Migrate from Chat Completions to Responses API

.env.sample

@@ -1,5 +1,4 @@
-AZURE_OPENAI_API_VERSION=2024-02-15-preview
 AZURE_OPENAI_ENDPOINT=https://YOUR-ENDPOINT-HERE.openai.azure.com/
 # Name of the Azure OpenAI GPT deployment (different from the model name)
-AZURE_OPENAI_CHAT_DEPLOYMENT=gpt4o-mini
-AZURE_TENANT_ID=YOUR-TENANT-ID
+AZURE_OPENAI_CHAT_DEPLOYMENT=gpt-4o-mini
+AZURE_TENANT_ID=YOUR-TENANT-ID

.github/workflows/azure-dev.yaml

@@ -34,7 +34,6 @@ jobs:
       AZURE_OPENAI_DEPLOYMENT_CAPACITY: ${{ vars.AZURE_OPENAI_DEPLOYMENT_CAPACITY }}
       AZURE_OPENAI_DEPLOYMENT_SKU_NAME: ${{ vars.AZURE_OPENAI_DEPLOYMENT_SKU_NAME }}
       AZURE_OPENAI_SKU_NAME: ${{ vars.AZURE_OPENAI_SKU_NAME }}
-      AZURE_OPENAI_API_VERSION: ${{ vars.AZURE_OPENAI_API_VERSION }}
       CREATE_AZURE_OPENAI: ${{ vars.CREATE_AZURE_OPENAI }}
       AZURE_OPENAI_ENDPOINT: ${{ vars.AZURE_OPENAI_ENDPOINT }}
       SERVICE_ACA_RESOURCE_EXISTS: ${{ vars.SERVICE_ACA_RESOURCE_EXISTS }}

infra/aca.bicep

@@ -9,7 +9,6 @@ param serviceName string = 'aca'
 param exists bool
 param openAiDeploymentName string
 param openAiEndpoint string
-param openAiApiVersion string
 
 resource acaIdentity 'Microsoft.ManagedIdentity/userAssignedIdentities@2023-01-31' = {
   name: identityName
@@ -25,10 +24,6 @@ var env = [
     name: 'AZURE_OPENAI_ENDPOINT'
     value: openAiEndpoint
   }
-  {
-    name: 'AZURE_OPENAI_API_VERSION'
-    value: openAiApiVersion
-  }
   {
     name: 'RUNNING_IN_PRODUCTION'
     value: 'true'

infra/main.bicep

@@ -55,7 +55,6 @@ param openAiResourceGroupName string = ''
 })
 param openAiResourceLocation string
 param openAiSkuName string = ''
-param openAiApiVersion string = '' // Used by the SDK in the app code
 param disableKeyBasedAuth bool = true
 
 // Parameters for the specific Azure OpenAI deployment:
@@ -152,7 +151,6 @@ module aca 'aca.bicep' = {
     containerRegistryName: containerApps.outputs.registryName
     openAiDeploymentName: openAiDeploymentName
     openAiEndpoint: createAzureOpenAi ? openAi.outputs.endpoint : openAiEndpoint
-    openAiApiVersion: openAiApiVersion
     exists: acaExists
   }
 }
@@ -185,7 +183,6 @@ output AZURE_TENANT_ID string = tenant().tenantId
 output AZURE_OPENAI_RESOURCE_GROUP string = openAiResourceGroup.name
 output AZURE_OPENAI_RESOURCE_NAME string = openAi.outputs.name
 output AZURE_OPENAI_CHAT_DEPLOYMENT string = openAiDeploymentName
-output AZURE_OPENAI_API_VERSION string = openAiApiVersion
 output AZURE_OPENAI_ENDPOINT string = createAzureOpenAi ? openAi.outputs.endpoint : openAiEndpoint
 
 output SERVICE_ACA_IDENTITY_PRINCIPAL_ID string = aca.outputs.SERVICE_ACA_IDENTITY_PRINCIPAL_ID
@@ -196,4 +193,3 @@ output SERVICE_ACA_IMAGE_NAME string = aca.outputs.SERVICE_ACA_IMAGE_NAME
 output AZURE_CONTAINER_ENVIRONMENT_NAME string = containerApps.outputs.environmentName
 output AZURE_CONTAINER_REGISTRY_ENDPOINT string = containerApps.outputs.registryLoginServer
 output AZURE_CONTAINER_REGISTRY_NAME string = containerApps.outputs.registryName
-

infra/main.parameters.json

@@ -38,9 +38,6 @@
       "openAiSkuName": {
         "value": "${AZURE_OPENAI_SKU_NAME}"
       },
-      "openAiApiVersion": {
-        "value": "${AZURE_OPENAI_API_VERSION}"
-      },
       "createAzureOpenAi": {
         "value": "${CREATE_AZURE_OPENAI=true}"
       },

src/pyproject.toml

@@ -13,7 +13,7 @@ dependencies = [
     "httptools",
     # Used by uvicorn for reload functionality
     "watchfiles",
-    "openai",
+    "openai>=1.108.1",
     "azure-identity",
     "aiohttp>=3.11.0",
     "python-dotenv",

src/quartapp/chat.py

@@ -7,7 +7,7 @@
     ManagedIdentityCredential,
     get_bearer_token_provider,
 )
-from openai import AsyncAzureOpenAI
+from openai import AsyncOpenAI
 from quart import (
     Blueprint,
     Response,
@@ -22,7 +22,6 @@
 
