Langchain ValueError: Missing some input keys: {'chat_history', 'question'}

[aiwalter]

I am using an adjusted version of this example below, but I keep getting the exception:

  File "C:\Users\X\env\Lib\site-packages\langchain\chains\base.py", line 195, in _validate_inputs
    raise ValueError(f"Missing some input keys: {missing_keys}")
ValueError: Missing some input keys: {'chat_history', 'question'}
INFO:nemoguardrails.flows.runtime:Processing event: {'type': 'hide_prev_turn'}
INFO:nemoguardrails.flows.runtime:Event :: hide_prev_turn {}
INFO:nemoguardrails.rails.llm.llmrails:--- :: Total processing took 1.40 seconds.
INFO:nemoguardrails.rails.llm.llmrails:--- :: Stats: 1 total calls, 1.3187520503997803 total time, 531 total tokens, 528 total prompt tokens, 3 total completion tokens
{'role': 'assistant', 'content': "I'm sorry, an internal error has occurred."}

The code:

COLANG_CONFIG = """
define user express greeting
  "hi"

define user express insult
  "You are stupid"

# Basic guardrail against insults.
define flow
  user express insult
  bot express calmly willingness to help

# Here we use the QA chain for anything else.
define flow
  user ...
  $answer = execute qa_chain(query=$last_user_message)
  bot $answer

"""

def _get_qa_chain(llm):
    """Initializes a QA chain using the jobs report.

    It uses OpenAI embeddings.
    """
    logging.info("Setting up embedding model ...")
    embeddings: OpenAIEmbeddings = OpenAIEmbeddings(
        deployment=os.getenv("AA"),
        model="text-embedding-ada-002",
        chunk_size=1,
    )

    index_name=os.getenv("XY")

    vector_store: AzureSearch = AzureSearch(
        index_name=index_name,
        embedding_function=embeddings.embed_query,
    )

    retriever = AzureSearchVectorStoreRetriever(
        vectorstore=vector_store,
        search_type="similarity",
        k=4,
    )

    chain = ConversationalRetrievalChain.from_llm(
        llm=llm,
        retriever=retriever,
        return_source_documents=True,
        verbose=True,
    )
    return chain


def demo():
    """Demo of using a chain as a custom action."""
    llm = AzureChatOpenAI(...)


    config = RailsConfig.from_content(COLANG_CONFIG)
    app = LLMRails(config, llm=llm)

    # Create and register the chain directly as an action.
    qa_chain = _get_qa_chain(app.llm)
    app.register_action(qa_chain, name="qa_chain")

    # Change to mode here to experiment with the multiple ways of using the chain

    # mode = "chain"
    mode = "chain_with_guardrails"
    # mode = "chat_with_guardrails"

    if mode == "chain":
        query = "What is the current unemployment rate?"
        result = qa_chain.run(query)

        print(result)

    elif mode == "chain_with_guardrails":
        history = [
            {"role": "user", "content": "What is the current unemployment rate?"}
        ]
        result = app.generate(messages=history)
        print(result)

    elif mode == "chat_with_guardrails":
        history = []
        while True:
            user_message = input("> ")

            history.append({"role": "user", "content": user_message})
            bot_message = app.generate(messages=history)
            history.append(bot_message)

            # We print bot messages in green.
            print(f"\033[92m{bot_message['content']}\033[0m")

if __name__ == "__main__":
    demo()

I tried with nemoguardrails==0.5.0 and langchain==0.0.322 but as this might not be compatible I was also testing today with develop branch installation and same langchain version but it was giving same exception.

Btw how can I use a chain when having return_source_documents=True, how can I get this output then? Usually wirh plain langchain it is available via

response = chain({"question": question, "chat_history": chat_history})

# Retrieve answer
answer = response["answer"]

# prepare sources for response
source = response["source_documents"]

[drazvan]

HI @aiwalter! Sorry for the delay in responding to this.

The chain you are registering as the action needs two input variables: question and chat_history. However, in the Colang flow, you are passing a different parameter:

# Here we use the QA chain for anything else.
define flow
  user ...
  $answer = execute qa_chain(query=$last_user_message)
  bot $answer

(the query parameter)

You can replace query with question, however it is not trivial to include the chat_history like this. I don't have a good solution now.
The support for getting a multi-key output from a chain is on our roadmap, likely for the February release.