Azure Foundry model catalog vectorizer

Important

This vectorizer is in public preview under Supplemental Terms of Use. To use this feature, we recommend the latest preview version of Indexes - Create Or Update (REST API).

The Azure AI Foundry model catalog vectorizer connects to an embedding model that was deployed via the Azure AI Foundry model catalog to an Azure Machine Learning endpoint. Your data is processed in the Geo where your model is deployed.

Vectorizer parameters

Parameters are case sensitive. The parameters you use depend on what authentication your model provider requires, if any.

Parameter name	Description
`uri`	(Required for key authentication) The target URI of the serverless deployment from the Foundry model catalog or the scoring URI of the AML online endpoint. Only the HTTPS URI scheme is allowed.
`key`	(Required for key authentication) The API key of the model provider.
`resourceId`	(Required for token authentication) The Azure Resource Manager resource ID of the model provider. For an AML online endpoint, use the `subscriptions/{guid}/resourceGroups/{resource-group-name}/Microsoft.MachineLearningServices/workspaces/{workspace-name}/onlineendpoints/{endpoint_name}` format.
`region`	(Optional for token authentication) The region in which the model provider is deployed. Required if the region is different from the region of the search service.
`timeout`	(Optional) The timeout for the HTTP client making the API call. It must be formatted as an XSD "dayTimeDuration" value (a restricted subset of an ISO 8601 duration value). For example, `PT60S` for 60 seconds. If not set, a default value of 30 seconds is chosen. The timeout can be set to a maximum of 230 seconds and a minimum of 1 second.

What authentication parameters to use

The Azure Foundry model catalog vectorizer provides two authentication options:

Key-based authentication. You provide a static key to authenticate scoring requests from the vectorizer. Set the uri and key parameters for this connection.
Token-based authentication. The Foundry hub-based project or AML online endpoint is deployed using token-based authentication. The Azure AI Search service must have a managed identity and a role assignment on the model provider. The vectorizer then uses the search service identity to authenticate against the model provider, with no static keys required. The search service identity must have the Owner or Contributor role. Set the resourceId parameter, and if the search service is in a different region from the model provider, set the region parameter.

Supported vector query types

Which vector query types are supported by the Azure Foundry model catalog vectorizer depends on the modelName that is configured.

Embedding model	Supports `text` query	Supports `imageBinary` query
Cohere-embed-v3-english	X	X
Cohere-embed-v3-multilingual	X	X
Cohere-embed-v4	X	X

Expected field dimensions

The expected field dimensions for a vector field configured with a Azure Foundry model catalog vectorizer depend on the modelName that is configured.

`modelName`	Expected dimensions
Cohere-embed-v3-english	1024
Cohere-embed-v3-multilingual	1024
Cohere-embed-v4	256-1536

Sample definition

Suggested model names in the Azure AI Foundry model catalog consist of the base model plus a random three-letter suffix. The name of your model will be different from the one shown in this example.

"vectorizers": [
    {
        "name": "my-model-catalog-vectorizer",
        "kind": "aml",
        "amlParameters": {
            "uri": "https://Cohere-embed-v3-multilingual-hin.chinaeast.models.ai.azure.cn",
            "key": "aaaaaaaa-0b0b-1c1c-2d2d-333333333333",
            "timeout": "PT60S",
            "modelName": "Cohere-embed-v3-multilingual-hin",
            "resourceId": null,
            "region": null,
        },
    }
]