diff --git a/.github/workflows/CI.yml b/.github/workflows/CI.yml index 103f8222..44d6f403 100644 --- a/.github/workflows/CI.yml +++ b/.github/workflows/CI.yml @@ -196,7 +196,7 @@ jobs: # Increase the TPM for the Azure OpenAI models echo "Increasing TPM for Azure OpenAI models..." openai_gpt_deployment_url="/subscriptions/${{ secrets.AZURE_SUBSCRIPTION_ID }}/resourceGroups/${{ env.rg_name }}/providers/Microsoft.CognitiveServices/accounts/$openai_resource_name/deployments/gpt-4o-mini?api-version=2023-05-01" - az rest -m put -u "$openai_gpt_deployment_url" -b "{'sku':{'name':'Standard','capacity':${{ env.GPT_CAPACITY }}},'properties': {'model': {'format': 'OpenAI','name': 'gpt-4o-mini','version': '2024-07-18'}}}" + az rest -m put -u "$openai_gpt_deployment_url" -b "{'sku':{'name':'GlobalStandard','capacity':${{ env.GPT_CAPACITY }}},'properties': {'model': {'format': 'OpenAI','name': 'gpt-4o-mini','version': '2024-07-18'}}}" if [ $? -ne 0 ]; then echo "Failed to increase TPM for GPT deployment." exit 1 @@ -204,7 +204,7 @@ jobs: echo "Successfully increased TPM for GPT deployment." fi openai_embedding_deployment_url="/subscriptions/${{ secrets.AZURE_SUBSCRIPTION_ID }}/resourceGroups/${{ env.rg_name }}/providers/Microsoft.CognitiveServices/accounts/$openai_resource_name/deployments/text-embedding-large?api-version=2023-05-01" - az rest -m put -u "$openai_embedding_deployment_url" -b "{'sku':{'name':'Standard','capacity': ${{ env.TEXT_EMBEDDING_CAPACITY }}},'properties': {'model': {'format': 'OpenAI','name': 'text-embedding-3-large','version': '1'}}}" + az rest -m put -u "$openai_embedding_deployment_url" -b "{'sku':{'name':'GlobalStandard','capacity': ${{ env.TEXT_EMBEDDING_CAPACITY }}},'properties': {'model': {'format': 'OpenAI','name': 'text-embedding-3-large','version': '1'}}}" if [ $? -ne 0 ]; then echo "Failed to increase TPM for Text Embedding deployment." exit 1 diff --git a/Deployment/bicep/azureopenaiservicemodel.bicep b/Deployment/bicep/azureopenaiservicemodel.bicep index 712a4397..74ac1ee6 100644 --- a/Deployment/bicep/azureopenaiservicemodel.bicep +++ b/Deployment/bicep/azureopenaiservicemodel.bicep @@ -10,7 +10,7 @@ resource gpt4Deployment 'Microsoft.CognitiveServices/accounts/deployments@2023-0 parent: openAIService name: name sku: { - name: 'Standard' + name: 'GlobalStandard' capacity: model.capacity } properties: { diff --git a/Deployment/checkquota.ps1 b/Deployment/checkquota.ps1 index 7cc9eecb..9002df52 100644 --- a/Deployment/checkquota.ps1 +++ b/Deployment/checkquota.ps1 @@ -45,8 +45,8 @@ Write-Host "✅ Azure subscription set successfully." # Define models and their minimum required capacities $MIN_CAPACITY = @{ - "OpenAI.Standard.gpt-4o-mini" = $GPT_MIN_CAPACITY - "OpenAI.Standard.text-embedding-3-large" = $TEXT_EMBEDDING_MIN_CAPACITY + "OpenAI.GlobalStandard.gpt-4o-mini" = $GPT_MIN_CAPACITY + "OpenAI.GlobalStandard.text-embedding-3-large" = $TEXT_EMBEDDING_MIN_CAPACITY } $VALID_REGION = "" diff --git a/Deployment/main.bicep b/Deployment/main.bicep index a7edfdb6..f1c88558 100644 --- a/Deployment/main.bicep +++ b/Deployment/main.bicep @@ -119,7 +119,7 @@ module gs_openaiservicemodels_gpt4o 'bicep/azureopenaiservicemodel.bicep' = { version: '2024-07-18' raiPolicyName: '' capacity: 1 - scaleType: 'Standard' + scaleType: 'GlobalStandard' } } @@ -139,7 +139,7 @@ module gs_openaiservicemodels_text_embedding 'bicep/azureopenaiservicemodel.bice version: '1' raiPolicyName: '' capacity: 1 - scaleType: 'Standard' + scaleType: 'GlobalStandard' } } dependsOn: [