Skip to main content

LangChain Integration with Qdrant

LangChain is a framework for developing applications powered by large language models (LLMs). It simplifies every stage of the LLM application lifecycle, i.e., Development, Productionization, and Deployment. You can integrate Qdrant with LangChain to leverage Qdrant's capabilities to enhance the search and retrieval functionalities within the LangChain ecosystem.

Pre-requisites​

Installing LangChain and Qdrant-Client​

  • For installing LangChain:

    pip install langchain

  • For installing Qdrant-Client:

    pip install qdrant-client

Importing Libraries​

  • First, we need to import some libraries provided by LangChain to help us in converting text into chunks and then embeddings and putting them into the Qdrant Vector Database.

    from langchain_community.document_loaders import TextLoader
    from langchain_text_splitters import CharacterTextSplitter
    from langchain.embeddings import HuggingFaceEmbeddings
    from langchain_community.vectorstores import Qdrant

Reading the File​

  • We need to read the contents of a file and store it in a variable. To do this, we will be using the TextLoader library.

    loader = TextLoader("<<FILE_NAME>>")
    documents = loader.load()

Chunking the Text​

  • The next step is to chunk the text from the file so that at the time of query from Qdrant, only the chunk of text that is the most similar to the query is displayed.

    text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
    docs = text_splitter.split_documents(documents)

Embedding and Inserting into Qdrant​

  • Now, we need to convert these chunks of text into vector embeddings and insert them into Qdrant. For converting chunks into vector embeddings, we will be using a HuggingFace Model as it is compatible with the LangChain library. You also need to specify your API Key and URL for accessing the Qdrant cluster in the code below.

    Note:

    You can get your Qdrant URL and API Key from the Overview tab in Vector Database on TIR. See the Qdrant Details subsection for more information.

    embeddings = HuggingFaceEmbeddings(
        model_name="sentence-transformers/all-mpnet-base-v2"
    )

    host = "<<YOUR_QDRANT_ENDPOINT_URL>>"
    # HTTP Port = 6333
    # gRPC Port = 6334
    port = "<<YOUR_PORT>>"
    api_key = "<<YOUR_QDRANT_API_KEY>>"
    qdrant = Qdrant.from_documents(
        docs,
        embeddings,
        host=host,
        port=port,
        api_key=api_key,
        collection_name="my_documents",
    )

    Note:

    You can specify some extra parameters like force_recreate=True to recreate your collection or prefer_grpc=True to use gRPC requests.

  • The last step is to give a prompt to the database and get the most similar output with the score that you can then give to your LLM as inference. If you don't want the score, you can use the similarity_search function instead.

    query = "<<QUERY>>"  # Question related to the text inserted in Qdrant.
    found_docs = qdrant.similarity_search_with_score(query)
    document, score = found_docs[0]

Maximum Marginal Relevance Search (MMR)​

  • If you’d like to look for some similar documents but also want to receive diverse results, MMR is a method you should consider. Maximal marginal relevance optimizes for similarity to query and diversity among selected documents.

    query = "<<QUERY>>"  # Question related to the text inserted in Qdrant.
    found_docs = qdrant.max_marginal_relevance_search(query, k=2, fetch_k=10)
    for i, doc in enumerate(found_docs):
        print(f"{i + 1}.", doc.page_content, "\n")

Now, you can pass the output text with the original query to your inference to get the context-specific response.

What's Next?​

This was a quick start guide to get you started with the Qdrant Integration with LangChain.