Support for SPLADE sparse models in Milvus

[vrunm]

Is there a way to use SPLADE sparse embedding models in the milvus hybrid retriever?

[dosubot[bot]]

Yes, you can use SPLADE sparse embedding models in the Milvus hybrid retriever. The retriever is designed to handle both dense and sparse embeddings, allowing for hybrid search capabilities. You can implement a sparse embedding function for SPLADE and use it in the retriever setup, similar to how the BM25 algorithm is used for sparse embeddings ^[1].

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other}

[vrunm]

@dosu a Separate class similar to the BM25 (https://python.langchain.com/api_reference/_modules/langchain_milvus/utils/sparse.html#BM25SparseEmbedding) would have to be created?

[dosubot[bot]]

Yes, creating a separate class for SPLADE sparse embeddings, similar to the BM25SparseEmbedding class, would be a good approach. This class would handle the specific logic for generating and managing SPLADE sparse embeddings, allowing you to integrate it seamlessly into the Milvus hybrid retriever setup. You can use the existing BM25SparseEmbedding class as a reference for structuring your SPLADE implementation.

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other}