feat: Developed scripts and Implement ACR deployment (#2298)

Co-authored-by: Vamshi-Microsoft <v-vamolla@microsoft.com>
Co-authored-by: Ajit Padhi <v-ajpadhi@microsoft.com>

Author: 3 people

azure.yaml

@@ -28,7 +28,13 @@ hooks:
         Write-Host "1. Login to Azure CLI (required for post-deployment script):" -ForegroundColor Yellow
         Write-Host "   az login" -ForegroundColor Cyan
         Write-Host ""
-        Write-Host "2. Run the post-deployment setup script:" -ForegroundColor Yellow
+        Write-Host "2. If you deployed with container hosting, run the combined ACR build/push/update script:" -ForegroundColor Yellow
+        Write-Host "   ./scripts/acr_build_push_update.ps1 -ResourceGroupName `"$env:AZURE_RESOURCE_GROUP`"" -ForegroundColor Cyan
+        Write-Host ""
+        Write-Host "   Or using Bash:" -ForegroundColor Yellow
+        Write-Host "   bash scripts/acr_build_push_update.sh `"$env:AZURE_RESOURCE_GROUP`"" -ForegroundColor Cyan
+        Write-Host ""
+        Write-Host "3. Run the post-deployment setup script:" -ForegroundColor Yellow
         Write-Host "   ./scripts/post_deployment_setup.ps1 -ResourceGroupName `"$env:AZURE_RESOURCE_GROUP`"" -ForegroundColor Cyan
         Write-Host ""
         Write-Host "   Or using Bash:" -ForegroundColor Yellow
@@ -47,7 +53,10 @@ hooks:
         echo "1. Login to Azure CLI (required for post-deployment script):"
         echo "   az login"
         echo ""
-        echo "2. Run the post-deployment setup script:"
+        echo "2. If you deployed with container hosting, run the combined ACR build/push/update script:"
+        echo "   bash scripts/acr_build_push_update.sh \"$AZURE_RESOURCE_GROUP\""
+        echo ""
+        echo "3. Run the post-deployment setup script:"
         echo "   bash scripts/post_deployment_setup.sh \"$AZURE_RESOURCE_GROUP\""
         echo ""
       continueOnError: false

code/backend/api/chat_history.py

@@ -9,6 +9,7 @@
 )
 from backend.batch.utilities.helpers.config.config_helper import ConfigHelper
 from backend.batch.utilities.helpers.env_helper import EnvHelper
+from backend.batch.utilities.helpers.openai_utils import build_completion_kwargs
 from backend.batch.utilities.chat_history.database_factory import DatabaseFactory
 from backend.batch.utilities.loggers.event_utils import track_event_if_configured
 
@@ -516,7 +517,7 @@ async def generate_title(conversation_messages):
             model=env_helper.AZURE_OPENAI_MODEL,
             messages=messages,
             temperature=1,
-            max_tokens=64,
+            **build_completion_kwargs(env_helper.AZURE_OPENAI_MODEL, 64),
         )
 
         # Ensure response contains valid choices and content

code/backend/batch/utilities/helpers/llm_helper.py

@@ -10,6 +10,7 @@
 from azure.ai.ml import MLClient
 from .azure_credential_utils import get_azure_credential
 from .env_helper import EnvHelper
+from .openai_utils import build_completion_kwargs
 
 logger = logging.getLogger(__name__)
 
@@ -45,27 +46,33 @@ def __init__(self):
         logger.info("Initializing LLMHelper completed")
 
     def get_llm(self):
+        completion_kwargs = build_completion_kwargs(
+            self.llm_model, self.llm_max_tokens
+        )
         if self.auth_type_keys:
             return AzureChatOpenAI(
                 deployment_name=self.llm_model,
                 temperature=0,
-                max_tokens=self.llm_max_tokens,
                 openai_api_version=self.openai_client._api_version,
                 azure_endpoint=self.env_helper.AZURE_OPENAI_ENDPOINT,
                 api_key=self.env_helper.OPENAI_API_KEY,
+                **completion_kwargs,
             )
         else:
             return AzureChatOpenAI(
                 deployment_name=self.llm_model,
                 temperature=0,
-                max_tokens=self.llm_max_tokens,
                 openai_api_version=self.openai_client._api_version,
                 azure_endpoint=self.env_helper.AZURE_OPENAI_ENDPOINT,
                 azure_ad_token_provider=self.token_provider,
+                **completion_kwargs,
             )
 
