Langchain elastic search error when retrieving using RetrievalQA

[raghuldeva]

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

from langchain_community.vectorstores import ElasticsearchStore
from langchain_community.embeddings import HuggingFaceBgeEmbeddings
import torch
from langchain.chains import RetrievalQA
from langchain_openai import ChatOpenAI

device = "cpu"
model_name = "BAAI/bge-large-en"
model_kwargs = {'device': device}
encode_kwargs = {'normalize_embeddings': False}
embeddings = HuggingFaceBgeEmbeddings(model_name=model_name,model_kwargs=model_kwargs,encode_kwargs=encode_kwargs)


db = ElasticsearchStore(es_url="http://localhost:9200", index_name="test-basic", embedding=embeddings)

qa = RetrievalQA.from_chain_type(
    llm=ChatOpenAI(temperature=0, openai_api_key="my-api-key"),
    chain_type="stuff",
    retriever=db.as_retriever(),
)


query = "What is compliance?"
result = qa.run(query)

print("Result is :", result)

Description

I'm trying to use langchain elasticsearch to retrieve context from document stored in index in elastic search using the above code .

the error I'm getting is

elasticsearch.BadRequestError: BadRequestError(400, 'parsing_exception', 'Unknown key for a START_OBJECT in [knn].')

System Info

langchain == 0.1.7
langchain-community == 0.0.20
elasticsearch == 8.12.0
python == 3.11.4

[dosubot[bot]]

🤖

Hey @raghuldeva, great to see you back diving into the depths of LangChain! 🚀

Based on the information you've provided and the context from the LangChain repository, it seems like the issue might be related to the version of Elasticsearch you're using or the format of your Elasticsearch index.

Firstly, LangChain v0.1.7 supports Elasticsearch version 8.10.0 as per the pyproject.toml file in the repository. You're currently using Elasticsearch version 8.12.0, which might be causing compatibility issues. I would recommend downgrading to version 8.10.0 and see if the issue persists.

Secondly, the error message BadRequestError(400, 'parsing_exception', 'Unknown key for a START_OBJECT in [knn].') suggests that there might be an issue with the format of your Elasticsearch index. LangChain's ElasticsearchStore expects a specific format for the Elasticsearch index, which varies depending on the retrieval strategy used (ApproxRetrievalStrategy, ExactRetrievalStrategy, or SparseRetrievalStrategy). You can find more details about the expected format in the elasticsearch.py file in the repository.

If you're still facing issues after trying these suggestions, could you please provide more details about the retrieval strategy you're using and the format of your Elasticsearch index? This will help in diagnosing the issue further.

Sources

• Can't use the ElasticVectorSearch store with the provided notion q-a example
• libs/community/langchain_community/vectorstores/elasticsearch.py
• templates/rag-elasticsearch/pyproject.toml

About Dosu

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[likhith00]

Hi @raghuldeva ,

could you mention some details ?

1. What are the versions of these libraries: torch, sentence_transformers?
2. elastic search version ?
3. you have used huggingface bge embeddings for vector store and tried to retreive using chat openai.... (I never tried it before). is there any particular reason ?

[ArnaudWald]

Hello,

I know this is a bit old, but for anyone else running into this error, here's a possible fix:

I had to change the DenseVectorStrategy of my ElasticsearchStore.

For some reason, it wasn't working with hybrid=True parameter.

Hope that helps !