langchain_core.output_parsers import StrOutputParser, JsonOutputParser
from langchain_core.runnables import RunnableParallel

from langchain_community.callbacks import OpenAICallbackHandler
from langchain_community.document_loaders import WebBaseLoader
from langchain_community.retrievers import GoogleVertexAISearchRetriever
from langchain_community.vectorstores import FAISS

from langgraph.checkpoint.memory import MemorySaver
from langgraph.graph import START, MessagesState, StateGraph

from langchain_google_community import VertexAISearchRetriever
from pydantic import BaseModel, Field

import nltk
from nltk.corpus import stopwords
from nltk.tokenize import word_tokenize, sent_tokenize
from string import punctuation

from heapq import nlargest
from bs4 import BeautifulSoup

from json_repair import repair_json

from langchain.memory import ConversationBufferMemory

list_history = [
    {"input": "would you give me about the latest sustainable pack for package food"}, 
    {"output": "The landscape of sustainable packaging for food has seen significant advancements, driven by both consumer demand and regulatory pressures. ..."}
]

def chat_to_memory(input_array: List, n_used_memory: int):
    
    memory = []
    
    for i in range(0, len(input_array)):
        if i % 2 == 0:
            memory.append(HumanMessage(content=input_array[i]['input']))
        elif i % 2 == 1:
            memory.append(AIMessage(content=input_array[i]['output']))
            
    selected_memory = trim_messages(
        memory,
        token_counter=len,
        max_tokens=n_used_memory,
        strategy="last",
        start_on="human",
        allow_partial=False,
    )
    
    return selected_memory

memory = chat_to_memory(list_history, 2)

# Initialize the memory with the conversation history
custom_memory = ConversationBufferMemory(return_messages=True)

# Add messages to the memory
for i in range(0, len(memory)):
    if i % 2 == 0:
        custom_memory.save_context({"input": memory[i].content}, {"output":memory[+1].content})

#Defining LLM and retriever

llm = ChatGPT() 

retriever = VertexAISearchRetriever(
    project_id=project_id,
    location_id=VERTEX_AI_SEARCH_LOCATION,
    data_store_id=VERTEX_AI_SEARCH_DATASTORE_ID,
    max_documents=6,
    engine_data_type = 2
) 

retrieval_qa_with_sources = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=retriever, return_source_documents=True, memory=custom_memory)

QUERY = "Could you make it into bullets format"

retrieval_qa_with_sources.invoke(QUERY)

Description

I'm trying to use LangChain RetrievalQA, my goal is to passing the memory into the RetrievalQA Chain

Initially I have a custom memory look like this

[HumanMessage(content='would you give me about the latest sustainable pack for package food', additional_kwargs={}, response_metadata={}),
AIMessage(content="The landscape of sustainable packaging for food has seen significant advancements, driven by both consumer demand and regulatory ... ", additional_kwargs={}, response_metadata={})]

Then I process it to ConversationBufferMemory that look like this then later I called it custom memory

custom_memory = ConversationBufferMemory(return_messages=True)

for i in range(0, len(memory)):
    if i % 2 == 0:
        custom_memory.save_context({"input": memory[i].content}, {"output":memory[+1].content})

Initialize my chain then later I invoke it

retrieval_qa_with_sources = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=retriever, return_source_documents=True, memory=custom_memory)

retrieval_qa_with_sources.invoke(SEARCH_QUERY)

When I invoke, I got error

ValueError: Got multiple output keys: dict_keys(['result', 'source_documents']), cannot determine which to store in memory. Please set the 'output_key' explicitly.

System Info

flask
google-cloud-storage
google-cloud-aiplatform
google-cloud-bigquery[pandas]
google-cloud-discoveryengine
google-cloud-aiplatform
google-api-python-client
google-auth
langchain
langchain-community
langchain-google-vertexai
langchain-openai
langchain-google-community
tiktoken
nest-asyncio
bs4
faiss-cpu
nltk
lxml

python version 3.9.20