{Cognitive Services} Manage Compute Deployment Support

src/azure-cli/HISTORY.rst

@@ -268,6 +268,7 @@ Release History
 * `az cognitiveservices agent logs show`: Add console log streaming for hosted agents (#32701)
 * `az cognitiveservices agent create`: Add `--show-logs` flag for deployment troubleshooting (#32701)
 * `az cognitiveservices agent start`: Add `--show-logs` and `--timeout` flags (#32701)
+* [PREVIEW] `az cognitiveservices account managed-compute-deployment`: Add new command group for managing GPU-backed managed compute deployments with create, show, list, update, and delete operations
 
 **Container app**
 

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

@@ -113,5 +113,9 @@ def cf_project_capability_hosts(cli_ctx, *_):
     return get_cognitiveservices_management_client(cli_ctx).project_capability_hosts
 
 
+def cf_managed_compute_deployments(cli_ctx, *_):
+    return get_cognitiveservices_management_client(cli_ctx).managed_compute_deployments
+
+
 def cf_project_connections(cli_ctx, *_):
     return get_cognitiveservices_management_client(cli_ctx).project_connections

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

@@ -371,6 +371,86 @@
     text: az cognitiveservices account deployment list -g yuanyang-test-sdk -n yytest-oai
 """
 
+helps[
+    "cognitiveservices account managed-compute-deployment"
+] = """
+type: group
+short-summary: Manage managed compute deployments for Azure Cognitive Services accounts.
+"""
+
+helps[
+    "cognitiveservices account managed-compute-deployment create"
+] = """
+type: command
+short-summary: Create a managed compute deployment for Azure Cognitive Services account.
+long-summary: Create a GPU-backed managed compute deployment associated with a Cognitive Services account.
+examples:
+  - name: Create a managed compute deployment.
+    text: >
+        az cognitiveservices account managed-compute-deployment create
+        -g myResourceGroup -n myAccount
+        --deployment-name gpt-oss-120b-gpu
+        --model "azureml://registries/azureml-openai-oss/models/gpt-oss-120b/versions/4"
+        --deployment-template "azureml://registries/azureml-openai-oss/deploymenttemplates/gpt-oss-120b-short-context/versions/1"
+        --accelerator-type H100_80GB
+        --sku-name GlobalManagedCompute
+        --sku-capacity 1
+        --tags environment=production team=nlp
+"""
+
+helps[
+    "cognitiveservices account managed-compute-deployment show"
+] = """
+type: command
+short-summary: Show a managed compute deployment for Azure Cognitive Services account.
+examples:
+  - name: Show a managed compute deployment.
+    text: >
+        az cognitiveservices account managed-compute-deployment show
+        -g myResourceGroup -n myAccount
+        --deployment-name gpt-oss-120b-gpu
+"""
+
+helps[
+    "cognitiveservices account managed-compute-deployment list"
+] = """
+type: command
+short-summary: List all managed compute deployments for Azure Cognitive Services account.
+examples:
+  - name: List all managed compute deployments.
+    text: >
+        az cognitiveservices account managed-compute-deployment list
+        -g myResourceGroup -n myAccount
+"""
+
+helps[
+    "cognitiveservices account managed-compute-deployment update"
+] = """
+type: command
+short-summary: Update a managed compute deployment for Azure Cognitive Services account.
+long-summary: Only SKU (name/capacity) and tags can be updated. Model and accelerator type are immutable after creation.
+examples:
+  - name: Update SKU capacity of a managed compute deployment.
+    text: >
+        az cognitiveservices account managed-compute-deployment update
+        -g myResourceGroup -n myAccount
+        --deployment-name gpt-oss-120b-gpu
+        --sku-capacity 2
+"""
+
+helps[
+    "cognitiveservices account managed-compute-deployment delete"
+] = """
+type: command
+short-summary: Delete a managed compute deployment from Azure Cognitive Services account.
+examples:
+  - name: Delete a managed compute deployment.
+    text: >
+        az cognitiveservices account managed-compute-deployment delete
+        -g myResourceGroup -n myAccount
+        --deployment-name gpt-oss-120b-gpu
+"""
+
 helps[
     "cognitiveservices commitment-tier"
 ] = """

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

@@ -672,6 +672,24 @@ def load_arguments(self, _):
             'scale_settings_capacity', options_list=['--scale-capacity', '--scale-settings-capacity'],
             help='Cognitive Services account deployment scale settings capacity.')
 
