Embeddings API

This documentation is valid for:

Gets a vector representation of a given input that can be easily consumed by machine learning models and algorithms.

Check the generic variables needed to use the API and the supported models.

Endpoints

Method	Path	Description
POST	/embeddings	Creates an embedding vector representing the input

POST/embeddings

Creates an embedding vector representing the input provided.

Parameter	Type	Description
input	string or array	Input to embed, encoded as a string⁽⁴⁾. To embed multiple inputs in a single request, pass an array of strings. The input must not exceed the max input tokens for the model and cannot be an empty string
model	string	provider/modelId to use
encoding_format⁽¹⁾	string	The format to return the embeddings. It can be either float (default) or base64 (optional)
dimensions⁽¹⁾	integer	The number of dimensions the resulting output embeddings should have. Only supported in text-embedding-3* and later models (optional)
user⁽³⁾	string	A unique identifier representing your end-user
input_type⁽²⁾	string	Defines how the input data will be used when generating embeddings (optional)
timeout	integer	The maximum time, in seconds, to wait for the API to respond. Defaults to 600 seconds

Check the valid providers in the Providers column of the Supported Chat Models.

⁽¹⁾: Only supported by openAI provider. The maximum input token is 8191 (roughly ten standard pages of text).

⁽²⁾: Check if the selected embeddings model supports this option.

⁽³⁾: openAI specific.

⁽⁴⁾: If the embedding model support image embeddings; pass in a base64 encoded image string to the input parameter with the associated mime type (check the samples section).

You can use the X-Saia-Cache-Enabled (true or false default) to cache the embeddings for the model; it applies by Organization/Project.

{
    "model": "openai/text-embedding-3-small",
    "input" : "Help me with Globant Enterprise AI!"
}

{
    "model": "openai/text-embedding-3-small",
    "input" : ["Help me with", "Globant Enterprise AI!"]
}

Response

The expected response is similar to the following:

{
    "model": "text-embedding-3-small",
    "data": [
        {
            "embedding": [...],
            "index": 0,
            "object": "embedding"
        },...
    ],
    "object": "list",
    "usage": {
        "completion_tokens": number,
        "prompt_tokens": number,
        "total_tokens": number,
        "completion_tokens_details": null
    }
}

Samples

OpenAI using one string
OpenAI and cache
VertexAI and optional parameters
Nvidia and optional parameters
Bedrock text and Image content

curl --location '$BASE_URL/embeddings' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer $SAIA_PROJECT_APITOKEN' \
--data '{    
    "model": "openai/text-embedding-3-small",
    "input" : "Help me with Globant Enterprise AI!"
}'

curl --location '$BASE_URL/embeddings' \
--header 'X-Saia-Cache-Enabled: true' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer $SAIA_PROJECT_APITOKEN' \
--data '{    
    "model": "openai/text-embedding-3-small",
    "input" : "Help me with", "Globant Enterprise AI!"   
}'

curl --location '$BASE_URL/embeddings' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer $SAIA_PROJECT_APITOKEN' \
--data '{
    "model": "vertex_ai/text-embedding-004",
    "input_type": "SEMANTIC_SIMILARITY", // RETRIEVAL_QUERY, RETRIEVAL_DOCUMENT and other options
    "input" : "Help me with", "Globant Enterprise AI!"
}'

curl --location '$BASE_URL/embeddings' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer $SAIA_PROJECT_APITOKEN' \
--data '{
    "model": "nvidia/nv-embedqa-e5-v5",
    "input" : "identify", "yourself!",
    "input_type": "passage" // passage or query
}'