Skip to main content
Version: User Guides (Cloud)


Everything you need to know about operations on collections, partitions, indexes, and similarity searches.

Use Sparse Vector [READ MORE]

Sparse vectors represent words or phrases using vector embeddings where most elements are zero, with only one non-zero element indicating the presence of a specific word. Sparse vector models, such as SPLADEv2, outperform dense models in out-of-domain knowledge search, keyword-awareness, and interpretability. They are particularly useful in information retrieval, natural language processing, and recommendation systems, where combining sparse vectors for recall with a large model for ranking can significantly improve retrieval results.