Technical GlossaryData Engineering and AI Infrastructure

Embedding Versioning

An approach for managing different embedding models or updated embedding-generation processes through versions.

Embedding versioning makes changes in embedding generation traceable so retrieval quality can be preserved. When the model changes, dimensionality shifts, or preprocessing rules are updated, old and new embeddings must not be mixed blindly. This is important for both reindexing strategy and comparative quality evaluation. In vector infrastructure, using embeddings without version control can lead to silent quality degradation.