Discussion
Vectors, meet Virtuoso.
Native HNSW vector indexing has landed in the Virtuoso Open Source develop/7 branch, with ONNX-based text embeddings, optimizer tuning, Windows build support and the new VVec console alongside.
Similarity search across native relational tables via SQL and RDF graphs via SPARQL, one engine.
Branch-level today, ahead of a release. Enjoy!
The commits, straight from the source:
HNSW indexing: https://github.com/openlink/virtuoso-opensource/commit/3d6a0d5
Merge into develop/7: https://github.com/openlink/virtuoso-opensource/commit/d166c69
Full history: https://github.com/openlink/virtuoso-opensource/commits/develop/7/
Architectural visualization.
#VectorSearch #VirtuosoRDBMS #SQL #SPARQL #AI #LLMs #KnowledgeGraph #Database
ONNX provides the bridge btw Virtuoso and local embedding models. By default, local embeddings are IBM’s granite-embedding-30m-english model based. Users can also use alternative models.
Embeddings from OpenAI, OpenAI-compatible APIs, etc. r also options for vector generation.
Showcasing the impact of vector similarity search on a DBpedia instance where embeddings have been generated and indexed on literal objects in a DBpedia knowledge graph.
#VectorSimilarity #DBpedia #SPARQL #VirtuosoRDBMS #HowTo #AI
Install Fedizen
Get the full app experience
Similarity search and SPARQL in one engine is the bit that matters for entity work. Find candidates by embedding, then confirm identity against the graph in the same query, instead of shuttling IDs between a vector store and a triple store and hoping they still agree.