+    with self.argument_context('cognitiveservices account managed-compute-deployment') as c:
+        c.argument('deployment_name', help='Managed compute deployment name.')
+
+    with self.argument_context('cognitiveservices account managed-compute-deployment create') as c:
+        c.argument('model', help='AzureML registry model URI '
+                   '(e.g., azureml://registries/{registry}/models/{model}/versions/{version}).')
+        c.argument('deployment_template', options_list=['--deployment-template'],
+                   help='AzureML registry deployment template URI '
+                   '(e.g., azureml://registries/{registry}/deploymenttemplates/{template}/versions/{version}).')
+        c.argument('accelerator_type', options_list=['--accelerator-type'],
+                   help='GPU accelerator type (e.g., H100_80GB).')
+        c.argument('version_upgrade_option', options_list=['--version-upgrade-option'],
+                   help='Version upgrade policy. Allowed values: OnceNewDefaultVersionAvailable, '
+                   'OnceCurrentVersionExpired, NoAutoUpgrade.')
+
+    with self.argument_context('cognitiveservices account managed-compute-deployment update') as c:
+        c.argument('tags', tags_type)
+
     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')

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

@@ -7,7 +7,8 @@
 from azure.cli.command_modules.cognitiveservices._client_factory import cf_accounts, cf_resource_skus, \
     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_network_settings, cf_managed_network_provisions, cf_outbound_rule, \
+    cf_managed_compute_deployments
 
 
 def load_command_table(self, _):
@@ -197,3 +198,17 @@ def load_command_table(self, _):
             setter_name='update',
             setter_arg_name='connection',
             custom_func_name='account_connection_update')
+
+    managed_compute_deployments_type = CliCommandType(
+        operations_tmpl='azure.mgmt.cognitiveservices.operations#ManagedComputeDeploymentsOperations.{}',
+        client_factory=cf_managed_compute_deployments
+    )
+
+    with self.command_group(
+            'cognitiveservices account managed-compute-deployment', managed_compute_deployments_type,
+            client_factory=cf_managed_compute_deployments, is_preview=True) as g:
+        g.custom_command('create', 'managed_compute_deployment_create')
+        g.custom_command('show', 'managed_compute_deployment_show')
+        g.custom_command('list', 'managed_compute_deployment_list')
+        g.custom_command('update', 'managed_compute_deployment_update')
+        g.custom_command('delete', 'managed_compute_deployment_delete')

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

@@ -394,6 +394,68 @@ def deployment_begin_create_or_update(
     return client.begin_create_or_update(resource_group_name, account_name, deployment_name, dpy, polling=False)
 
 
+def managed_compute_deployment_create(
+        client, resource_group_name, account_name, deployment_name,
+        model, deployment_template=None, accelerator_type=None,
+        version_upgrade_option=None,
+        sku_name=None, sku_capacity=None, tags=None):
+    """
+    Create a managed compute deployment for Azure Cognitive Services account.
+    """
+    from azure.mgmt.cognitiveservices.models import ManagedComputeDeployment, ManagedComputeDeploymentProperties
+    properties = ManagedComputeDeploymentProperties(
+        model=model,
+        deployment_template=deployment_template,
+        accelerator_type=accelerator_type,
+        version_upgrade_option=version_upgrade_option,
+    )
+    deployment = ManagedComputeDeployment(properties=properties)
+    if sku_name is not None:
+        deployment.sku = Sku(name=sku_name, capacity=sku_capacity)
+    if tags is not None:
+        deployment.tags = tags
+    return client.begin_create_or_update(
+        resource_group_name, account_name, deployment_name, deployment)
+
+
+def managed_compute_deployment_update(
+        client, resource_group_name, account_name, deployment_name,
+        sku_name=None, sku_capacity=None, tags=None):
+    """
+    Update a managed compute deployment for Azure Cognitive Services account.
+    Only SKU (name/capacity) and tags can be updated.
+    """
+    from azure.mgmt.cognitiveservices.models import PatchResourceSku
+    patch = PatchResourceSku()
+    if sku_name is not None or sku_capacity is not None:
+        patch.sku = Sku(name=sku_name, capacity=sku_capacity)
+    if tags is not None:
+        patch.tags = tags
+    return client.begin_update(
+        resource_group_name, account_name, deployment_name, patch)
+
+
+def managed_compute_deployment_show(client, resource_group_name, account_name, deployment_name):
+    """
+    Show a managed compute deployment for Azure Cognitive Services account.
+    """
+    return client.get(resource_group_name, account_name, deployment_name)
+
+
+def managed_compute_deployment_list(client, resource_group_name, account_name):
+    """
+    List managed compute deployments for Azure Cognitive Services account.
+    """
+    return client.list(resource_group_name, account_name)
+
+
+def managed_compute_deployment_delete(client, resource_group_name, account_name, deployment_name):
+    """
+    Delete a managed compute deployment from Azure Cognitive Services account.
+    """
+    return client.begin_delete(resource_group_name, account_name, deployment_name)
+
+
 def commitment_plan_create_or_update(
     client,
     resource_group_name,