     # TODO: This needs to have a custom callback to stream back to the UI
     def get_streaming_llm(self):
+        completion_kwargs = build_completion_kwargs(
+            self.llm_model, self.llm_max_tokens
+        )
         if self.auth_type_keys:
             return AzureChatOpenAI(
                 azure_endpoint=self.env_helper.AZURE_OPENAI_ENDPOINT,
@@ -74,8 +81,8 @@ def get_streaming_llm(self):
                 callbacks=[StreamingStdOutCallbackHandler],
                 deployment_name=self.llm_model,
                 temperature=0,
-                max_tokens=self.llm_max_tokens,
                 openai_api_version=self.openai_client._api_version,
+                **completion_kwargs,
             )
         else:
             return AzureChatOpenAI(
@@ -85,9 +92,9 @@ def get_streaming_llm(self):
                 callbacks=[StreamingStdOutCallbackHandler],
                 deployment_name=self.llm_model,
                 temperature=0,
-                max_tokens=self.llm_max_tokens,
                 openai_api_version=self.openai_client._api_version,
                 azure_ad_token_provider=self.token_provider,
+                **completion_kwargs,
             )
 
     def get_embedding_model(self):
@@ -143,6 +150,7 @@ def get_chat_completion_with_functions(
             messages=messages,
             functions=functions,
             function_call=function_call,
+            **build_completion_kwargs(self.llm_model, self.llm_max_tokens),
         )
 
     def get_chat_completion(
@@ -151,7 +159,7 @@ def get_chat_completion(
         return self.openai_client.chat.completions.create(
             model=model or self.llm_model,
             messages=messages,
-            max_tokens=self.llm_max_tokens,
+            **build_completion_kwargs(model or self.llm_model, self.llm_max_tokens),
             **kwargs
         )
 
@@ -174,12 +182,13 @@ def get_sk_chat_completion_service(self, service_id: str):
             )
 
     def get_sk_service_settings(self, service: AzureChatCompletion):
+        completion_kwargs = build_completion_kwargs(self.llm_model, self.llm_max_tokens)
         return cast(
             AzureChatPromptExecutionSettings,
             service.instantiate_prompt_execution_settings(
                 service_id=service.service_id,
                 temperature=0,
-                max_tokens=self.llm_max_tokens,
+                **completion_kwargs,
             ),
         )
 

code/backend/batch/utilities/helpers/openai_utils.py

@@ -0,0 +1,15 @@
+from typing import Any
+
+
+def build_completion_kwargs(model_name: str | None, max_tokens: int | None, **kwargs: Any) -> dict[str, Any]:
+    """Return request kwargs that are compatible with newer Azure OpenAI models."""
+    if max_tokens is None:
+        return kwargs
+
+    normalized_model = (model_name or "").lower()
+    if "gpt-5" in normalized_model or normalized_model.startswith(("o1", "o3", "o4")):
+        kwargs["max_completion_tokens"] = max_tokens
+    else:
+        kwargs["max_tokens"] = max_tokens
+
+    return kwargs

code/create_app.py

@@ -21,6 +21,7 @@
 from backend.batch.utilities.helpers.azure_search_helper import AzureSearchHelper
 from backend.batch.utilities.helpers.orchestrator_helper import Orchestrator
 from backend.batch.utilities.helpers.config.config_helper import ConfigHelper
+from backend.batch.utilities.helpers.openai_utils import build_completion_kwargs
 from backend.batch.utilities.helpers.config.conversation_flow import ConversationFlow
 from backend.batch.utilities.helpers.prompt_utils import get_current_date_suffix
 from backend.api.chat_history import bp_chat_history_response
@@ -202,8 +203,11 @@ def conversation_with_data(conversation: Request, env_helper: EnvHelper):
         model=env_helper.AZURE_OPENAI_MODEL,
         messages=messages,
         temperature=float(env_helper.AZURE_OPENAI_TEMPERATURE),
-        max_tokens=int(env_helper.AZURE_OPENAI_MAX_TOKENS),
         top_p=float(env_helper.AZURE_OPENAI_TOP_P),
+        **build_completion_kwargs(
+            env_helper.AZURE_OPENAI_MODEL,
+            int(env_helper.AZURE_OPENAI_MAX_TOKENS),
+        ),
         stop=(
             env_helper.AZURE_OPENAI_STOP_SEQUENCE.split("|")
             if env_helper.AZURE_OPENAI_STOP_SEQUENCE
@@ -366,8 +370,11 @@ def conversation_without_data(conversation: Request, env_helper: EnvHelper):
         model=env_helper.AZURE_OPENAI_MODEL,
         messages=messages,
         temperature=float(env_helper.AZURE_OPENAI_TEMPERATURE),
-        max_tokens=int(env_helper.AZURE_OPENAI_MAX_TOKENS),
         top_p=float(env_helper.AZURE_OPENAI_TOP_P),
+        **build_completion_kwargs(
+            env_helper.AZURE_OPENAI_MODEL,
+            int(env_helper.AZURE_OPENAI_MAX_TOKENS),
+        ),
         stop=(
             env_helper.AZURE_OPENAI_STOP_SEQUENCE.split("|")
             if env_helper.AZURE_OPENAI_STOP_SEQUENCE

code/tests/functional/tests/backend_api/default/test_conversation.py

@@ -317,6 +317,7 @@ def test_post_makes_correct_call_to_openai_chat_completions_with_functions(
                         },
                     },
                 ],
+                "max_tokens": int(app_config.get("AZURE_OPENAI_MAX_TOKENS")),
             },
             headers={
                 "Accept": "application/json",

code/tests/utilities/helpers/test_llm_helper.py

@@ -108,6 +108,22 @@ def test_get_sk_service_settings():
     assert settings.max_tokens == int(AZURE_OPENAI_MAX_TOKENS)
 
