Provision GPUs and enable Vertex AI pre-trained APIs

Each Vertex AI service requires different computing resources, including graphics processing units (GPUs), as described in Capacity planning and computing requirements. This page describes how to provision GPU resources to run artificial intelligence (AI) and machine learning (ML) workloads in a GPU environment. This page also walks you through enabling the Vertex AI pre-trained APIs on Google Distributed Cloud (GDC) air-gapped appliance so you can start implementing Vertex AI capabilities.

Most tasks to configure GPU resources and enable or deactivate Vertex AI pre-trained APIs require administrator access. If you lack the necessary permissions, ask your administrator to enable GPUs and the Vertex AI pre-trained APIs on your behalf.

Vertex AI on GDC air-gapped appliance includes three APIs, one for each of its pre-trained models. To learn more about these pre-trained models, see the following documentation:

• Optical Character Recognition (OCR): Learn about character recognition features.
• Speech-to-Text: Learn about speech recognition features.
• Vertex AI Translation: Learn about translation features.

Use the GDC console to enable, deactivate, and view the endpoints of the Vertex AI pre-trained APIs.

Before you begin

To get the permissions that you need to enable pre-trained APIs, ask your Organization IAM Admin or Project IAM Admin to grant you the AI Platform Admin (ai-platform-admin) role in the project namespace.

For information about this role, see Prepare IAM permissions. To learn how to grant permissions to a subject, see Grant and revoke access.

Follow these steps to provision GPUs before enabling the pre-trained APIs:

1. Set up a project to use Vertex AI.
2. Ensure that your project has the adequate ingress communication configured.
3. Manage GPU workloads.
4. Sign in to the GDC console. If you can't sign in, see Recover device if identity provider is unavailable.

Enable pre-trained APIs

You can enable the OCR, Speech-to-Text, and Vertex AI Translation pre-trained APIs using the GDC console.

Caution: GDC air-gapped appliance includes a single NVIDIA A100 80GB GPU. This limits the device to running only one GPU-dependent Vertex AI pre-trained API at a time. If you attempt to enable both Vertex AI Translation and OCR, the second API will fail to enable with an error message indicating insufficient GPU resources. You can, however, run Speech-to-Text alongside either Vertex AI Translation or OCR, as Speech-to-Text only requires CPU resources.

After meeting the prerequisites, follow these steps to enable the pre-trained APIs:

1. Sign in to the GDC console.
2. In the navigation menu, click Vertex AI > Pre-trained APIs.

3. On the Pre-trained APIs page, click Enable on a specific service to enable that API.

   Note: The GDC console doesn't display the buttons to enable the pre-trained APIs if you don't have the correct permissions.

4. In the confirmation dialog, click Enable. A progress message displays.

The enablement duration varies. It might take between 15 and 45 minutes to finish.

If you want to view the status of the pre-trained APIs, view the service status and endpoints.

Note: You can't enable an API while you wait for another API to finish deactivating.

Deactivate pre-trained APIs

You can deactivate the OCR, Speech-to-Text, and Vertex AI Translation pre-trained APIs using the GDC console.

After meeting the prerequisites, follow these steps to deactivate the pre-trained APIs:

1. Sign in to the GDC console.
2. In the navigation menu, click Vertex AI > Pre-trained APIs.

3. On the Pre-trained APIs page, click Disable on a specific service to deactivate that API.

   Note: The GDC console doesn't display the buttons to deactivate the pre-trained APIs if you don't have the correct permissions.

4. In the confirmation dialog, enter disable in the text field to confirm that you want to take that action. Then, click Disable. A progress message displays.

If you want to view the status of the pre-trained APIs, view the service status and endpoints.

Note: You can't deactivate an API while you wait for another API to finish enabling.