Azure GPT-4o Deployment and Provisioning Explained

Azure GPT-4o deployment and provisioning is a straightforward process that allows you to quickly set up and manage your models.

You can deploy GPT-4o models as a container in Azure Kubernetes Service (AKS), which provides a scalable and highly available environment for your models.

Azure GPT-4o supports automated provisioning, which means you can easily set up and configure your models with minimal manual intervention.

Provisioning a GPT-4o model in Azure involves creating a resource group, selecting a region, and choosing a pricing tier, among other steps.

Azure GPT-4o deployment is a crucial step in utilizing this powerful AI model. You must create an Azure OpenAI Resource before deploying a model.

To deploy a model, you can use the Deployment APIs, which allow you to specify the model you wish to use. The model you choose will determine the type of deployment you can create. For example, gpt-4o-2024-08-06 and gpt-4o-mini-2024-07-18 models are supported for data zone standard and global provisioned deployments.

GPT-4o mini is available for standard and global standard deployment in the East US and Sweden Central regions. It's also available for global batch deployment in East US, Sweden Central, and West US regions.

Here's a summary of the supported deployment types for GPT-4o:

Note that the specific models and deployment types available may change over time, so be sure to check the latest information on model availability and deployment options.

Provisioned deployments allow you to reserve model processing capacity for high and predictable throughput using Azure global infrastructure.

GPT-4o mini is available for provisioned deployments in Canada East, East US, East US2, North Central US, and Sweden Central regions.

GPT-4o mini provisioned deployments support both text and image/vision inference requests, but it's essential to check the model regional availability for the latest information.

GPT-4o provisioned deployments, on the other hand, are available for both standard and provisioned deployments and accept both text and image/vision inference requests.

The following models are available for provisioned deployments:

Keep in mind that the provisioned version of GPT-4 Turbo doesn't support image/vision inference requests, only text input.

GPT-4o mini is significantly smarter than GPT-3.5 Turbo, scoring 82% on Measuring Massive Multitask Language Understanding (MMLU) compared to 70%, and is more than 60% cheaper.

The o1 series models are currently in preview, which means they're still being tested and refined. This also means they don't have all the features available in other models, such as image understanding and structured outputs.

For many tasks, the generally available GPT-4o models might still be more suitable. This is because they have more features and capabilities compared to the o1 series models.

The o1 series models are missing some features that are available in the latest GPT-4o model. If you're looking for image understanding or structured outputs, you might want to consider using a different model.

For information on global standard quota, consult the quota and limits page.

Microsoft has announced that GPT-4o is available within Azure OpenAI Service now. This means you can start using the new model right away.

If you're already using Azure OpenAI, you can create a new deployment of GPT-4o in the Azure OpenAI Studio starting today.

Fine-tuning with Azure GPT-4o has become Generally Available (GA), allowing you to add images to your JSONL training data.

Vision fine-tuning enables you to include image inputs within your training data, either as URLs or base64 encoded images. This feature is now available for Azure OpenAI in public preview in North Central US and Sweden Central.

GPT-4o fine-tuning is now available in public preview, and the latest GA release of GPT-4 Turbo has been announced.

Fine-tuning billing is now token-based, resulting in a significant cost reduction for some training runs. This change makes estimating fine-tuning costs much easier.

Here are the key updates:

Fine-tuning with GPT-4o is now Generally Available (GA), allowing you to add images to your JSONL training data. This is a big deal for anyone looking to improve the visual aspect of their AI models.

GPT-4o fine-tuning is also available in public preview in North Central US and Sweden Central, giving you more options for testing and refining your models.

You can now include images in your training data as URLs or as base64 encoded images, making it easier to incorporate visual inputs into your AI models. This is a game-changer for anyone working with image-based data.

GPT-4 fine-tuning is available in public preview, offering more flexibility and options for fine-tuning your models.

Here are some key updates to the GPT-4 fine-tuning API:

Fine-tuning now supports multi-turn chat training examples, making it easier to create more realistic and conversational AI models. Try out Whisper by following a quickstart to see it in action.

As of September 19, the default user interface for Azure OpenAI Studio will change, and you'll no longer see the legacy studio UI by default.

You can still access the old experience for a couple of weeks by using the Switch to the old look toggle in the top bar of the UI.

The new experience is being actively monitored, and filling out the feedback form will help improve it for you and others.

If you do switch back to the legacy Studio UI, it's helpful to let us know why by filling out the feedback form.