Atlas AI - Upgrading Existing Models in your Azure OpenAI service (Atlas 6.0)

NOTE: This article makes the assumption that you already have an operative Azure OpenAI service in your Azure subscription that is being used to work with the Atlas AI features, and making use of models that are not compatible with Atlas 6.0 - Atlas 6.0 is compatible with GPT-4o and GPT-4o-mini models. If you have not created the Azure OpenAI service in your subscription yet, please refer to the related articles that you can find at the bottom of this article to create a new one.

Atlas comes with a set of AI features that make use of the OpenAI service provided by Microsoft in Azure.

Prior to Atlas 6.0, during the deployment of Atlas and in case you would like to use the Atlas AI features, you were given the option to automatically create the Azure OpenAI service during the deployment or to bring your own service created on your own. The OpenAI service was configured with models that are no longer compatible with Atlas 6.0. Instead, Atlas 6.0 is leveraging the GPT-4o and GPT-4o-mini models' capabilities.

Due to this situation, you need to update your models so the Atlas AI features work as expected. In this article, we will guide you on how to create the new GPT models that are compatible with Atlas AI features in Atlas 6.0, and what options you can use among the ones available in the Azure OpenAI service for this purpose.

IMPORTANT: Before proceeding, please make sure to review the region where your Azure OpenAI service is located to verify whether or not is compatible with the new GPT-4o and GPT-4o-mini models. For this version of Atlas, we specifically require the use of GPT-4o (version 2024-08-06) and GPT-4o-mini (version 2024-07-18) model versions, so please make sure to check the compatible regions for these models here: Azure OpenAI Service models - Model summary table and region availability

We also offer a separate article that adds a bit more of context and information related to the compatible regions. You can find it at the bottom of this article.

To follow the steps in the next sections, you must have assigned the RBAC "Owner" role over the Atlas' resource group in your Azure subscription and over the Azure OpenAI service resource that you are using with Atlas.

Create and configure Azure OpenAI models

In this section, you will see the models that you must create to make the Azure OpenAI service compatible with the AI features that come along with Atlas 6.0. Three models will be created in this section:

• gpt-4o
• gpt-4o-mini

Creating a model in your Azure OpenAI service is also known as "creating a new deployment", which is really like creating an instance of a given model from the ones available in your Azure OpenAI service. This means that you may have, for example, two deployments (or instances) of the model "gpt-4o" or "gpt-4o-mini" that may coexist because you may be working with different products that leverage your Azure OpenAI service created in your Azure subscription.

The next steps will provide the guidelines to create these deployments and configure them to make them compatible with the use of Atlas AI. You will end up with two new deployments of the models that will be compatible with our Atlas AI features, apart from the "text-embedding-ada-002" model that you should have already and must keep from the previous deployments (it is crucial for Atlas AI to work properly, for content synchronization tasks), so these will be the ones used by Atlas when configuring it during a deployment to work with your Azure OpenAI service:

1. Open your browser and go to Azure AI Foundry
2. Click on the "Sign In" button at the top-right corner, if you are not already signed in, with your work account
   
   
   
3. Once in the Home dashboard of the "Azure AI Foundry | Azure OpenAI Service" portal, make sure to select the Azure OpenAI Service resource you want to work with using the select dropdown at the top as shown in the screenshot
   
4. Now, under "Shared resources", click on "Deployments" in the left menu, and then click on the "+ Deploy model" button:
   
   
   
5. A new window will appear showing up all the available models for your Azure OpenAI to create a new deployment. So, first, we need to select the model we want and then click on Confirm:
   
   
   
6. A new form will appear to create the deployment with the configuration you want. Because you must create two deployments (repeat this process as needed), one per each of the models required, follow the next tables and screenshots to fill in the information required for each field, and click on "Create" each time to save your deployment.
   
   IMPORTANT: you should use the same values for the "Deployment name" and "Model version" from the tables below for your deployment; otherwise, Atlas AI may not work when using your Azure OpenAI resource. However, for the rest of the fields, the values that we suggest are just recommendations to get a good performance.
   
   Please, pay special attention to the "Deployment type" field because this option cannot be edited after creating the deployment. If you need a different deployment type because your data is geo-sensitive or if you need more information before continuing, please check the documentation here: Azure OpenAI deployment types
   
   WARNING: A bug has recently been detected in the User Interface of this form. When changing the model version, the "Model Name" field reset to default, so please leave the "Model Name" field to the end just before applying to avoid misconfigurations.
   
   
   1. Model gpt-4o:
      
      

      Deployment name	atlas-gpt-4o
      Deployment Type	Global Standard
      Deployment Details
      Model version	2024-08-06
      Tokens per Minute Rate Limit (thousands)	150K
      Content Filter	DefaultV2

      
      
      
      
      
   2. Model gpt-4o-mini:
      
      

      Deployment name	atlas-gpt-4o-mini
      Deployment Type	Global Standard
      Deployment Details
      Model version	2024-07-18 (Default)
      Tokens per Minute Rate Limit (thousands)	300K
      Content Filter	DefaultV2

      
      
7. Once you create the deployments, your "Deployments" summary should look like the screenshot below listing, at least, the next models deployments (notice you might have other deployments in your Azure OpenAI resource):
   
   1. atlas-embedding (text-embedding-ada-002)
   2. atlas-gpt-4o (gpt-4o)
   3. atlas-gpt-4o-mini (gpt-4o-mini
      
      
      

NOTE: Once the Atlas deployment is complete, and taking this as a post-deployment step, you can consider deleting the old Azure OpenAI model used by Atlas in previous versions. In the Deployments list, the name of the model deployment to delete should be "atlas-gpt", based on gpt-4-32k (0613).

Now that you have completed all the steps to upgrade the Azure OpenAI models, remember to take note of the name of your Azure OpenAI resource to make use of it when updating Atlas in your Azure subscription. Additionally, from now on, if you need to modify the technical configuration of the OpenAI models for any reason at any moment, please check the related article referred to this matter down below.

Related articles:

• Atlas AI: Review access to Azure OpenAI in your Azure subscription (and request access to Microsoft)
• Atlas AI - Atlas 6.0: Compatible regions for the Azure OpenAI service
• Atlas AI - Atlas 6.0: Manual creation of Azure OpenAI in your subscription
• Atlas AI - Atlas 6.0: Changing the configuration of the Open AI models for Atlas