Azure OpenAI Provider

The Azure OpenAI provider allows you to use Azure OpenAI Service through OneLLM’s unified interface.

Configuration

The Azure provider requires a configuration file (azure.json) that contains your Azure OpenAI deployment details. The provider will look for this file in the following locations:

1. Path specified via azure_config_path parameter when initializing the provider
2. Path specified via AZURE_OPENAI_CONFIG_PATH environment variable
3. Default location: azure.json in the project root

Configuration File Format

{
    "key1": "your-primary-api-key",
    "key2": "your-secondary-api-key",
    "region": "uksouth",
    "endpoint": "https://your-resource.openai.azure.com/",
    "deployment": {
        "gpt-4o-mini": {
            "endpoint": "https://your-resource.openai.azure.com/",
            "model_name": "gpt-4o-mini",
            "deployment": "gpt-4o-mini-deployment",
            "subscription_key": "deployment-specific-key",
            "api_version": "2024-12-01-preview"
        },
        "o4-mini": {
            "endpoint": "https://another-resource.openai.azure.com/",
            "model_name": "o4-mini",
            "deployment": "o4-mini-deployment",
            "subscription_key": "another-deployment-key",
            "api_version": "2024-12-01-preview"
        }
    }
}

Configuration Fields

• key1, key2: Primary and secondary API keys for your Azure OpenAI resource
• region: Azure region where your resource is deployed
• endpoint: Base endpoint URL for your Azure OpenAI resource
• deployment: Object containing deployment-specific configurations
  • Each deployment can have its own endpoint, API key, and API version
  • This allows you to use multiple Azure OpenAI resources and deployments

Usage

Basic Usage

from onellm import Client

# Initialize client (will auto-detect azure.json)
client = Client()

# Use Azure OpenAI with the azure/ prefix
response = await client.chat.completions.create(
    model="azure/gpt-4o-mini",
    messages=[
        {"role": "user", "content": "Hello, Azure!"}
    ]
)

Direct Provider Usage

from onellm.providers import get_provider

# Initialize with custom config path
azure = get_provider("azure", azure_config_path="/path/to/azure.json")

# Use the provider directly
response = await azure.create_chat_completion(
    messages=[{"role": "user", "content": "Hello!"}],
    model="gpt-4o-mini"
)

Environment Variables

You can set the configuration file path via environment variable:

export AZURE_OPENAI_CONFIG_PATH=/path/to/your/azure.json

Supported Features

The Azure provider supports all features available in Azure OpenAI Service:

Chat Completions

• All GPT models deployed in your Azure resource
• Streaming support
• Function calling
• JSON mode (for supported models)

Vision Models

• GPT-4 Vision models (gpt-4v, gpt-4o)
• Image input in chat messages

Audio

• Whisper models for transcription and translation
• Text-to-speech with various voices

Image Generation

• DALL-E 2 and DALL-E 3 models
• Various image sizes and quality settings

Embeddings

• Text embedding models (text-embedding-ada-002, etc.)

Model Naming

When using Azure OpenAI through OneLLM, prefix your model names with azure/:

• azure/gpt-4o-mini
• azure/gpt-4-turbo
• azure/o4-mini
• azure/whisper-1
• azure/dall-e-3

The model name after the prefix should match either:

1. A deployment name defined in your azure.json configuration
2. A standard Azure OpenAI deployment name

Advanced Configuration

Multiple Deployments

The Azure provider supports using multiple deployments with different configurations:

{
    "deployment": {
        "fast-model": {
            "endpoint": "https://eastus-resource.openai.azure.com/",
            "deployment": "gpt-35-turbo",
            "subscription_key": "eastus-key",
            "api_version": "2024-12-01-preview"
        },
        "powerful-model": {
            "endpoint": "https://westus-resource.openai.azure.com/",
            "deployment": "gpt-4-turbo",
            "subscription_key": "westus-key",
            "api_version": "2024-12-01-preview"
        }
    }
}

API Versions

You can specify different API versions for different deployments or set a default:

azure = get_provider("azure", api_version="2024-12-01-preview")

Error Handling

The Azure provider handles Azure-specific errors and maps them to OneLLM’s standard error types:

• AuthenticationError: Invalid API key or authentication issues
• RateLimitError: Rate limit exceeded
• InvalidRequestError: Invalid request parameters
• ResourceNotFoundError: Deployment not found

Best Practices

1. Use deployment-specific configurations for better control over routing and API versions
2. Store your azure.json securely and never commit it to version control
3. Use secondary keys for key rotation without downtime
4. Monitor usage through Azure Portal to track costs and performance
5. Set appropriate timeouts for long-running operations

Limitations

• File upload/download operations are not supported (Azure uses blob storage separately)
• Some OpenAI-specific features may not be available in Azure OpenAI
• API versions and available features depend on your Azure region and deployment