def _Run(args, version):
"""Create a new Vertex AI endpoint."""
validation.ValidateDisplayName(args.display_name)
region_ref = args.CONCEPTS.region.Parse()
args.region = region_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version,
region=args.region):
endpoints_client = client.EndpointsClient(version=version)
operation_client = operations.OperationsClient()
if version == constants.GA_VERSION:
op = endpoints_client.Create(
region_ref, args.display_name,
labels_util.ParseCreateArgs(
args, endpoints_client.messages.
GoogleCloudAiplatformV1Endpoint.LabelsValue),
args.description, args.network, args.endpoint_id)
else:
op = endpoints_client.CreateBeta(
region_ref, args.display_name,
labels_util.ParseCreateArgs(
args, endpoints_client.messages.
GoogleCloudAiplatformV1beta1Endpoint.LabelsValue),
args.description, args.network, args.endpoint_id)
response_msg = operations_util.WaitForOpMaybe(
operation_client, op, endpoints_util.ParseOperation(op.name))
if response_msg is not None:
response = encoding.MessageToPyValue(response_msg)
if 'name' in response:
log.status.Print(('Created Vertex AI endpoint: {}.').format(
response['name']))
return response_msg
def _Run(args, version):
"""Run Vertex AI online prediction."""
endpoint_ref = args.CONCEPTS.endpoint.Parse()
args.region = endpoint_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version,
region=args.region):
if args.request.startswith('@'):
request = console_io.ReadFromFileOrStdin(args.request[1:],
binary=True)
else:
request = args.request.encode('utf-8')
endpoints_client = client.EndpointsClient(version=version)
_, response = endpoints_client.RawPredict(endpoint_ref,
args.http_headers, request)
# Workaround since gcloud only supports protobufs as JSON objects. Since
# raw predict can return anything, write raw bytes to stdout.
if not args.IsSpecified('format'):
sys.stdout.buffer.write(response)
return None
# If user asked for formatting, assume it's a JSON object.
try:
return json.loads(response.decode('utf-8'))
except ValueError:
raise core_exceptions.Error(
'No JSON object could be decoded from the '
'HTTP response body:\n' + six.text_type(response))
def _Run(args, version):
"""Deploy a model to an existing Vertex AI endpoint."""
validation.ValidateDisplayName(args.display_name)
if version != constants.GA_VERSION:
validation.ValidateAutoscalingMetricSpecs(
args.autoscaling_metric_specs)
endpoint_ref = args.CONCEPTS.endpoint.Parse()
args.region = endpoint_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version,
region=args.region):
endpoints_client = client.EndpointsClient(version=version)
operation_client = operations.OperationsClient()
if version == constants.GA_VERSION:
op = endpoints_client.DeployModel(
endpoint_ref,
args.model,
args.region,
args.display_name,
machine_type=args.machine_type,
accelerator_dict=args.accelerator,
min_replica_count=args.min_replica_count,
max_replica_count=args.max_replica_count,
enable_access_logging=args.enable_access_logging,
disable_container_logging=args.disable_container_logging,
service_account=args.service_account,
traffic_split=args.traffic_split,
deployed_model_id=args.deployed_model_id)
else:
op = endpoints_client.DeployModelBeta(
endpoint_ref,
args.model,
args.region,
args.display_name,
machine_type=args.machine_type,
accelerator_dict=args.accelerator,
min_replica_count=args.min_replica_count,
max_replica_count=args.max_replica_count,
autoscaling_metric_specs=args.autoscaling_metric_specs,
enable_access_logging=args.enable_access_logging,
enable_container_logging=args.enable_container_logging,
service_account=args.service_account,
traffic_split=args.traffic_split,
deployed_model_id=args.deployed_model_id)
response_msg = operations_util.WaitForOpMaybe(
operation_client, op, endpoints_util.ParseOperation(op.name))
if response_msg is not None:
response = encoding.MessageToPyValue(response_msg)
if 'deployedModel' in response and 'id' in response[
'deployedModel']:
log.status.Print(('Deployed a model to the endpoint {}. '
'Id of the deployed model: {}.').format(
endpoint_ref.AsDict()['endpointsId'],
response['deployedModel']['id']))
return response_msg
def _Run(args, version):
"""Undeploy a model fro man existing AI Platform endpoint."""
