dyna-record
    Preparing search index...

    Interface EmbeddingModelDescriptor

    Pure-data description of an embedding model: the model name, the dimensions of the vectors it produces, the distance function the vector index should use, and the conversion from a raw distance score to a similarity value.

    Descriptors carry no client, credentials, or SDK dependency. Only the descriptor's name is ever serialized (EX: through metadata()).

    interface EmbeddingModelDescriptor {
        dimensions: number;
        distanceFunction: VectorDistanceFunction;
        name: string;
        scoreToSimilarity: (score: number) => number;
    }
    Index

    Properties

    dimensions distanceFunction name scoreToSimilarity

    Properties

    dimensions: number

    The number of dimensions of the vectors the model produces.

    distanceFunction: VectorDistanceFunction

    The distance function the vector index should be provisioned with.

    name: string

    The model name (EX: a Bedrock model id). This is the only part of the descriptor that appears in serialized metadata.

    scoreToSimilarity: (score: number) => number

    Converts a raw distance score returned by DynamoDB to a similarity value.

    Settings

    Member Visibility

    On This Page

    Properties