Creates an embedding vector from the input text, image, or text and image.

post

https://integrate.api.nvidia.com/v1/embeddings

Recent Requests

Time	Status	User Agent
Retrieving recent requests…

Loading…

Body Params

input

required

Input to embed. Each item is text, one image, or text followed by one image. Send an image as a base64 data URL, for example data:image/png;base64,<data>, or as an <img src="data:image/png;base64,<data>"> tag. PNG and JPEG images are supported, with one image per item. Max length is 10240 tokens.

model

string

required

Defaults to nvidia/nemotron-3.5-embed-8b

ID of the embedding model.

input_type

string

enum

nvidia/nemotron-3.5-embed-8b operates in passage or query mode, and thus requires the input_type parameter. passage is used when generating embeddings during indexing. query is used when generating embeddings during querying. Inputs that contain an image must use passage. It is very important to use the correct input_type. Failure to do so will result in large drops in retrieval accuracy.

Allowed:

modality

string

enum

The modality of each input item: text, image, or text_image. Give one value for all items, or one value per item. If omitted, the modality is inferred from the input.

Allowed:

encoding_format

string

enum

Defaults to float

The format to return the embeddings in.

Allowed:

truncate

string

enum

Defaults to END

Specifies how inputs longer than the maximum token length (10240) are handled. END discards the end of the text until the input fits. NONE returns an error when the input exceeds the maximum. START is not supported. The image and the passage prefix must always fit; an image block that alone exceeds the maximum is rejected in every mode.

Allowed:

dimensions

integer

1 to 4096

Number of dimensions of the returned embedding, from 1 through the native dimension of 4096. Shortened vectors are L2-normalized again. Omit to receive the native 4096-dimensional embedding.

embedding_type

string

enum

Defaults to float

Output representation of each embedding value. Quantized representations pack several dimensions into each output element.

Allowed:

longest_edge

integer

256 to 1540

Maximum image size before patch-grid alignment. Larger images are downsized while preserving the aspect ratio; smaller images are not enlarged. Omit to use the model default of 1120. Use the same value when indexing and querying.

user

string

length ≤ 512

Optional identifier for the end user. If provided it must be non-blank and at most 512 characters. It does not select the model or authorize the request.

Responses

400

Bad request – invalid parameters.

401

Unauthorized – invalid or missing API key.

404

Not found – unknown model.

413

Payload too large – the request or an image exceeds the size limit.

422

Unprocessable entity – validation error.

500

Internal server error.

503

Service unavailable.

Language

Credentials

Bearer

Loading…

Response

Click Try It! to start a request and see the response here! Or choose an example:

application/json

200Successful response containing embedding vectors.