11.5.0-beta.5 (2023-10-09)
Features Added
- Added support for
VectorSearch.Vectorizers
, which contains configuration options for vectorizing text vector queries, andVectorSearch.Profiles
, which define combinations of configurations to use with vector search. - Added the
VectorSearchAlgorithmConfiguration
base type, containing configuration options specific to the algorithm used during indexing and/or querying. Derived classes includeExhaustiveKnnVectorSearchAlgorithmConfiguration
andHnswVectorSearchAlgorithmConfiguration
. - Added the
SearchOptions.VectorQueries
base type, which is used for the query parameters for vector and hybrid search queries. Derived classes includeVectorizableTextQuery
andRawVectorQuery
. WithRawVectorQuery
, users can pass raw vector values for vector search, whileVectorizableTextQuery
allows the passing of text values to be vectorized for vector search. - Added
SearchOptions.VectorFilterMode
, determining whether filters are applied before or after vector search is executed. - Added
SearchOptions.SemanticQuery
, which enables the setting of a dedicated search query for semantic reranking, semantic captions, and semantic answers. - Added support for
AzureOpenAIEmbeddingSkill
, which enables the generation of vector embeddings for given text inputs using the Azure Open AI service. - Added
SearchIndexStatistics.VectorIndexSize
, which reports the amount of memory consumed by vectors in the index. - Added
KnowledgeStore.Parameters
, which defines a dictionary of knowledge store-specific configuration properties. - Added
SearchIndexerSkillset.IndexProjections
, which specifies additional projections to secondary search indexes.
Breaking Changes
- In
SearchOptions
, theIList<SearchQueryVector> Vectors
property has been removed in favor of the abstract base typeIList<VectorQuery> VectorQueries
. - In
SearchField
, thevectorSearchConfiguration
property has been removed in favor of the newVectorSearchProfile
property. - In
VectorSearch
,AlgorithmConfigurations
has been renamed toAlgorithms
.