Collection configuration messages

Messages on this page are deprecation warnings from the Weaviate Python client. Nothing has failed: the collection was created or updated as you asked. The way the request described its vectors is on its way out, so the fix is always a rewrite of the configuration, never a change to your data. If your message is not here, the message index lists the other groups.

One or more of these warnings when you create or update a collection:

Dep024: You are using the `vectorizer_config` argument in `collection.config.create()`, which is deprecated.
            Use the `vector_config` argument instead.

Dep025: You are using the `vector_index_config` argument in `collection.config.create()`, which is deprecated.
            Use the `vector_config` argument instead defining `vector_index_config` as a sub-argument.

Dep017: You are using the `vector_index_config` argument in the `collection.config.update()` method, which is deprecated.
            Use the `vector_config` argument instead.

Dep023: You are using the `vectorizer_config` argument in the `collection.config.update()` method with a collection with named vectors, which is deprecated.
            Use the `vector_config` argument instead.

A collection used to describe its embeddings as two independent settings: the model that produces the vector, and the index that stores it. That shape cannot express a collection with several differently configured vectors, which Weaviate supports. Both settings now live inside a single vector definition, and a collection holds one or more of them, each with a name. The old arguments still work, and a collection created with them keeps the older single-vector shape.

Move the model and the index settings into one vector definition when you create the collection:

from weaviate.classes.config import Configure

client.collections.create(
    "Article",
    vector_config=Configure.Vectors.text2vec_weaviate(
        vector_index_config=Configure.VectorIndex.hnsw(),
    ),
)

from weaviate.classes.config import Configure

client.collections.create(
    "Article",
    vectorizer_config=Configure.Vectorizer.text2vec_weaviate(),
    vector_index_config=Configure.VectorIndex.hnsw(),
)

Updates name the vector they change. A collection created with the current argument, and no explicit vector name, has a single vector called default:

from weaviate.classes.config import Reconfigure

client.collections.get("Article").config.update(
    vector_config=Reconfigure.Vectors.update(
        name="default",
        vector_index_config=Reconfigure.VectorIndex.hnsw(ef=128),
    ),
)

from weaviate.classes.config import Reconfigure

client.collections.get("Article").config.update(
    vector_index_config=Reconfigure.VectorIndex.hnsw(ef=128),
)

Rewriting a collection definition does not re-embed anything and does not touch stored vectors.

A warning when you add a vector to an existing collection, quoted exactly as the client prints it, stray backtick included. The vector name in it is your own:

Dep026: You are using the named vector syntax for vector title_vector, e.g. `Configure.NamedVectors` in `collection.config.add_vector()`, which is deprecated.
            Use `Configure.Vectors` or `Configure.MultiVectors` instead.`

When named vectors were introduced they had their own set of builders, kept apart from the single-vector ones. Every vector is now a named vector, so the two sets were merged into one, split instead by whether the vector is a single vector or a multi-vector.

from weaviate.classes.config import Configure

collection = client.collections.get("Article")
collection.config.add_vector(
    vector_config=Configure.Vectors.text2vec_weaviate(
        name="title_vector",
        source_properties=["title"],
    ),
)

from weaviate.classes.config import Configure

collection = client.collections.get("Article")
collection.config.add_vector(
    vector_config=Configure.NamedVectors.text2vec_weaviate(
        name="title_vector",
        source_properties=["title"],
    ),
)

For a multi-vector embedding, such as ColBERT or ColPali, use the multi-vector builders instead. If your warning names the encoding argument rather than the named vector syntax, it is the other message that ships under this same id: see deprecated multi-vector index settings.

One or both of these warnings when you create a collection with a multi-vector embedding. You get both at once if you set an encoding inside the index configuration, because that nests one deprecated argument inside the other:

Dep027: You are using the `multi_vector` argument in `Configure.VectorIndex.hnsw()`, which is deprecated.
            Use the `multi_vector` argument inside `Configure.MultiVectors.module()` instead.

Dep026: You are using the `encoding` argument in `Configure.VectorIndex.MultiVectors.multi_vector()`, which is deprecated.
            Use the `encoding` argument inside `Configure.MultiVectors.module()` instead.

Whether a vector is a multi-vector is a property of the vector, not of the index that stores it, and the multi-vector builder you chose already implies it. Declaring it a second time on the index left two places that could disagree, so the setting moved onto the vector definition.

Move the setting onto the vector definition. The self_provided builder below stands for whichever multi-vector builder you use:

from weaviate.classes.config import Configure

client.collections.create(
    "Document",
    vector_config=Configure.MultiVectors.self_provided(
        name="page_vector",
        multi_vector_config=Configure.VectorIndex.MultiVector.multi_vector(),
        vector_index_config=Configure.VectorIndex.hnsw(),
    ),
)

from weaviate.classes.config import Configure

client.collections.create(
    "Document",
    vector_config=Configure.MultiVectors.self_provided(
        name="page_vector",
        vector_index_config=Configure.VectorIndex.hnsw(
            multi_vector=Configure.VectorIndex.MultiVector.multi_vector(),
        ),
    ),
)

An encoding such as MUVERA moves the same way, onto the vector definition:

from weaviate.classes.config import Configure

client.collections.create(
    "Document",
    vector_config=Configure.MultiVectors.self_provided(
        name="page_vector",
        encoding=Configure.VectorIndex.MultiVector.Encoding.muvera(),
        vector_index_config=Configure.VectorIndex.hnsw(),
    ),
)

Have a question or feedback? Here's how to reach us.