Is your feature request related to a problem? Please describe.
The Hugging Face API integration supports dense embeddings but not TEI's /embed_sparse endpoint, requiring custom components for sparse or hybrid retrieval.
Describe the solution you'd like
Add:
HuggingFaceAPISparseTextEmbedderHuggingFaceAPISparseDocumentEmbedder
These should call /embed_sparse and return/populate Haystack SparseEmbedding objects.
Describe alternatives you've considered
Using a custom component or a different sparse embedding backend such as Sentence Transformers or FastEmbed.
Additional context
TEI supports sparse embeddings through /embed_sparse, including SPLADE-compatible models. Haystack already supports SparseEmbedding and Document.sparse_embedding, so the endpoint maps naturally to existing Haystack types.
Is your feature request related to a problem? Please describe.
The Hugging Face API integration supports dense embeddings but not TEI's
/embed_sparseendpoint, requiring custom components for sparse or hybrid retrieval.Describe the solution you'd like
Add:
HuggingFaceAPISparseTextEmbedderHuggingFaceAPISparseDocumentEmbedderThese should call
/embed_sparseand return/populate HaystackSparseEmbeddingobjects.Describe alternatives you've considered
Using a custom component or a different sparse embedding backend such as Sentence Transformers or FastEmbed.
Additional context
TEI supports sparse embeddings through
/embed_sparse, including SPLADE-compatible models. Haystack already supportsSparseEmbeddingandDocument.sparse_embedding, so the endpoint maps naturally to existing Haystack types.