Migrate vector DBs to a new embedding model without re-embedding. Linear maps from ~2K calibration texts, ~87-91% retrieval retention (BEIR-benchmarked).
-
Updated
Sep 8, 2026 - Python
Migrate vector DBs to a new embedding model without re-embedding. Linear maps from ~2K calibration texts, ~87-91% retrieval retention (BEIR-benchmarked).
Recall falls below both ends while you migrate a vector index to a new embedding model. Reproducible experiments in numpy/hnswlib and a real Vespa deployment.
To associate your repository with the embedding-migration topic, visit your repo's landing page and select "manage topics."