Skip to main content

​
Overview

Azure OpenAI Service provides access to OpenAI models through Microsoft Azure with enterprise features, compliance, and regional deployment. Model Class: OpenAIChatModel (uses Azure provider)

​
Authentication

​
Environment Variables

export AZURE_OPENAI_ENDPOINT="https://your-resource.openai.azure.com/"
export AZURE_OPENAI_API_KEY="..."
export OPENAI_API_VERSION="2024-02-15-preview"

​
Using infer_model

from upsonic import infer_model

model = infer_model("azure/gpt-4o")

​
Manual Configuration

from upsonic.models.openai import OpenAIChatModel, OpenAIChatModelSettings
from upsonic.providers.azure import AzureProvider

# Using provider
provider = AzureProvider()
model = OpenAIChatModel(
    model_name="gpt-4o",
    provider=provider
)

# Or specify provider string
model = OpenAIChatModel(
    model_name="gpt-4o",
    provider="azure"
)

​
Examples

​
Basic Usage

from upsonic import Agent, Task, infer_model

model = infer_model("azure/gpt-4o")
agent = Agent(model=model)

task = Task("Analyze customer feedback data")
result = agent.do(task)

​
With Custom Deployment

from upsonic.models.openai import OpenAIChatModel
from openai import AsyncAzureOpenAI

# Custom deployment name
client = AsyncAzureOpenAI(
    azure_endpoint="https://your-resource.openai.azure.com/",
    api_key="your-api-key",
    api_version="2024-02-15-preview"
)

from upsonic.providers.azure import AzureProvider
provider = AzureProvider(client=client)

model = OpenAIChatModel(
    model_name="your-deployment-name",  # Your deployment name in Azure
    provider=provider
)

​
With Managed Identity

from azure.identity import DefaultAzureCredential
from openai import AsyncAzureOpenAI

# Using managed identity (no API key needed)
credential = DefaultAzureCredential()
token = credential.get_token("https://cognitiveservices.azure.com/.default")

client = AsyncAzureOpenAI(
    azure_endpoint="https://your-resource.openai.azure.com/",
    azure_ad_token=token.token,
    api_version="2024-02-15-preview"
)

from upsonic.providers.azure import AzureProvider
provider = AzureProvider(client=client)

model = OpenAIChatModel(
    model_name="gpt-4o",
    provider=provider
)

​
With Virtual Network

from upsonic.models.openai import OpenAIChatModel, OpenAIChatModelSettings

# Azure resources can be behind VNet
settings = OpenAIChatModelSettings(
    max_tokens=2048,
    timeout=120.0
)

model = OpenAIChatModel(
    model_name="gpt-4o",
    provider="azure",
    settings=settings
)

​
With Content Filtering

from upsonic import Agent, Task, infer_model

# Azure includes built-in content filtering
model = infer_model("azure/gpt-4o")
agent = Agent(model=model)

task = Task("Generate safe content for children")
result = agent.do(task)

​
Prompt Caching

Azure OpenAI supports the same prompt caching capabilities as OpenAI: 
from upsonic import Agent, Task, infer_model

# Long system prompt will be cached
system_prompt = """
You are an enterprise assistant with access to:
- Company policies and procedures
- HR handbook
- IT security guidelines
... (extensive context)
"""

model = infer_model("azure/gpt-4o")
agent = Agent(
    model=model,
    system_prompt=system_prompt
)

# First request builds cache
task1 = Task("What's the vacation policy?")
result1 = agent.do(task1)

# Subsequent requests use cache
task2 = Task("How do I report a security incident?")
result2 = agent.do(task2)

​
Cache Benefits

  • Cost Reduction: 50% savings on cached input tokens
  • Latency: 2-4x faster with cache hits
  • Automatic: No configuration needed
  • Minimum Size: 1024 tokens recommended

​
Model Parameters

​
Base Settings

ParameterTypeDescriptionDefault
max_tokensintMaximum tokens to generateModel-specific
temperaturefloatSampling temperature (0.0-2.0)1.0
top_pfloatNucleus sampling1.0
seedintRandom seedNone
stop_sequenceslist[str]Stop sequencesNone
presence_penaltyfloatToken presence penalty0.0
frequency_penaltyfloatToken frequency penalty0.0
logit_biasdict[str, int]Token likelihood modifierNone
parallel_tool_callsboolAllow parallel toolsTrue
timeoutfloatRequest timeout (seconds)600

​
Azure-Specific Features

Azure OpenAI includes additional enterprise features:
  • Content Filtering: Automatic content safety checks
  • Managed Identity: Azure AD authentication
  • Private Endpoints: VNet integration
  • Data Residency: Regional deployment
  • Compliance: SOC 2, HIPAA, ISO certifications

​
Example Configuration

from upsonic.models.openai import OpenAIChatModel, OpenAIChatModelSettings

settings = OpenAIChatModelSettings(
    max_tokens=4096,
    temperature=0.7,
    top_p=0.9,
    seed=42,
    presence_penalty=0.1,
    frequency_penalty=0.1,
    parallel_tool_calls=True,
    timeout=120.0
)

model = OpenAIChatModel(
    model_name="gpt-4o",
    provider="azure",
    settings=settings
)

​
Available Models

Azure OpenAI provides access to OpenAI models through regional deployments:

​
GPT-4o Models

  • gpt-4o: Latest GPT-4o
  • gpt-4o-mini: Efficient variant

​
GPT-4 Models

  • gpt-4: Original GPT-4
  • gpt-4-32k: Extended context
  • gpt-4-turbo: Faster variant

​
GPT-3.5 Models

  • gpt-3.5-turbo: Cost-effective
  • gpt-3.5-turbo-16k: Extended context
Note: Model availability varies by region. Check Azure portal for your region’s offerings.

​
Deployment Configuration

​
Creating a Deployment

  1. Create Azure OpenAI resource
  2. Deploy a model (create deployment)
  3. Note deployment name (used as model_name)
  4. Get endpoint and keys
# Use your deployment name
model = OpenAIChatModel(
    model_name="my-gpt4o-deployment",  # Your deployment name
    provider="azure"
)

​
Regional Considerations

Choose region based on:
  • Data residency requirements
  • Latency requirements
  • Model availability
  • Compliance needs
Example regions:
  • East US
  • West Europe
  • UK South
  • Australia East

​
Best Practices

  1. Use Managed Identity: More secure than API keys
  2. Deploy in Same Region: Minimize latency
  3. Use Private Endpoints: For sensitive data
  4. Monitor Usage: Track through Azure portal
  5. Set Up Alerts: For quota and cost management
  6. Version API Calls: Use specific API versions
  7. Leverage Content Filtering: Built-in safety features
  8. Use Deployment Names: Not model IDs

​
Enterprise Features

​
Content Filtering

Azure includes multi-level content filtering:
  • Hate and fairness
  • Sexual content
  • Violence
  • Self-harm
Levels: Low, Medium, High filtering

​
Compliance

Azure OpenAI is compliant with:
  • SOC 2 Type II
  • ISO 27001, 27018, 27701
  • HIPAA
  • FedRAMP (select regions)

​
Data Privacy

  • Data stays in your Azure tenant
  • No data used for model training
  • Customer Lockbox support
  • Encryption at rest and in transit

​
Monitoring and Logging

from upsonic import Agent, Task, infer_model

model = infer_model("azure/gpt-4o")
agent = Agent(model=model)

# Azure automatically logs to Azure Monitor
task = Task("Process this request")
result = agent.do(task)

# View metrics in Azure Portal:
# - Token usage
# - Latency
# - Error rates
# - Content filtering actions