Implementation:Langchain ai Langchain FastEmbedSparse

Knowledge Sources	Langchain_ai_Langchain
Domains	Vector Store, Embeddings, Qdrant, Sparse Embeddings
Last Updated	2026-02-11 00:00 GMT

Overview

A sparse embedding implementation using the FastEmbed library for generating sparse text embeddings compatible with Qdrant vector store.

Description

FastEmbedSparse is a concrete implementation of the SparseEmbeddings abstract base class in the langchain-qdrant partner package. It wraps the fastembed.SparseTextEmbedding model to produce sparse vector representations of text, which are useful for hybrid search in Qdrant. The class supports configurable model names (defaulting to Qdrant/bm25), batch sizes, cache directories, threading, ONNX execution providers, and data-parallel encoding for large datasets.

Usage

Import this class when you need sparse text embeddings for hybrid (dense + sparse) vector search in Qdrant. Requires the fastembed or fastembed-gpu package to be installed.

Code Reference

Source Location

• Repository: Langchain_ai_Langchain
• File: libs/partners/qdrant/langchain_qdrant/fastembed_sparse.py
• Lines: 1-84

Signature

class FastEmbedSparse(SparseEmbeddings):
    """An interface for sparse embedding models to use with Qdrant."""

    def __init__(
        self,
        model_name: str = "Qdrant/bm25",
        batch_size: int = 256,
        cache_dir: str | None = None,
        threads: int | None = None,
        providers: Sequence[Any] | None = None,
        parallel: int | None = None,
        **kwargs: Any,
    ) -> None:
        ...

    def embed_documents(self, texts: list[str]) -> list[SparseVector]:
        ...

    def embed_query(self, text: str) -> SparseVector:
        ...

Import

from langchain_qdrant import FastEmbedSparse

I/O Contract

Inputs (Constructor Parameters)

Name	Type	Required	Description
model_name	str	No	The name of the sparse embedding model. Defaults to "Qdrant/bm25".
batch_size	int	No	Batch size for encoding. Defaults to 256.
cache_dir	None	No	Path to the model cache directory. Can also be set via FASTEMBED_CACHE_PATH env variable.
threads	None	No	Number of threads for the ONNX runtime session.
providers	None	No	List of ONNX execution providers.
parallel	None	No	Data-parallel encoding setting. >1 enables parallelism, 0 uses all cores, None uses default threading.
**kwargs	Any	No	Additional options passed to fastembed.SparseTextEmbedding.

embed_documents

Inputs

Name	Type	Required	Description
texts	list[str]	Yes	List of text documents to embed.

Outputs

Name	Type	Description
return	list[SparseVector]	List of sparse vectors, each containing indices and values lists.

embed_query

Inputs

Name	Type	Required	Description
text	str	Yes	Single query text to embed.

Outputs

Name	Type	Description
return	SparseVector	A sparse vector with indices and values lists.

Usage Examples

Basic Usage

from langchain_qdrant import FastEmbedSparse

# Initialize with default BM25 model
sparse_embeddings = FastEmbedSparse(model_name="Qdrant/bm25")

# Embed documents
docs = ["LangChain is a framework for LLM applications", "Qdrant is a vector database"]
sparse_vectors = sparse_embeddings.embed_documents(docs)

# Embed a query
query_vector = sparse_embeddings.embed_query("What is LangChain?")
print(query_vector.indices)  # Non-zero dimension indices
print(query_vector.values)   # Corresponding values

With Custom Configuration

from langchain_qdrant import FastEmbedSparse

sparse_embeddings = FastEmbedSparse(
    model_name="Qdrant/bm25",
    batch_size=128,
    threads=4,
    parallel=2,
    cache_dir="/tmp/fastembed_cache",
)

vectors = sparse_embeddings.embed_documents(["document text"])

Related Pages

• Langchain_ai_Langchain_SparseEmbeddings
• Langchain_ai_Langchain_Qdrant_Vector_Utils