Azure OpenAI GPT-4o Structured Outputs with .NET 9

The goal of this sample is to demonstrate how to use the new JsonSchemaExporter type in .NET 9 to simplify the creation of JSON schemas for Azure OpenAI GPT-4o structured outputs.

How it works

JsonSchemaExporter is simply a built in .NET mechanism to generate valid JSON schema objects from .NET types.

For the .NET OpenAI SDK however, the JSON schema must be in a specific format to be considered valid. On top of the default generated schemas using the exporter, the current known additional rules must be applied:

• The base type must be an object only by setting the TreatNullObliviousAsNonNullable = true option in the JsonSchemaExporterOptions.
• Any properties that are of type object must additionally include the following properties:
  • additionalProperties set to false.
  • required which is an array of all the keys in the object.

The StructuredOutputsExtensions class provides a generic CreateJsonSchemaFormat method that can be passed a model object type and will return the ChatResponseFormat object that can be used by the OpenAI SDK. This method will apply the additional rules to the generated schema using a Transform function.

Note

For complex object nesting where there are self-referencing types, this implementation does not support. It would be possible to provide this capability by simplifying the nested classes as $defs at the root of the schema, and then referencing them in the properties with '$ref': '#/$defs/ClassName'.

In addition to the generic CreateJsonSchemaFormat method, a generic CompleteChat extension is provided for the ChatClient that will automatically deserialize the structured output into the expected model object type.

Example

class Family
{
    public List<Person> Parents { get; set; }
    public List<Person>? Children { get; set; }

    class Person
    {
        public string Name { get; set; }
        public int Age { get; set; }
    }
}

ChatCompletionOptions options = new()
{
    ResponseFormat = StructuredOutputsExtensions.CreateJsonSchemaFormat<Family>("family", jsonSchemaIsStrict: true),
    MaxOutputTokenCount = 4096,
    Temperature = 0.1f,
    TopP = 0.1f
};

List<ChatMessage> messages =
[
    new SystemChatMessage("You are an AI assistant that creates families."),
    new UserChatMessage("Create a family with 2 parents and 2 children.")
];

var family = chatClient.CompleteChat<Family>(messages, options);

Running the sample

• Install the latest .NET 9 SDK.
• Install PowerShell Core.
• Install the Azure CLI.
• Install Visual Studio Code.

Additionally, you will require:

• An Azure subscription. If you don't have an Azure subscription, create an account.

To setup a local development environment, follow these steps:

Note

For the most optimal sample experience, it is recommended to deploy the necessary infrastructure in a region that supports GPT-4o version 2024-08-06. Find out more about region availability for the GPT-4o model.

az login

./Setup-Environment.ps1 -DeploymentName <UniqueDeploymentName> -Location <AzureRegion> -SkipInfrastructure $false

The script will deploy the following resources to your Azure subscription:

• Azure AI Services, a managed service for all Azure AI Services, including Azure OpenAI.
  • Note: GPT-4o will be deployed as Global Standard with 10K TPM quota allocation. This can be adjusted based on your quota availability in the main.bicep file.

Note

Resources are secured by default with Microsoft Entra ID using Azure RBAC. Your user client ID will be added with the necessary least-privilege roles to access the resources created.

After the script completes, a .env file will be crated in the src/StructuredOutputs folder and you can run the sample project by following these steps:

cd src/StructuredOutputs
dotnet run

Resources

• Azure OpenAI GPT-4o Structured Outputs Supported Schemas and Limitations
• How to use chat completions with structured outputs in OpenAI .NET SDK
• .NET 9 JSON Schema Exporter

License

This project is licensed under the MIT License - see the LICENSE file for details.