Skip to content

DeployWithLimitedQuota.md

Latest commit

 

History

History
98 lines (67 loc) · 3.17 KB

DeployWithLimitedQuota.md

File metadata and controls

98 lines (67 loc) · 3.17 KB

Deploying with Limited OpenAI Quota

This document provides guidance on deploying the Document Knowledge Mining Solution Accelerator when you have limited Azure OpenAI model quota available.

Overview

By default, the solution requires:

  • GPT model: 100,000 Tokens Per Minute (TPM)
  • Embedding model: 200,000 TPM

If your Azure OpenAI service has lower quota limits, you can modify the deployment to work with reduced capacity.

Prerequisites

Before proceeding, ensure you have:

  • Azure Developer CLI (azd) installed
  • Access to your Azure OpenAI service quota settings
  • Knowledge of your current TPM limits

Deployment Options

You have two approaches to deploy with less quota:

Option 1: Remove Quota Validation

Remove the metadata section (lines 94-102) from the infra/main.bicep file:

@metadata({
  azd: {
    type: 'location'
    usageName: [
      'OpenAI.GlobalStandard.gpt4.1-mini,150'
      'OpenAI.GlobalStandard.text-embedding-3-large,100'
    ]
  }
})

Option 2: Modify Quota Thresholds (Recommended)

Update the values on lines 98-99 in infra/main.bicep to match your available quota:

@metadata({
  azd: {
    type: 'location'
    usageName: [
        'OpenAI.GlobalStandard.gpt4.1-mini, 50'           // Changed from 150
        'OpenAI.GlobalStandard.text-embedding-3-large, 50'  // Changed from 100
    ]
  }
})

Configuration Steps

After modifying the Bicep file, configure your deployment capacity:

azd env set AZURE_ENV_MODEL_CAPACITY="50"
azd env set AZURE_ENV_EMBEDDING_MODEL_CAPACITY="50"

Note: Adjust the values (50) to match your actual available quota.

Deploy the Solution

Once configured, proceed with deployment:

azd up

Performance Considerations

⚠️ Important: Using reduced TPM limits may impact application performance:

For optimal performance, we recommend maintaining at least 200,000 TPM for GPT models when possible.

Additional Resources

For more detailed information, refer to:

Why we need to do this?

  • The solution uses built-in Azure Developer CLI (azd) quota validation to prevent deployment failures. Specifically, azd performs pre-deployment checks to ensure sufficient quota is available i.e. 200k TPM for gpt model and 80k TPM for embedding model.

  • These quota thresholds are hardcoded in the infrastructure file because azd's quota checking mechanism doesn't currently support parameterized values. If your Azure OpenAI service has quota below these thresholds, the deployment will fail during the validation phase rather than proceeding and failing later in the process.

  • By following the steps above, you can either:

    1. Bypass quota validation entirely by removing the metadata block
    2. Lower the validation thresholds to match your available quota (e.g., 50,000 TPM)

  • This ensures successful deployment while working within your quota constraints.