Refactor collection creation to prefer named vectors #1654
Description
Currently we have two ways of creating vector indices:
- legacy:
vector_index_config=Configure.VectorIndex.hnsw(),
vectorizer_config=Configure.Vectorizer.none(),
- named vectors
vectorizer_config=[
wvc.config.Configure.NamedVectors.text2vec_contextionary(
"title",
source_properties=["title"],
vectorize_collection_name=False,
vector_index_config=wvc.config.Configure.VectorIndex.flat(
distance_metric=wvc.config.VectorDistances.HAMMING,
quantizer=wvc.config.Configure.VectorIndex.Quantizer.bq(rescore_limit=10),
),
),
wvc.config.Configure.NamedVectors.none(
"custom",
),
wvc.config.Configure.NamedVectors.text2vec_contextionary(
"default",
vectorize_collection_name=False, # needed as contextionary cant handle "_" in collection names
),
],
Instead we should:
- deprecate both ways of setting vectorizers in
Configure - deprecate
vectorizer_config - replace it with NamedVectorsByDefault that does not contain named vectors
Metadata
Metadata
Assignees
Labels
No labels