BaseVectorSearchCompression interface

Package:

@azure/search-documents

Contains configuration options specific to the compression method used during indexing or querying.

Properties

compressionName	The name to associate with this particular configuration.
kind	Polymorphic discriminator, which specifies the different types this object can be
rescoringOptions	Contains the options for rescoring.
truncationDimension	The number of dimensions to truncate the vectors to. Truncating the vectors reduces the size of the vectors and the amount of data that needs to be transferred during search. This can save storage cost and improve search performance at the expense of recall. It should be only used for embeddings trained with Matryoshka Representation Learning (MRL) such as OpenAI text-embedding-3-large (small). The default value is null, which means no truncation.

Property Details

compressionName

The name to associate with this particular configuration.

compressionName: string

Property Value

string

kind

Polymorphic discriminator, which specifies the different types this object can be

kind: "scalarQuantization" | "binaryQuantization"

Property Value

"scalarQuantization" | "binaryQuantization"

rescoringOptions

Contains the options for rescoring.

rescoringOptions?: RescoringOptions

Property Value

RescoringOptions

truncationDimension

The number of dimensions to truncate the vectors to. Truncating the vectors reduces the size of the vectors and the amount of data that needs to be transferred during search. This can save storage cost and improve search performance at the expense of recall. It should be only used for embeddings trained with Matryoshka Representation Learning (MRL) such as OpenAI text-embedding-3-large (small). The default value is null, which means no truncation.

truncationDimension?: number

Property Value

number