Configure GlobalStandard as default deployment sku for gpt-4o

.devcontainer/devcontainer.json

@@ -23,7 +23,7 @@
     "forwardPorts": [
         50505
     ],
-    "postCreateCommand": "",
+    "postCreateCommand": "python3 -m pip install -r requirements-dev.txt",
     "remoteUser": "vscode",
     "hostRequirements": {
         "memory": "8gb"

azure.yaml

@@ -132,3 +132,12 @@ hooks:
         run: ./scripts/auth_update.sh;./scripts/prepdocs.sh
         interactive: true
         continueOnError: false
+    predown:
+      windows:
+        shell: pwsh
+        run: ./scripts/pre-down.ps1
+        continueOnError: true
+      posix:
+        shell: sh
+        run: ./scripts/pre-down.sh
+        continueOnError: true

infra/main.bicep

@@ -110,32 +110,37 @@ param chatGptModelName string = ''
 param chatGptDeploymentName string = ''
 param chatGptDeploymentVersion string = ''
 param chatGptDeploymentCapacity int = 0
+param chatGptDeploymentSkuName string = ''
 var chatGpt = {
   modelName: !empty(chatGptModelName)
     ? chatGptModelName
     : startsWith(openAiHost, 'azure') ? 'gpt-35-turbo' : 'gpt-3.5-turbo'
   deploymentName: !empty(chatGptDeploymentName) ? chatGptDeploymentName : 'chat'
   deploymentVersion: !empty(chatGptDeploymentVersion) ? chatGptDeploymentVersion : '0613'
   deploymentCapacity: chatGptDeploymentCapacity != 0 ? chatGptDeploymentCapacity : 30
+  deploymentSkuName: !empty(chatGptDeploymentSkuName) ? chatGptDeploymentSkuName : 'Standard'
 }
 
 param embeddingModelName string = ''
 param embeddingDeploymentName string = ''
 param embeddingDeploymentVersion string = ''
 param embeddingDeploymentCapacity int = 0
 param embeddingDimensions int = 0
+param embeddingDeploymentSkuName string = ''
 var embedding = {
   modelName: !empty(embeddingModelName) ? embeddingModelName : 'text-embedding-ada-002'
   deploymentName: !empty(embeddingDeploymentName) ? embeddingDeploymentName : 'embedding'
   deploymentVersion: !empty(embeddingDeploymentVersion) ? embeddingDeploymentVersion : '2'
   deploymentCapacity: embeddingDeploymentCapacity != 0 ? embeddingDeploymentCapacity : 30
+  deploymentSkuName: !empty(embeddingDeploymentSkuName) ? embeddingDeploymentSkuName : 'Standard'
   dimensions: embeddingDimensions != 0 ? embeddingDimensions : 1536
 }
 
 param gpt4vModelName string = 'gpt-4o'
 param gpt4vDeploymentName string = 'gpt-4o'
 param gpt4vModelVersion string = '2024-05-13'
 param gpt4vDeploymentCapacity int = 10
+param gpt4vDeploymentSkuName string = 'GlobalStandard'
 
 param tenantId string = tenant().tenantId
 param authTenantId string = ''
@@ -453,7 +458,7 @@ var defaultOpenAiDeployments = [
       version: chatGpt.deploymentVersion
     }
     sku: {
-      name: 'Standard'
+      name: chatGpt.deploymentSkuName
       capacity: chatGpt.deploymentCapacity
     }
   }
@@ -465,7 +470,7 @@ var defaultOpenAiDeployments = [
       version: embedding.deploymentVersion
     }
     sku: {
-      name: 'Standard'
+      name: embedding.deploymentSkuName
       capacity: embedding.deploymentCapacity
     }
   }
@@ -483,7 +488,7 @@ var openAiDeployments = concat(
             version: gpt4vModelVersion
           }
           sku: {
-            name: 'Standard'
+            name: gpt4vDeploymentSkuName
             capacity: gpt4vDeploymentCapacity
           }
         }

