Adding support for custom Azure deployments (#1273)

* Adding support for custom Azure deployments

* Fixing formatting

* Fixing using Pamela's suggestions

* Fixing format

* Better clarity

---------

Co-authored-by: andredewes <[email protected]>
Co-authored-by: Pamela Fox <[email protected]>

Author: 3 people

app/backend/app.py

@@ -223,8 +223,10 @@ async def setup_clients():
     AZURE_OPENAI_SERVICE = os.getenv("AZURE_OPENAI_SERVICE")
     AZURE_OPENAI_GPT4V_DEPLOYMENT = os.environ.get("AZURE_OPENAI_GPT4V_DEPLOYMENT")
     AZURE_OPENAI_GPT4V_MODEL = os.environ.get("AZURE_OPENAI_GPT4V_MODEL")
-    AZURE_OPENAI_CHATGPT_DEPLOYMENT = os.getenv("AZURE_OPENAI_CHATGPT_DEPLOYMENT") if OPENAI_HOST == "azure" else None
-    AZURE_OPENAI_EMB_DEPLOYMENT = os.getenv("AZURE_OPENAI_EMB_DEPLOYMENT") if OPENAI_HOST == "azure" else None
+    AZURE_OPENAI_CHATGPT_DEPLOYMENT = (
+        os.getenv("AZURE_OPENAI_CHATGPT_DEPLOYMENT") if OPENAI_HOST.startswith("azure") else None
+    )
+    AZURE_OPENAI_EMB_DEPLOYMENT = os.getenv("AZURE_OPENAI_EMB_DEPLOYMENT") if OPENAI_HOST.startswith("azure") else None
     AZURE_VISION_ENDPOINT = os.getenv("AZURE_VISION_ENDPOINT", "")
     # Used only with non-Azure OpenAI deployments
     OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
@@ -297,12 +299,18 @@ async def setup_clients():
     # Used by the OpenAI SDK
     openai_client: AsyncOpenAI
 
-    if OPENAI_HOST == "azure":
+    if OPENAI_HOST.startswith("azure"):
         token_provider = get_bearer_token_provider(azure_credential, "https://cognitiveservices.azure.com/.default")
+
+        if OPENAI_HOST == "azure_custom":
+            endpoint = os.environ["AZURE_OPENAI_CUSTOM_URL"]
+        else:
+            endpoint = f"https://{AZURE_OPENAI_SERVICE}.openai.azure.com"
+
         # Store on app.config for later use inside requests
         openai_client = AsyncAzureOpenAI(
             api_version="2023-07-01-preview",
-            azure_endpoint=f"https://{AZURE_OPENAI_SERVICE}.openai.azure.com",
+            azure_endpoint=endpoint,
             azure_ad_token_provider=token_provider,
         )
     elif OPENAI_HOST == "local":

infra/main.bicep

@@ -39,8 +39,9 @@ param storageSkuName string // Set in main.parameters.json
 
 param appServiceSkuName string // Set in main.parameters.json
 
-@allowed([ 'azure', 'openai' ])
+@allowed([ 'azure', 'openai', 'azure_custom' ])
 param openAiHost string // Set in main.parameters.json
+param azureOpenAiCustomUrl string = ''
 
 param openAiServiceName string = ''
 param openAiResourceGroupName string = ''
@@ -87,7 +88,7 @@ param computerVisionSkuName string = 'S1'
 param chatGptDeploymentName string // Set in main.parameters.json
 param chatGptDeploymentCapacity int = 30
 param chatGpt4vDeploymentCapacity int = 10
-param chatGptModelName string = (openAiHost == 'azure') ? 'gpt-35-turbo' : 'gpt-3.5-turbo'
+param chatGptModelName string = startsWith(openAiHost, 'azure') ? 'gpt-35-turbo' : 'gpt-3.5-turbo'
 param chatGptModelVersion string = '0613'
 param embeddingDeploymentName string // Set in main.parameters.json
 param embeddingDeploymentCapacity int = 30
@@ -245,6 +246,7 @@ module backend 'core/host/appservice.bicep' = {
       APPLICATIONINSIGHTS_CONNECTION_STRING: useApplicationInsights ? monitoring.outputs.applicationInsightsConnectionString : ''
       // Shared by all OpenAI deployments
       OPENAI_HOST: openAiHost
+      AZURE_OPENAI_CUSTOM_URL: azureOpenAiCustomUrl
       AZURE_OPENAI_EMB_MODEL_NAME: embeddingModelName
       AZURE_OPENAI_CHATGPT_MODEL: chatGptModelName
       AZURE_OPENAI_GPT4V_MODEL: gpt4vModelName

infra/main.parameters.json

@@ -80,6 +80,9 @@
     "openAiHost": {
       "value": "${OPENAI_HOST=azure}"
     },
+    "azureOpenAiCustomUrl":{
+      "value": "${AZURE_OPENAI_CUSTOM_URL}"
+    },
     "openAiApiKey": {
       "value": "${OPENAI_API_KEY}"
     },