Skip to content

feat: sparse embeddings in vector stores #493

Open
@ludwiktrammer

Description

@ludwiktrammer

Currently, Ragbits supports sparse embeddings (via the SparseVector and SparseEmbedder classes) but none of our vector stores supports this.

  1. Change the typing of VectorStoreResult to accept SparseVector (should now be `list[float] | SparseVector)
  2. Add support for sparse embeddings to Qdrant and in_memory vector stores. They should be able to take Embedder or SparseEmbedder and then return the vector as list[float] or SparseVector (depending ont he type of embedder).
  3. Write HOW-TO documentation on using sparse vectors with vector stores
  4. Update the HOW-TO documentation on hybrid search - resolve both the TODO comments that reference sparse vectors.

Metadata

Metadata

Assignees

No one assigned

    Labels

    featureNew feature or request

    Type

    No type

    Projects

    Status

    Ready

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions