[Cognitive Services] az cognitiveservices account adapter-deployment: Add new command group (preview)

src/azure-cli/azure/cli/command_modules/cognitiveservices/_client_factory.py

@@ -134,6 +134,10 @@ def cf_managed_compute_deployments(cli_ctx, *_):
     return get_cognitiveservices_management_client(cli_ctx).managed_compute_deployments
 
 
+def cf_adapter_deployments(cli_ctx, *_):
+    return get_cognitiveservices_management_client(cli_ctx).adapter_deployments
+
+
 def cf_computes(cli_ctx, *_):
     return get_cognitiveservices_management_client(cli_ctx).computes
 

src/azure-cli/azure/cli/command_modules/cognitiveservices/_help.py

@@ -451,6 +451,76 @@
         --deployment-name gpt-oss-120b-gpu
 """
 
+helps[
+    "cognitiveservices account adapter-deployment"
+] = """
+type: group
+short-summary: Manage LoRA adapter deployments for Azure Cognitive Services accounts.
+long-summary: |
+    An adapter deployment is a lightweight LoRA weight patch that attaches to an existing
+    managed compute (parent) deployment. It owns no SKU, capacity, base model, or runtime
+    - all inference resources come from the parent managed compute deployment referenced
+    by --target-deployment-name.
+"""
+
+helps[
+    "cognitiveservices account adapter-deployment create"
+] = """
+type: command
+short-summary: Create (or replace / re-target) a LoRA adapter deployment attached to a managed compute deployment.
+long-summary: |
+    The adapter must reference a protected Foundry fine-tuning LoRA output as its source model.
+    The target deployment is the name of an existing managed compute deployment in the same
+    account. Re-issuing create with the same adapter name and a different --target-deployment-name
+    re-targets the adapter; --source-model-id is immutable.
+examples:
+  - name: Create an adapter deployment attached to an existing managed compute deployment.
+    text: >
+        az cognitiveservices account adapter-deployment create
+        -g myResourceGroup -n myAccount
+        --adapter-deployment-name my-lora-adapter
+        --source-model-id "azureai://accounts/myAccount/projects/myProject/models/my-fine-tuned-lora/versions/1"
+        --target-deployment-name gpt-oss-120b-gpu
+"""
+
+helps[
+    "cognitiveservices account adapter-deployment show"
+] = """
+type: command
+short-summary: Show an adapter deployment for Azure Cognitive Services account.
+examples:
+  - name: Show an adapter deployment.
+    text: >
+        az cognitiveservices account adapter-deployment show
+        -g myResourceGroup -n myAccount
+        --adapter-deployment-name my-lora-adapter
+"""
+
+helps[
+    "cognitiveservices account adapter-deployment list"
+] = """
+type: command
+short-summary: List all adapter deployments for Azure Cognitive Services account.
+examples:
+  - name: List all adapter deployments.
+    text: >
+        az cognitiveservices account adapter-deployment list
+        -g myResourceGroup -n myAccount
+"""
+
+helps[
+    "cognitiveservices account adapter-deployment delete"
+] = """
+type: command
+short-summary: Delete an adapter deployment from Azure Cognitive Services account.
+examples:
+  - name: Delete an adapter deployment.
+    text: >
+        az cognitiveservices account adapter-deployment delete
+        -g myResourceGroup -n myAccount
+        --adapter-deployment-name my-lora-adapter
+"""
+
 helps[
     "cognitiveservices commitment-tier"
 ] = """
@@ -1042,6 +1112,8 @@
     examples:
     - name: Create a compute resource.
       text: az cognitiveservices account compute create -g MyResourceGroup -n MyAccount --compute-name my-compute --location westus2 --pool-name my-pool --instance-type Standard_DS3_v2 --node-count 1
+    - name: Create a Spot compute resource without waiting for completion.
+      text: az cognitiveservices account compute create -g MyResourceGroup -n MyAccount --compute-name my-compute --location westus2 --pool-name my-pool --instance-type Standard_DS3_v2 --node-count 1 --vm-priority Spot --no-wait
 """
 
 helps['cognitiveservices account compute delete'] = """
@@ -1050,6 +1122,8 @@
     examples:
     - name: Delete a compute resource.
       text: az cognitiveservices account compute delete -g MyResourceGroup -n MyAccount --compute-name my-compute
+    - name: Delete a compute resource without waiting for completion.
+      text: az cognitiveservices account compute delete -g MyResourceGroup -n MyAccount --compute-name my-compute --no-wait
 """
 
 helps['cognitiveservices account compute show'] = """
@@ -1067,3 +1141,13 @@
     - name: List compute resources.
       text: az cognitiveservices account compute list -g MyResourceGroup -n MyAccount
 """
+
+helps['cognitiveservices account compute wait'] = """
+    type: command
+    short-summary: Wait for a compute resource to reach a specified condition.
+    examples:
+    - name: Wait for a compute resource to finish creating.
+      text: az cognitiveservices account compute wait -g MyResourceGroup -n MyAccount --compute-name my-compute --created
+    - name: Wait for a compute resource to be deleted.
+      text: az cognitiveservices account compute wait -g MyResourceGroup -n MyAccount --compute-name my-compute --deleted
+"""

src/azure-cli/azure/cli/command_modules/cognitiveservices/_params.py

@@ -17,7 +17,7 @@
 )
 from azure.cli.core.util import shell_safe_json_parse, CLIError
 
-from azure.cli.core.commands.validators import validate_tag
+from azure.cli.core.commands.validators import get_default_location_from_resource_group, validate_tag
 from azure.cli.core.decorators import Completer
 
 from azure.cli.command_modules.cognitiveservices._client_factory import cf_resource_skus
@@ -26,6 +26,7 @@
     KeyName,
     DeploymentScaleType,
     HostingModel,
+    VmPriority,
 )
 
 logger = get_logger(__name__)
@@ -690,6 +691,20 @@ def load_arguments(self, _):
     with self.argument_context('cognitiveservices account managed-compute-deployment update') as c:
         c.argument('tags', tags_type)
 
+    with self.argument_context('cognitiveservices account adapter-deployment') as c:
+        c.argument('adapter_deployment_name',
+                   options_list=['--adapter-deployment-name', '--adapter-name'],
+                   help='Adapter deployment name.')
+
+    with self.argument_context('cognitiveservices account adapter-deployment create') as c:
+        c.argument('source_model_id', options_list=['--source-model-id'],
+                   help='Immutable Foundry fine-tuning LoRA output model URI '
+                   '(e.g., azureai://accounts/{account}/projects/{project}/models/{model}/versions/{version}).')
+        c.argument('target_deployment_name', options_list=['--target-deployment-name'],
+                   help='Name of the existing managed compute parent deployment (resolved within '
+                   'the same account) that will serve this adapter. Re-specify to re-target an '
+                   'existing adapter deployment.')
+
     with self.argument_context('cognitiveservices account commitment-plan') as c:
         c.argument('commitment_plan_name', help='Cognitive Services account commitment plan name')
         c.argument('plan_type', help='Cognitive Services account commitment plan type')
@@ -803,13 +818,16 @@ def load_arguments(self, _):
         c.argument('compute_name', help='Cognitive Services account compute name.')
 
     with self.argument_context('cognitiveservices account compute create') as c:
-        c.argument('location', arg_type=get_location_type(self.cli_ctx),
-                   help='Location for the compute resource.')
+        c.argument(
+            'location', arg_type=get_location_type(self.cli_ctx), required=False,
+            validator=get_default_location_from_resource_group,
+            help='Location for the compute resource. If omitted, uses the configured default '
+                 'location or the resource group location.')
         c.argument('pool_name', arg_group='Pool', help='Name of the compute pool.')
         c.argument('instance_type', arg_group='Pool',
                    help='VM instance type for the pool (e.g. Standard_DS3_v2).')
         c.argument('node_count', arg_group='Pool', type=int,
                    help='Number of nodes in the pool.')
-        c.argument('vm_priority', arg_group='Pool',
-                   help='VM priority for the pool (e.g. Dedicated, LowPriority). '
-                        'If omitted, the service default is used.')
+        c.argument(
+            'vm_priority', arg_group='Pool', arg_type=get_enum_type(VmPriority),
+            help='VM priority for the pool.')

src/azure-cli/azure/cli/command_modules/cognitiveservices/commands.py

@@ -8,7 +8,7 @@
     cf_deleted_accounts, cf_deployments, cf_commitment_plans, cf_commitment_tiers, cf_models, cf_usages, \
     cf_ai_projects, cf_account_connections, cf_projects, cf_project_connections, \
     cf_managed_network_settings, cf_managed_network_provisions, cf_outbound_rule, \
-    cf_managed_compute_deployments, cf_computes
+    cf_managed_compute_deployments, cf_adapter_deployments, cf_computes
 
 
 def load_command_table(self, _):
@@ -213,6 +213,19 @@ def load_command_table(self, _):
         g.custom_command('update', 'managed_compute_deployment_update')
         g.custom_command('delete', 'managed_compute_deployment_delete')
 
