Overview
Abstract base class and data model defining the interface for sparse embedding models used with Qdrant vector store.
Description
This module in the langchain-qdrant partner package defines two core abstractions: SparseVector, a Pydantic model representing a sparse vector with indices and values fields, and SparseEmbeddings, an abstract base class that specifies the interface for sparse embedding implementations. The SparseEmbeddings ABC requires implementations for embed_documents and embed_query, and provides default async versions (aembed_documents and aembed_query) that delegate to the sync methods via run_in_executor.
Usage
Import SparseEmbeddings when implementing a custom sparse embedding model for use with Qdrant hybrid search. Import SparseVector when you need to construct or type-hint sparse vector data structures.
Code Reference
Source Location
- Repository: Langchain_ai_Langchain
- File:
libs/partners/qdrant/langchain_qdrant/sparse_embeddings.py
- Lines: 1-33
Signature
class SparseVector(BaseModel, extra="forbid"):
"""Sparse vector structure."""
indices: list[int] = Field(..., description="indices must be unique")
values: list[float] = Field(..., description="values and indices must be the same length")
class SparseEmbeddings(ABC):
"""An interface for sparse embedding models to use with Qdrant."""
@abstractmethod
def embed_documents(self, texts: list[str]) -> list[SparseVector]:
"""Embed search docs."""
@abstractmethod
def embed_query(self, text: str) -> SparseVector:
"""Embed query text."""
async def aembed_documents(self, texts: list[str]) -> list[SparseVector]:
"""Asynchronous Embed search docs."""
...
async def aembed_query(self, text: str) -> SparseVector:
"""Asynchronous Embed query text."""
...
Import
from langchain_qdrant.sparse_embeddings import SparseEmbeddings, SparseVector
I/O Contract
SparseVector Fields
| Name |
Type |
Required |
Description
|
| indices |
list[int] |
Yes |
List of non-zero dimension indices. Must be unique.
|
| values |
list[float] |
Yes |
List of values corresponding to each index. Must be the same length as indices.
|
embed_documents
Inputs
| Name |
Type |
Required |
Description
|
| texts |
list[str] |
Yes |
List of document texts to embed.
|
Outputs
| Name |
Type |
Description
|
| return |
list[SparseVector] |
List of sparse vectors, one per input document.
|
embed_query
Inputs
| Name |
Type |
Required |
Description
|
| text |
str |
Yes |
Single query text to embed.
|
Outputs
| Name |
Type |
Description
|
| return |
SparseVector |
A sparse vector representation of the query.
|
Usage Examples
Implementing a Custom Sparse Embeddings Class
from langchain_qdrant.sparse_embeddings import SparseEmbeddings, SparseVector
class MySparseEmbeddings(SparseEmbeddings):
def embed_documents(self, texts: list[str]) -> list[SparseVector]:
# Custom sparse embedding logic
return [
SparseVector(indices=[0, 1, 2], values=[0.1, 0.5, 0.3])
for _ in texts
]
def embed_query(self, text: str) -> SparseVector:
# Custom sparse query embedding logic
return SparseVector(indices=[0, 1], values=[0.4, 0.6])
Using SparseVector Directly
from langchain_qdrant.sparse_embeddings import SparseVector
vec = SparseVector(indices=[10, 42, 99], values=[0.8, 0.3, 0.5])
print(vec.indices) # [10, 42, 99]
print(vec.values) # [0.8, 0.3, 0.5]
Related Pages