 
+def test_get_chat_completion_uses_max_completion_tokens_for_gpt_5_models(
+    azure_openai_mock, env_helper_mock
+):
+    # given
+    env_helper_mock.AZURE_OPENAI_MODEL = "gpt-5.1"
+    llm_helper = LLMHelper()
+
+    # when
+    llm_helper.get_chat_completion([{"role": "user", "content": "hello"}])
+
+    # then
+    call_kwargs = azure_openai_mock.return_value.chat.completions.create.call_args.kwargs
+    assert call_kwargs["max_completion_tokens"] == int(AZURE_OPENAI_MAX_TOKENS)
+    assert "max_tokens" not in call_kwargs
+
+
 def test_generate_embeddings_embeds_input(azure_openai_mock):
     # given
     llm_helper = LLMHelper()

docs/AVMPostDeploymentGuide.md

@@ -23,13 +23,26 @@ cd chat-with-your-data-solution-accelerator
 bash scripts/post_deployment_setup.sh "<your-resource-group-name>"
 ```
 
-### Step 2: Configure App Authentication
+### Step 2: Build, Push, and Update Container Images (Container Model Only)
+
+> **📌 Skip this step** if you deployed with the default `hostingModel=code`.
+
+When deploying with `hostingModel=container`, the App Services start with a placeholder hello-world image. Run the combined container workflow to build and push the application images to your Azure Container Registry and update the App Services to use them.
+
+**Run the combined ACR build/push/update script (remote build, no Docker required):**
+```bash
+bash scripts/acr_build_push_update.sh "<your-resource-group-name>"
+```
+
+> The script configures managed-identity based authentication between the App Services and your private ACR, then restarts all services. To build locally with Docker instead, use `--mode local`.
+
+### Step 3: Configure App Authentication
 
 1. After deployment is complete, navigate to your Azure App Service in the Azure portal
 2. Follow the detailed instructions in [Set Up Authentication in Azure App Service](./azure_app_service_auth_setup.md) to add authentication to your web app
 3. This will ensure only authorized users can access your application
 
-### Step 3: Access and Configure the Admin Site
+### Step 4: Access and Configure the Admin Site
 
 1. **Navigate to the admin site** using the following URL pattern:
    ```
@@ -47,7 +60,7 @@ bash scripts/post_deployment_setup.sh "<your-resource-group-name>"
    - Wait for the documents to be processed and indexed
    - Verify successful ingestion through the admin interface
 
-### Step 4: Access the Chat Application
+### Step 5: Access the Chat Application
 
 1. **Navigate to the main chat application** using this URL pattern:
    ```

docs/LOCAL_DEPLOYMENT.md

@@ -334,20 +334,45 @@ bash scripts/post_deployment_setup.sh "<your-resource-group-name>"
 
 > **Note:** The script auto-discovers all resources in the resource group. It handles private networking (WAF) deployments by temporarily enabling public access, performing the setup, then restoring the original state.
 
-### 5.2 Configure Authentication (Required for Chat Application)
+### 5.2 Build, Push, and Update Container Images (Container Model Only)
+
+> **📌 Skip this step** if you deployed with the default `hostingModel=code`.
+
+When deploying with `hostingModel=container`, the App Services start with a placeholder hello-world image. After provisioning, run the combined container workflow to build and push the application images to your Azure Container Registry and update the App Services to use them.
+
+*PowerShell (Windows):*
+```powershell
+.\scripts\acr_build_push_update.ps1 -ResourceGroupName "<your-resource-group-name>"
+```
+
+*Bash (Linux/macOS/WSL):*
+```bash
+bash scripts/acr_build_push_update.sh "<your-resource-group-name>"
+```
+
+This script:
+- Builds and pushes the images to your ACR
+- Updates each App Service to pull its image from your private ACR using managed-identity authentication
+- Restarts all services
+
+> By default, images are built remotely using `az acr build` (no local Docker required). To build locally with Docker instead, use `-Mode local` in PowerShell or `--mode local` in Bash. You can also set a custom tag with `-Tag` or `--tag`.
+
+> **Re-deployment note:** If you re-run `azd provision`, run this script again to restore the correct container images.
+
+### 5.3 Configure Authentication (Required for Chat Application)
 
 **This step is mandatory for Chat Application access:**
 
 1. Follow [App Authentication Configuration](./azure_app_service_auth_setup.md)
 2. Wait up to 10 minutes for authentication changes to take effect
 
-### 5.3 Verify Deployment
+### 5.4 Verify Deployment
 
 1. Access your application using the URL from Step 4.3
 2. Confirm the application loads successfully
 3. Verify you can sign in with your authenticated account
 
-### 5.4 Test the Application
+### 5.5 Test the Application
 
 **Quick Test Steps:**
 1. Navigate to the admin site, where you can upload documents. Then select Ingest Data and add your data. You can find sample data in the [data](../data) directory.