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

@@ -88,5 +88,23 @@ cognitiveservices agent logs bulk-set:
   rule_exclusions:
     - missing_command_test_coverage
 cognitiveservices agent logs remove:
+  rule_exclusions:
+    - missing_command_test_coverage
+cognitiveservices account managed-compute-deployment:
+  rule_exclusions:
+    - missing_command_test_coverage
+cognitiveservices account managed-compute-deployment create:
+  rule_exclusions:
+    - missing_command_test_coverage
+cognitiveservices account managed-compute-deployment show:
+  rule_exclusions:
+    - missing_command_test_coverage
+cognitiveservices account managed-compute-deployment list:
+  rule_exclusions:
+    - missing_command_test_coverage
+cognitiveservices account managed-compute-deployment update:
+  rule_exclusions:
+    - missing_command_test_coverage
+cognitiveservices account managed-compute-deployment delete:
   rule_exclusions:
     - missing_command_test_coverage

src/azure-cli/azure/cli/command_modules/cognitiveservices/tests/latest/test_managed_compute_deployment.py

@@ -0,0 +1,100 @@
+# --------------------------------------------------------------------------------------------
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT License. See License.txt in the project root for license information.
+# --------------------------------------------------------------------------------------------
+
+import unittest
+
+from azure.cli.testsdk import ScenarioTest, ResourceGroupPreparer
+from azure.cli.testsdk.decorators import serial_test
+
+
+class CognitiveServicesManagedComputeDeploymentTests(ScenarioTest):
+    @serial_test()
+    @ResourceGroupPreparer()
+    def test_cognitiveservices_managed_compute_deployment(self, resource_group):
+        sname = self.create_random_name(prefix='cs_cli_test_', length=16)
+
+        self.kwargs.update({
+            'sname': sname,
+            'kind': 'AIServices',
+            'sku': 'S0',
+            'location': 'eastus',
+            'deployment_name': 'test-mcd',
+            'model': 'azureml://registries/azureml-openai-oss/models/gpt-oss-120b/versions/4',
+            'deployment_template': 'azureml://registries/azureml-openai-oss/deploymenttemplates/'
+                                   'gpt-oss-120b-short-context/versions/1',
+            'accelerator_type': 'H100_80GB',
+            'sku_name': 'GlobalManagedCompute',
+            'sku_capacity': '1',
+        })
+
+        # create cognitive services account
+        self.cmd(
+            'az cognitiveservices account create -n {sname} -g {rg} '
+            '--kind {kind} --sku {sku} -l {location} --yes',
+            checks=[
+                self.check('name', '{sname}'),
+                self.check('properties.provisioningState', 'Succeeded'),
+            ])
+
+        # list should be empty initially
+        self.cmd(
+            'az cognitiveservices account managed-compute-deployment list '
+            '-n {sname} -g {rg}',
+            checks=[self.check('length(@)', 0)])
+
+        # create managed compute deployment
+        self.cmd(
+            'az cognitiveservices account managed-compute-deployment create '
+            '-n {sname} -g {rg} '
+            '--deployment-name {deployment_name} '
+            '--model "{model}" '
+            '--deployment-template "{deployment_template}" '
+            '--accelerator-type {accelerator_type} '
+            '--sku-name {sku_name} '
+            '--sku-capacity {sku_capacity} '
+            '--tags environment=test')
+
+        # show the deployment
+        self.cmd(
+            'az cognitiveservices account managed-compute-deployment show '
+            '-n {sname} -g {rg} '
+            '--deployment-name {deployment_name}',
+            checks=[
+                self.check('name', '{deployment_name}'),
+                self.check('properties.model', '{model}'),
+                self.check('sku.name', '{sku_name}'),
+            ])
+
+        # list should contain the deployment
+        self.cmd(
+            'az cognitiveservices account managed-compute-deployment list '
+            '-n {sname} -g {rg}',
+            checks=[self.check('length(@)', 1)])
+
+        # update sku capacity
+        self.cmd(
+            'az cognitiveservices account managed-compute-deployment update '
+            '-n {sname} -g {rg} '
+            '--deployment-name {deployment_name} '
+            '--sku-capacity 2')
+
+        # delete the deployment
+        self.cmd(
+            'az cognitiveservices account managed-compute-deployment delete '
+            '-n {sname} -g {rg} '
+            '--deployment-name {deployment_name}')
+
+        # verify deletion
+        self.cmd(
+            'az cognitiveservices account managed-compute-deployment list '
+            '-n {sname} -g {rg}',
+            checks=[self.check('length(@)', 0)])
+
+        # cleanup
+        self.cmd('az cognitiveservices account delete -n {sname} -g {rg}')
+
+
+if __name__ == '__main__':
+    unittest.main()