Introduction

This documentation describes how to go about deploying and configuring PA System Models. Unlike PA Custom Models that require manual configuration and maintenance, PA System Models are recommended and automatically maintained built-in models. An admin only needs to provide the LLM’s API token and API URL address.

All PA System Models are Azure OpenAI LLM configurations that are GDPR compliant if applying suggested default values.

These are the LLMs currently supported by the PA System Models:

• gpt-4o-mini

• gpt-4o

• dall-e-3

In addition, the text embedding model text-embedding-ada-002 is also supported.

Prerequisites

The following prerequisites must be fullfilled before starting on configuring the PA models:

• An Azure subscription - An active Azure Plan or Pay-as-you-go subscription is required. If one does not have one, one can sign up for one at https://portal.azure.com.

• A Sufficient large Azure OpenAI quota - Using one of the methods described in PA - Azure OpenAI Prerequisites, ensure that the quota is sufficiently large. If needed, request more following the instructions in PA - How to Request Azure OpenAI Quotas. See Azure OpenAI in Azure AI Foundry Models quotas and limits - Azure AI services for details.

• The Personal Assistant Administrator Role - Consult the Managing Users and Roles section of the Ayfie Locator Installation Guide for how to be made a Personal Assistant Admin.

Configuration

First-time Deployment

Obtain the Azure ARM Template

Acquire the template from the application by clicking the ARM Template link shown red circled below.

For on-premises Saga 7 and later deployments, the ARM Template can also be accessed using the URL provided below. No special role or even user access to the search application is necessary. The <SEARCH PAGE FQDN> is in Saga documentation known as the Gateway Hostname.

Open

Deploy models

Use the Deploy from Custom Template feature to deploy the Azure OpenAI models. It provides the preferred configuration parameters, such as the OpenAI model quota and the model deployment type. It is recommended to use Sweden Central as the deployment region due to the model's capacity and availability.

Deploy ARM template

1. Select Build your own template in the editor

   Open

   

2. Replace the selected text with ARM template definition and press Save

   Open

   

3. Fill in the form

   Open

   

4. If the deployment fails with error details listing the InsufficientQuota code, you have less available quota for the model(s) specified in the error than expected by the ARM template. In this case, please follow the instructions found in PA - How to Request Azure OpenAI Quotas

   Open

   

Obtain the API Token and the API URL

Once the deployment is completed, navigate to the selected resource group and the newly created Azure OpenAI resource.
Under Resource Management / Keys and Endpoint one will find the API token and the API URL

Open

Insert System Models

Provide the API token and the API URL and press Save. The application is now configured with recommended models and can be tested.

Open

Update Existing Deployment

Detection of New System Model Configuration

Once the system detects there is a new version of the recommended system model configuration, the Update system models option becomes available.

Open

Redeploying Models

The system will detect if the new recommended configuration requires a redeployment of the Azure OpenAI models. In that case, the red circled message seen below will be shown:

Open

1. Remove existing models

   1. Navigate to Azure AI services | Azure Open AI and identify the Azure Open AI resource.

   2. Delete the resource - that will only result in a soft delete.

   3. Select Managed deleted resources and purge the deleted resource.

      Open

      

2. Follow the instructions in “Deploy Models” to deploy new models

3. Follow the instructions in “Obtain API token, API URL and update the configuration” to finalize the model configuration.

Models Restoration

The application also provides the option to restore the default system model configuration. To do so, click the Restore system model option and provide the API token and the API URL in the pop up window before clicking Save.

Open