External Table support

Columns of type VECTOR can be included in external tables, providing the option to store vector embeddings used for AI workloads outside of the database while still using the database as a semantic search engine.

If using RAC instances, you must enable Patch 37244967 using Oracle RAC two-stage rolling updates.

Vectors in External Tables

Arithmetic and aggregate function support

The arithmetic operators for addition, subtraction, and multiplication along with the aggregate functions SUM and AVG can be used with vectors and columns of type VECTOR.

Arithmetic operators are supported for use with vectors in both SQL and PL/SQL.

If using RAC instances, you must enable Patch 37289468 using Oracle RAC two-stage rolling updates for support in PL/SQL.

Arithmetic Operators

Aggregate Functions

PL/SQL BINARY vector support

In addition to FLOAT32 (the default format for comma-separated string representations of vectors), FLOAT64, and INT8 dimension formats, you can also use the BINARY dimension format.

A BINARY vector represents each dimension as a single bit (0 or 1). BINARY vectors can now be created and declared in PL/SQL, supported in the same way as they are in SQL.

If using RAC instances, you must enable Patch 37289468 using Oracle RAC two-stage rolling updates.

BINARY Vectors

PL/SQL JACCARD distance metric support

The distance metric JACCARD can be used in PL/SQL for similarity searches with BINARY vectors.

In PL/SQL, there is no standalone shorthand JACCARD_DISTANCE function. Use JACCARD as the distance metric parameter of the VECTOR_DISTANCE function instead.

If using RAC instances, you must enable Patch 37289468 using Oracle RAC two-stage rolling updates.

VECTOR_DISTANCE

For additional information about VECTOR operations supported by PL/SQL, see Oracle Database PL/SQL Language Reference

Globally Distributed Database support

Columns of type VECTOR can be included in sharded tables and duplicated tables in a distributed database, and vector indexes are supported on sharded tables in a distributed database, with some restrictions.

Vectors in Distributed Database Tables

Vector Indexes in a Globally Distributed Database

"Restrictions" in Create Tables Using the VECTOR Data Type

In-Database Algorithms Support for VECTOR Data Type Predictors

The VECTOR data type is supported and can be used as an input to database machine learning algorithms such as classification, anomaly, regression, clustering and feature extraction.

Overview of Oracle AI Vector Search

More information on using VECTOR data type in machine learning can be found in Vector Data Type Support

Support for Image Transformer Models with AI Vector Search Using the In-Database ONNX Runtime

The Oracle Database ONNX runtime engine includes support of models for encoding images.

Hybrid Vector Index for JSON

New PATHS field is added to the DBMS_VECTOR_CHAIN.VECTORIZER that allows specification of an array of path objects : CREATE_PREFERENCE

New syntax in the DBMS_HYBRID_VECTOR.SEARCH API to allow searches restricted to paths. The VECTOR field accepts an INPATH parameter which an array of valid JSON paths to restrict the search : SEARCH

Support for the creation of a LOCAL index on a HVI when the underlying vector index_type is IVF : CREATE HYBRID VECTOR INDEX

Support for running optimization only on the $VR table for an HVI : Hybrid Vector Index Maintenance Operations

Included Columns in Neighbor Partition Vector Indexes

Included columns permit additional table (non-vector) columns to be stored in a Neighbor Partition vector index. The benefit is that query execution is optimized by removing the need to access the underlying base table to retrieve these columns.

If using RAC instances, you must enable Patch 37306139 using Oracle RAC two-stage rolling updates.

Included Columns