endpoint_ref = args.CONCEPTS.endpoint.Parse()
args.region = endpoint_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version,
region=args.region):
endpoints_client = client.EndpointsClient(version=version)
operation_client = operations.OperationsClient()
op = endpoints_client.UndeployModelBeta(endpoint_ref, args)
return operations_util.WaitForOpMaybe(
operation_client, op, endpoints_util.ParseOperation(op.name))
def _Run(args, version):
"""Delete an existing Vertex AI endpoint."""
endpoint_ref = args.CONCEPTS.endpoint.Parse()
args.region = endpoint_ref.AsDict()['locationsId']
endpoint_id = endpoint_ref.AsDict()['endpointsId']
with endpoint_util.AiplatformEndpointOverrides(version,
region=args.region):
endpoints_client = client.EndpointsClient(version=version)
operation_client = operations.OperationsClient()
console_io.PromptContinue(
'This will delete endpoint [{}]...'.format(endpoint_id),
cancel_on_no=True)
op = endpoints_client.Delete(endpoint_ref)
return operations_util.WaitForOpMaybe(
operation_client, op, endpoints_util.ParseOperation(op.name))
def _Run(args, version):
"""Run AI Platform online prediction."""
endpoint_ref = args.CONCEPTS.endpoint.Parse()
args.region = endpoint_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(
version, region=args.region, is_prediction=True):
endpoints_client = client.EndpointsClient(version=version)
instances_json = endpoints_util.ReadInstancesFromArgs(args.json_request)
results = endpoints_client.PredictBeta(endpoint_ref, instances_json)
if not args.IsSpecified('format'):
# default format is based on the response.
args.format = endpoints_util.GetDefaultFormat(results.predictions)
return results
def _Run(args, version):
"""Update an existing Vertex AI endpoint."""
validation.ValidateDisplayName(args.display_name)
endpoint_ref = args.CONCEPTS.endpoint.Parse()
args.region = endpoint_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version,
region=args.region):
endpoints_client = client.EndpointsClient(version=version)
def GetLabels():
return endpoints_client.Get(endpoint_ref).labels
try:
if version == constants.GA_VERSION:
op = endpoints_client.Patch(
endpoint_ref,
labels_util.ProcessUpdateArgsLazy(
args, endpoints_client.messages.
GoogleCloudAiplatformV1Endpoint.LabelsValue,
GetLabels),
display_name=args.display_name,
description=args.description,
traffic_split=args.traffic_split,
clear_traffic_split=args.clear_traffic_split)
else:
op = endpoints_client.PatchBeta(
endpoint_ref,
labels_util.ProcessUpdateArgsLazy(
args, endpoints_client.messages.
GoogleCloudAiplatformV1beta1Endpoint.LabelsValue,
GetLabels),
display_name=args.display_name,
description=args.description,
traffic_split=args.traffic_split,
clear_traffic_split=args.clear_traffic_split)
except errors.NoFieldsSpecifiedError:
available_update_args = [
'display_name', 'traffic_split', 'clear_traffic_split',
'update_labels', 'clear_labels', 'remove_labels', 'description'
]
if not any(args.IsSpecified(arg) for arg in available_update_args):
raise
log.status.Print('No update to perform.')
return None
else:
log.UpdatedResource(op.name, kind='Vertex AI endpoint')
return op
def _Run(args, version):
"""Create a new AI Platform endpoint."""