infra/main.parameters.json

@@ -101,6 +101,9 @@
     "chatGptDeploymentCapacity":{
       "value": "${AZURE_OPENAI_CHATGPT_DEPLOYMENT_CAPACITY}"
     },
+    "chatGptDeploymentSkuName":{
+      "value": "${AZURE_OPENAI_CHATGPT_DEPLOYMENT_SKU=Standard}"
+    },
     "embeddingModelName":{
       "value": "${AZURE_OPENAI_EMB_MODEL_NAME}"
     },
@@ -113,12 +116,27 @@
     "embeddingDeploymentCapacity":{
       "value": "${AZURE_OPENAI_EMB_DEPLOYMENT_CAPACITY}"
     },
+    "embeddingDeploymentSkuName":{
+      "value": "${AZURE_OPENAI_EMB_DEPLOYMENT_SKU=Standard}"
+    },
     "embeddingDimensions": {
       "value": "${AZURE_OPENAI_EMB_DIMENSIONS}"
     },
+    "gpt4vModelName": {
+      "value": "${AZURE_OPENAI_GPT4V_MODEL_NAME=gpt-4o}"
+    },
+    "gpt4vDeploymentName": {
+      "value": "${AZURE_OPENAI_GPT4V_DEPLOYMENT=gpt-4o}"
+    },
+    "gpt4vDeploymentVersion":{
+      "value": "${AZURE_OPENAI_GPT4V_DEPLOYMENT_VERSION=2024-05-13}"
+    },
     "gpt4vDeploymentCapacity":{
       "value": "${AZURE_OPENAI_GPT4V_DEPLOYMENT_CAPACITY=10}"
     },
+    "gpt4vDeploymentSkuName":{
+      "value": "${AZURE_OPENAI_GPT4V_DEPLOYMENT_SKU=GlobalStandard}"
+    },
     "openAiHost": {
       "value": "${OPENAI_HOST=azure}"
     },

requirements-dev.txt

@@ -1,4 +1,5 @@
 -r app/backend/requirements.txt
+azure-mgmt-cognitiveservices
 ruff
 black
 pytest

scripts/pre-down.ps1

@@ -0,0 +1,10 @@
+# Get the directory of the current script
+$scriptDir = Split-Path -Parent $MyInvocation.MyCommand.Definition
+
+# Load environment variables from azd env
+$subscriptionId = azd env get-value AZURE_SUBSCRIPTION_ID
+$resourceName = azd env get-value AZURE_OPENAI_SERVICE
+$resourceGroup = azd env get-value AZURE_OPENAI_RESOURCE_GROUP
+
+# Run the Python script with the retrieved values
+python "$scriptDir/pre-down.py" --subscription-id $subscriptionId --resource-name $resourceName --resource-group $resourceGroup

scripts/pre-down.py

@@ -0,0 +1,32 @@
+import argparse
+
+from azure.identity import DefaultAzureCredential
+from azure.mgmt.cognitiveservices import CognitiveServicesManagementClient
+
+# Set up argument parsing
+parser = argparse.ArgumentParser(description="Delete an Azure OpenAI deployment.")
+parser.add_argument("--resource-name", required=True, help="The name of the Azure OpenAI resource.")
+parser.add_argument("--resource-group", required=True, help="The name of the Azure resource group.")
+parser.add_argument("--subscription-id", required=True, help="The Azure subscription ID.")
+
+args = parser.parse_args()
+
+# Authenticate using DefaultAzureCredential
+credential = DefaultAzureCredential()
+
+# Initialize the Cognitive Services client
+client = CognitiveServicesManagementClient(credential, subscription_id=args.subscription_id)
+
+# List all deployments
+deployments = client.deployments.list(resource_group_name=args.resource_group, account_name=args.resource_name)
+
+# Delete each deployment and wait for the operation to complete
+for deployment in deployments:
+    deployment_name = deployment.name
+    if not deployment_name:
+        continue
+    poller = client.deployments.begin_delete(
+        resource_group_name=args.resource_group, account_name=args.resource_name, deployment_name=deployment_name
+    )
+    poller.result()
+    print(f"Deployment {deployment_name} deleted successfully.")

scripts/pre-down.sh

@@ -0,0 +1,12 @@
+#!/bin/bash
+
+# Get the directory of the current script
+script_dir=$(dirname "$0")
+
+# Load environment variables from azd env
+subscription_id=$(azd env get-value AZURE_SUBSCRIPTION_ID)
+resource_name=$(azd env get-value AZURE_OPENAI_SERVICE)
+resource_group=$(azd env get-value AZURE_OPENAI_RESOURCE_GROUP)
+
+# Run the Python script with the retrieved values
+python "$script_dir/pre-down.py" --subscription-id $subscription_id --resource-name $resource_name --resource-group $resource_group