 @bp.before_app_serving
 async def configure_openai():
-
     # Use ManagedIdentityCredential with the client_id for user-assigned managed identities
     user_assigned_managed_identity_credential = ManagedIdentityCredential(client_id=os.getenv("AZURE_CLIENT_ID"))
 
@@ -47,11 +46,11 @@ async def configure_openai():
     if not os.getenv("AZURE_OPENAI_CHAT_DEPLOYMENT"):
         raise ValueError("AZURE_OPENAI_CHAT_DEPLOYMENT is required for Azure OpenAI")
 
-    # Create the Asynchronous Azure OpenAI client
-    bp.openai_client = AsyncAzureOpenAI(
-        api_version=os.getenv("AZURE_OPENAI_API_VERSION") or "2024-02-15-preview",
-        azure_endpoint=os.getenv("AZURE_OPENAI_ENDPOINT"),
-        azure_ad_token_provider=token_provider,
+    # Create the Asynchronous OpenAI client with Azure OpenAI endpoint
+    azure_endpoint = os.getenv("AZURE_OPENAI_ENDPOINT").rstrip("/")
+    bp.openai_client = AsyncOpenAI(
+        base_url=f"{azure_endpoint}/openai/v1",
+        api_key=token_provider,
     )
     # Set the model name to the Azure OpenAI model deployment name
     bp.openai_model = os.getenv("AZURE_OPENAI_CHAT_DEPLOYMENT")
@@ -73,22 +72,33 @@ async def chat_handler():
 
     @stream_with_context
     async def response_stream():
-        # This sends all messages, so API request may exceed token limits
-        all_messages = [
-            {"role": "system", "content": "You are a helpful assistant."},
-        ] + request_messages
-
-        chat_coroutine = bp.openai_client.chat.completions.create(
+        # Build input items for Responses API
+        input_items = [
+            {"role": "system", "content": [{"type": "input_text", "text": "You are a helpful assistant."}]},
+        ]
+        for msg in request_messages:
+            content_type = "output_text" if msg["role"] == "assistant" else "input_text"
+            input_items.append(
+                {
+                    "role": msg["role"],
+                    "content": [{"type": content_type, "text": msg["content"]}],
+                }
+            )
+
+        response_stream = await bp.openai_client.responses.create(
             # Azure OpenAI takes the deployment name as the model name
             model=bp.openai_model,
-            messages=all_messages,
+            input=input_items,
             stream=True,
+            store=False,
         )
         try:
-            async for event in await chat_coroutine:
-                event_dict = event.model_dump()
-                if event_dict["choices"]:
-                    yield json.dumps(event_dict["choices"][0], ensure_ascii=False) + "\n"
+            async for event in response_stream:
+                if event.type == "response.output_text.delta":
+                    yield json.dumps({"delta": {"content": event.delta, "role": None}}, ensure_ascii=False) + "\n"
+                elif event.type == "response.completed":
+                    finish_event = {"delta": {"content": None, "role": None}, "finish_reason": "stop"}
+                    yield json.dumps(finish_event, ensure_ascii=False) + "\n"
         except Exception as e:
             current_app.logger.error(e)
             yield json.dumps({"error": str(e)}, ensure_ascii=False) + "\n"

src/requirements.txt

@@ -1,14 +1,14 @@
-#
-# This file is autogenerated by pip-compile with Python 3.12
-# by the following command:
-#
-#    pip-compile pyproject.toml
+#
+# This file is autogenerated by pip-compile with Python 3.12
+# by the following command:
+#
+#    pip-compile --output-file=requirements.txt pyproject.toml
 #
 aiofiles==24.1.0
     # via quart
 aiohappyeyeballs==2.6.1
     # via aiohttp
-aiohttp==3.12.15
+aiohttp==3.12.15
     # via quartapp (pyproject.toml)
 aiosignal==1.4.0
     # via aiohttp
@@ -64,19 +64,19 @@ h11==0.16.0
     #   hypercorn
     #   uvicorn
     #   wsproto
-h2==4.3.0
-    # via hypercorn
-hpack==4.1.0
+h2==4.3.0
+    # via hypercorn
+hpack==4.1.0
     # via h2
 httpcore==1.0.9
     # via httpx
 httptools==0.6.4
     # via quartapp (pyproject.toml)
-httpx==0.28.1
+httpx==0.28.1
     # via openai
 hypercorn==0.17.3
     # via quart
-hyperframe==6.1.0
+hyperframe==6.1.0
     # via h2
 idna==3.10
     # via
@@ -110,13 +110,13 @@ multidict==6.6.4
     # via
     #   aiohttp
     #   yarl
-openai==1.107.2
+openai==2.16.0
     # via quartapp (pyproject.toml)
 packaging==25.0
     # via gunicorn
 priority==2.0.0
     # via hypercorn
-propcache==0.3.2
+propcache==0.3.2
     # via
     #   aiohttp
     #   yarl
@@ -126,11 +126,11 @@ pydantic==2.11.9
     # via openai
 pydantic-core==2.33.2
     # via pydantic
-pyjwt[crypto]==2.10.1
-    # via
-    #   msal
-    #   pyjwt
-python-dotenv==1.1.1
+pyjwt[crypto]==2.10.1
+    # via
+    #   msal
+    #   pyjwt
+python-dotenv==1.1.1
     # via quartapp (pyproject.toml)
 pyyaml==6.0.2
     # via quartapp (pyproject.toml)