validation.ValidateDisplayName(args.display_name)
region_ref = args.CONCEPTS.region.Parse()
args.region = region_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version,
region=args.region):
endpoints_client = client.EndpointsClient(version=version)
operation_client = operations.OperationsClient()
op = endpoints_client.CreateBeta(region_ref, args)
response_msg = operations_util.WaitForOpMaybe(
operation_client, op, endpoints_util.ParseOperation(op.name))
if response_msg is not None:
response = encoding.MessageToPyValue(response_msg)
if 'name' in response:
log.status.Print(('Created AI Platform endpoint: {}.').format(
response['name']))
return response_msg
def _Run(args, version):
"""Undeploy a model fro man existing Vertex AI endpoint."""
endpoint_ref = args.CONCEPTS.endpoint.Parse()
args.region = endpoint_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version, region=args.region):
endpoints_client = client.EndpointsClient(version=version)
operation_client = operations.OperationsClient()
if version == constants.GA_VERSION:
op = endpoints_client.UndeployModel(
endpoint_ref,
args.deployed_model_id,
traffic_split=args.traffic_split)
else:
op = endpoints_client.UndeployModelBeta(
endpoint_ref,
args.deployed_model_id,
traffic_split=args.traffic_split)
return operations_util.WaitForOpMaybe(
operation_client, op, endpoints_util.ParseOperation(op.name))
def _Run(args, version):
"""Run AI Platform online explanation."""
endpoint_ref = args.CONCEPTS.endpoint.Parse()
args.region = endpoint_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(
version, region=args.region, is_prediction=True):
endpoints_client = client.EndpointsClient(version=version)
instances_json = endpoints_util.ReadInstancesFromArgs(args.json_request)
results = endpoints_client.ExplainBeta(endpoint_ref, instances_json, args)
if getattr(results, 'deployedModelId') is not None:
log.status.Print(
'Deployed model id to be used for explanation: {}'.format(
results.deployedModelId))
if not args.IsSpecified('format'):
# default format is based on the response.
args.format = endpoints_util.GetDefaultFormat(
results, key_name='explanations')
return results
def _Run(args, version):
"""Deploy a model to an existing AI Platform endpoint."""
validation.ValidateDisplayName(args.display_name)
endpoint_ref = args.CONCEPTS.endpoint.Parse()
args.region = endpoint_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version,
region=args.region):
endpoints_client = client.EndpointsClient(version=version)
operation_client = operations.OperationsClient()
op = endpoints_client.DeployModelBeta(endpoint_ref, args)
response_msg = operations_util.WaitForOpMaybe(
operation_client, op, endpoints_util.ParseOperation(op.name))
if response_msg is not None:
response = encoding.MessageToPyValue(response_msg)
if 'deployedModel' in response and 'id' in response[
'deployedModel']:
log.status.Print(('Deployed a model to the endpoint {}. '
'Id of the deployed model: {}.').format(
endpoint_ref.AsDict()['endpointsId'],
response['deployedModel']['id']))
return response_msg
def _Run(args, version):
"""Update an existing AI Platform endpoint."""
validation.ValidateDisplayName(args.display_name)
endpoint_ref = args.CONCEPTS.endpoint.Parse()
args.region = endpoint_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version,
region=args.region):
endpoints_client = client.EndpointsClient(version=version)
try:
op = endpoints_client.PatchBeta(endpoint_ref, args)
except errors.NoFieldsSpecifiedError:
available_update_args = [
'display_name', 'traffic_split', 'clear_traffic_split',
'update_labels', 'clear_labels', 'remove_labels', 'description'
]
if not any(args.IsSpecified(arg) for arg in available_update_args):
raise
log.status.Print('No update to perform.')
return None
else:
log.UpdatedResource(op.name, kind='AI Platform endpoint')
return op
def _Run(args, version):
region_ref = args.CONCEPTS.region.Parse()
args.region = region_ref.AsDict()['locationsId']
with endpoint_util.AiplatformEndpointOverrides(version,
region=args.region):
return client.EndpointsClient(version=version).List(region_ref)