Language

InferenceModelFactory.MultimodalMultiVectorEmbedding Method

Definition

The result for one item when vectorKind is multi: one vector per token, for late-interaction (MaxSim) scoring.

public static Azure.Data.AI.MultimodalMultiVectorEmbedding MultimodalMultiVectorEmbedding(int index = 0, string id = default, Azure.Data.AI.EmbeddingContentKind contentKind = default, int? width = default, int? height = default, Azure.Data.AI.EmbeddingImageFormat? format = default, long? tokens = default, long? visualTokens = default, bool? isTruncated = default, long? inputTokens = default, System.Collections.Generic.IEnumerable<int> shape = default, BinaryData multiEmbedding = default);
static member MultimodalMultiVectorEmbedding : int * string * Azure.Data.AI.EmbeddingContentKind * Nullable<int> * Nullable<int> * Nullable<Azure.Data.AI.EmbeddingImageFormat> * Nullable<int64> * Nullable<int64> * Nullable<bool> * Nullable<int64> * seq<int> * BinaryData -> Azure.Data.AI.MultimodalMultiVectorEmbedding
Public Shared Function MultimodalMultiVectorEmbedding (Optional index As Integer = 0, Optional id As String = Nothing, Optional contentKind As EmbeddingContentKind = Nothing, Optional width As Nullable(Of Integer) = Nothing, Optional height As Nullable(Of Integer) = Nothing, Optional format As Nullable(Of EmbeddingImageFormat) = Nothing, Optional tokens As Nullable(Of Long) = Nothing, Optional visualTokens As Nullable(Of Long) = Nothing, Optional isTruncated As Nullable(Of Boolean) = Nothing, Optional inputTokens As Nullable(Of Long) = Nothing, Optional shape As IEnumerable(Of Integer) = Nothing, Optional multiEmbedding As BinaryData = Nothing) As MultimodalMultiVectorEmbedding

Parameters

index
Int32

The zero-based position of the item in the request.

id
String

The item's id from the request, if it had one.

contentKind
EmbeddingContentKind

The item's content kind, as supplied in the request's kind.

width
Nullable<Int32>

For images: the width in pixels.

height
Nullable<Int32>

For images: the height in pixels.

format
Nullable<EmbeddingImageFormat>

For images: the detected image format.

tokens
Nullable<Int64>

The tokens the model read for this item, including any text the service adds to queries and, for images, the image prompt and image tokens. For a truncated item, the tokens embedded. Omitted when the model doesn't report it.

visualTokens
Nullable<Int64>

For images: how many of tokens were image tokens.

isTruncated
Nullable<Boolean>

true when the item was over the model's token limit and only its start was embedded; warnings then has an EMBEDDING_INPUT_TRUNCATED entry for it. Omitted otherwise.

inputTokens
Nullable<Int64>

For truncated items only: the item's tokens before truncation.

shape
IEnumerable<Int32>

The number of vectors and the number of dimensions: [vectors, dimensions].

multiEmbedding
BinaryData

The vectors: an array of arrays of numbers, or, when encodingFormat is base64, one base64 string of all their little-endian outputDataKind values in row-major order. Decode it and reshape it to shape.

Returns

A new MultimodalMultiVectorEmbedding instance for mocking.

Applies to