+    adapter_deployments_type = CliCommandType(
+        operations_tmpl='azure.mgmt.cognitiveservices.operations#AdapterDeploymentsOperations.{}',
+        client_factory=cf_adapter_deployments
+    )
+
+    with self.command_group(
+            'cognitiveservices account adapter-deployment', adapter_deployments_type,
+            client_factory=cf_adapter_deployments, is_preview=True) as g:
+        g.custom_command('create', 'adapter_deployment_create')
+        g.custom_show_command('show', 'adapter_deployment_show')
+        g.custom_command('list', 'adapter_deployment_list')
+        g.custom_command('delete', 'adapter_deployment_delete')
+
     computes_type = CliCommandType(
         operations_tmpl='azure.mgmt.cognitiveservices.operations#ComputesOperations.{}',
         client_factory=cf_computes
@@ -225,3 +238,4 @@ def load_command_table(self, _):
         g.custom_command('delete', 'compute_delete', supports_no_wait=True)
         g.custom_show_command('show', 'compute_show')
         g.custom_command('list', 'compute_list')
+        g.custom_wait_command('wait', 'compute_show')

src/azure-cli/azure/cli/command_modules/cognitiveservices/custom.py

@@ -458,6 +458,43 @@ def managed_compute_deployment_delete(client, resource_group_name, account_name,
     return client.begin_delete(resource_group_name, account_name, deployment_name)
 
 
+def adapter_deployment_create(
+        client, resource_group_name, account_name, adapter_deployment_name,
+        source_model_id, target_deployment_name):
+    """
+    Create (or replace / re-target) a LoRA adapter deployment attached to a managed compute deployment.
+    """
+    from azure.mgmt.cognitiveservices.models import AdapterDeployment, AdapterDeploymentProperties
+    properties = AdapterDeploymentProperties(
+        source_model_id=source_model_id,
+        target_deployment_name=target_deployment_name,
+    )
+    adapter_deployment = AdapterDeployment(properties=properties)
+    return client.begin_create_or_update(
+        resource_group_name, account_name, adapter_deployment_name, adapter_deployment)
+
+
+def adapter_deployment_show(client, resource_group_name, account_name, adapter_deployment_name):
+    """
+    Show an adapter deployment for Azure Cognitive Services account.
+    """
+    return client.get(resource_group_name, account_name, adapter_deployment_name)
+
+
+def adapter_deployment_list(client, resource_group_name, account_name):
+    """
+    List adapter deployments for Azure Cognitive Services account.
+    """
+    return client.list(resource_group_name, account_name)
+
+
+def adapter_deployment_delete(client, resource_group_name, account_name, adapter_deployment_name):
+    """
+    Delete an adapter deployment from Azure Cognitive Services account.
+    """
+    return client.begin_delete(resource_group_name, account_name, adapter_deployment_name)
+
+
 def commitment_plan_create_or_update(
     client,
     resource_group_name,
@@ -2650,6 +2687,7 @@ def compute_begin_create_or_update(
     Create a compute resource for Azure Cognitive Services account.
     """
     properties = ClusterComputeProperties(
+        location=location,
         pools=[
             Pool(
                 name=pool_name,
@@ -2659,11 +2697,7 @@ def compute_begin_create_or_update(
             )
         ],
     )
-    # The service also requires the location inside properties (not just on the outer
-    # Compute envelope). The SDK's ClusterComputeProperties class does not declare
-    # 'location' as a kwarg, so it is set via the underlying dict-backed model.
-    properties["location"] = location
-    resource = Compute(location=location, properties=properties)
+    resource = Compute(properties=properties)
     return sdk_no_wait(
         no_wait,
         client.begin_create_or_update,

src/azure-cli/azure/cli/command_modules/cognitiveservices/linter_exclusions.yml

@@ -111,6 +111,21 @@ cognitiveservices account managed-compute-deployment update:
 cognitiveservices account managed-compute-deployment delete:
   rule_exclusions:
     - missing_command_test_coverage
+cognitiveservices account adapter-deployment:
+  rule_exclusions:
+    - missing_command_test_coverage
+cognitiveservices account adapter-deployment create:
+  rule_exclusions:
+    - missing_command_test_coverage
+cognitiveservices account adapter-deployment show:
+  rule_exclusions:
+    - missing_command_test_coverage
+cognitiveservices account adapter-deployment list:
+  rule_exclusions:
+    - missing_command_test_coverage
+cognitiveservices account adapter-deployment delete:
+  rule_exclusions:
+    - missing_command_test_coverage
 cognitiveservices account compute:
   rule_exclusions:
     - missing_command_test_coverage