Not able to use Hybrid search in Qdrant Langchain

[pradhandebasish2046]

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

Embedding_Model = "BAAI/bge-base-en-v1.5"

def embeddings():
    modelPath = Embedding_Model
    model_kwargs = {'device':'cpu'}
    encode_kwargs = {'normalize_embeddings': True}

    generate_embeddings = HuggingFaceEmbeddings(
        model_name=modelPath,     
        model_kwargs=model_kwargs, 
        encode_kwargs=encode_kwargs 
    )

    return generate_embeddings

get_embeddings=embeddings()

from langchain_qdrant import FastEmbedSparse, RetrievalMode

sparse_embeddings = FastEmbedSparse(model_name="prithivida/Splade_PP_en_v1")
kb_path = "url_kb_hybrid"
collection_name = "url_kb_collection_hybrid"
qdrant = QdrantVectorStore.from_documents(
    documents,
    ids=uuids,
    embedding=get_embeddings,
    sparse_embeddings=sparse_embeddings,
    path=kb_path,
    collection_name=collection_name,
    RetrievalMode=RetrievalMode.HYBRID, 
    vector_name="custom_vector",  
    sparse_vector_name="custom_sparse_vector", 
)
query = 'building breakfast LangGraph soccer'
docs = qdrant.similarity_search_with_score(query)
for i in docs:
    print(i[0].page_content)
    print(i[1])

Description

I am trying to use hybrid search using qdrant langchain. Everytime I am doing retrieval using RetrievalMode.HYBRID or RetrievalMode.DENSE or RetrievalMode.SPARSE I am getting exact same result with the same score. So I believe the hybrid search is not working properly.

System Info

System Information

OS: Linux
OS Version: #1 SMP Fri Mar 29 23:14:13 UTC 2024
Python Version: 3.10.12 (main, Jul 29 2024, 16:56:48) [GCC 11.4.0]

Package Information

langchain_core: 0.2.29
langchain: 0.2.12
langchain_community: 0.2.11
langsmith: 0.1.98
langchain_qdrant: 0.1.3
langchain_text_splitters: 0.2.2

Optional packages not installed

langgraph
langserve

Other Dependencies

aiohttp: 3.10.2
async-timeout: 4.0.3
dataclasses-json: 0.6.7
fastembed: 0.2.7
jsonpatch: 1.33
numpy: 1.26.4
orjson: 3.10.7
packaging: 24.1
pydantic: 2.8.2
PyYAML: 6.0.2
qdrant-client: 1.10.1
requests: 2.32.3
SQLAlchemy: 2.0.32
tenacity: 8.5.0
typing-extensions: 4.12.2

[dosubot[bot]]

Hey, @pradhandebasish2046! 👋 I'm here to help you with your bug, answer your questions, and even guide you on becoming a contributor. Let's squash those bugs together!

To resolve the issue where hybrid search with Qdrant and LangChain returns the same result with the same score for RetrievalMode.HYBRID, RetrievalMode.DENSE, and RetrievalMode.SPARSE, ensure that you are correctly setting up the QdrantVectorStore with the appropriate embeddings for each retrieval mode. Specifically, for hybrid search, you need to provide both dense and sparse embeddings.

Here is an example of how to set up the QdrantVectorStore for each retrieval mode:

Dense Vector Search

from langchain_qdrant import QdrantVectorStore, RetrievalMode
from langchain_openai import OpenAIEmbeddings

embeddings = OpenAIEmbeddings(model="text-embedding-3-large")

qdrant_dense = QdrantVectorStore.from_documents(
    docs,
    embedding=embeddings,
    location=":memory:",
    collection_name="my_documents",
    retrieval_mode=RetrievalMode.DENSE,
)

query = "What did the president say about Ketanji Brown Jackson"
found_docs_dense = qdrant_dense.similarity_search(query)

Sparse Vector Search

from langchain_qdrant import FastEmbedSparse, QdrantVectorStore, RetrievalMode

sparse_embeddings = FastEmbedSparse(model_name="Qdrant/BM25")

qdrant_sparse = QdrantVectorStore.from_documents(
    docs,
    sparse_embedding=sparse_embeddings,
    location=":memory:",
    collection_name="my_documents",
    retrieval_mode=RetrievalMode.SPARSE,
)

query = "What did the president say about Ketanji Brown Jackson"
found_docs_sparse = qdrant_sparse.similarity_search(query)

Hybrid Vector Search

from langchain_qdrant import FastEmbedSparse, QdrantVectorStore, RetrievalMode
from langchain_openai import OpenAIEmbeddings

embeddings = OpenAIEmbeddings(model="text-embedding-3-large")
sparse_embeddings = FastEmbedSparse(model_name="Qdrant/BM25")

qdrant_hybrid = QdrantVectorStore.from_documents(
    docs,
    embedding=embeddings,
    sparse_embedding=sparse_embeddings,
    location=":memory:",
    collection_name="my_documents",
    retrieval_mode=RetrievalMode.HYBRID,
)

query = "What did the president say about Ketanji Brown Jackson"
found_docs_hybrid = qdrant_hybrid.similarity_search(query)

Ensure that you are using the correct embeddings and retrieval mode for each search type. If the setup is correct and the issue persists, it might be necessary to check the implementation details or configurations of the embeddings and the Qdrant setup. Specifically, verify that the similarity_search_with_score method in QdrantVectorStore correctly handles the different retrieval modes and combines the results appropriately for hybrid search ^[1][2].

To continue talking to Dosu, mention @dosu.

[minakshimathpal]

@dosu I want to disable indexing while uploading the vectors. After Vectors are uploaded I want to index them. How do i do that.

[minakshimathpal]

@pradhandebasish2046 How are you indexing. Indexing while uploading is taking lot of time and is not recommended either. Please hlep

[mathpal-minakshi]

@pradhandebasish2046

[mathpal-minakshi]

@dosu Would appreciate any input regarding my issue

[mrst2000]

how to create instance of QdrantVectorStore class with feature of hybrid search , to use in ParentDocument Retriver as vectorstore ??

[ziudeso]

Same issue here... how do I specify top k in sparse and in dense?