Skip to content

feat: Add --multihost-gpu-node-count to gcloud ai endpoints deploy-model command. #4927

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

feat: Add --multihost-gpu-node-count to gcloud ai endpoints deploy-model command. #4927

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
42 changes: 42 additions & 0 deletions google/cloud/aiplatform/models.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -252,6 +252,7 @@ def create(
reservation_affinity_values: Optional[List[str]] = None,
spot: bool = False,
required_replica_count: Optional[int] = 0,
multihost_gpu_node_count: Optional[int] = None,
) -> "DeploymentResourcePool":
"""Creates a new DeploymentResourcePool.

Expand Down Expand Up @@ -332,6 +333,9 @@ def create(
set, the model deploy/mutate operation will succeed once
available_replica_count reaches required_replica_count, and the
rest of the replicas will be retried.
multihost_gpu_node_count (int):
Optional. The number of nodes per replica for multihost GPU DeployedModel.
Required for multihost GPU deployments.

Returns:
DeploymentResourcePool
Expand Down Expand Up @@ -363,6 +367,7 @@ def create(
sync=sync,
create_request_timeout=create_request_timeout,
required_replica_count=required_replica_count,
multihost_gpu_node_count=multihost_gpu_node_count,
)

@classmethod
Expand All @@ -389,6 +394,7 @@ def _create(
sync=True,
create_request_timeout: Optional[float] = None,
required_replica_count: Optional[int] = 0,
multihost_gpu_node_count: Optional[int] = None,
) -> "DeploymentResourcePool":
"""Creates a new DeploymentResourcePool.

Expand Down Expand Up @@ -472,6 +478,9 @@ def _create(
set, the model deploy/mutate operation will succeed once
available_replica_count reaches required_replica_count, and the
rest of the replicas will be retried.
multihost_gpu_node_count (int):
Optional. The number of nodes per replica for multihost GPU DeployedModel.
Required for multihost GPU deployments.

Returns:
DeploymentResourcePool
Expand Down Expand Up @@ -505,6 +514,7 @@ def _create(
[autoscaling_metric_spec]
)

# TODO(joelletiangco): accelerator_type present here
if accelerator_type and accelerator_count:
utils.validate_accelerator_type(accelerator_type)
machine_spec.accelerator_type = accelerator_type
Expand Down Expand Up @@ -1327,6 +1337,7 @@ def deploy(
accelerator_type: Optional[str] = None,
accelerator_count: Optional[int] = None,
tpu_topology: Optional[str] = None,
multihost_gpu_node_count: Optional[int] = None,
service_account: Optional[str] = None,
explanation_metadata: Optional[aiplatform.explain.ExplanationMetadata] = None,
explanation_parameters: Optional[
Expand Down Expand Up @@ -1399,6 +1410,9 @@ def deploy(
tpu_topology (str):
Optional. The TPU topology to use for the DeployedModel.
Required for CloudTPU multihost deployments.
multihost_gpu_node_count (int):
Optional. The number of nodes per replica for multihost GPU DeployedModel.
Required for multihost GPU deployments.
service_account (str):
The service account that the DeployedModel's container runs as. Specify the
email address of the service account. If this service account is not
Expand Down Expand Up @@ -1500,6 +1514,7 @@ def deploy(
accelerator_type=accelerator_type,
accelerator_count=accelerator_count,
tpu_topology=tpu_topology,
multihost_gpu_node_count=multihost_gpu_node_count,
reservation_affinity_type=reservation_affinity_type,
reservation_affinity_key=reservation_affinity_key,
reservation_affinity_values=reservation_affinity_values,
Expand Down Expand Up @@ -1532,6 +1547,7 @@ def _deploy(
accelerator_type: Optional[str] = None,
accelerator_count: Optional[int] = None,
tpu_topology: Optional[str] = None,
multihost_gpu_node_count: Optional[int] = None,
reservation_affinity_type: Optional[str] = None,
reservation_affinity_key: Optional[str] = None,
reservation_affinity_values: Optional[List[str]] = None,
Expand Down Expand Up @@ -1601,6 +1617,9 @@ def _deploy(
tpu_topology (str):
Optional. The TPU topology to use for the DeployedModel.
Required for CloudTPU multihost deployments.
multihost_gpu_node_count (int):
Optional. The number of nodes per replica for multihost GPU DeployedModel.
Required for multihost GPU deployments.
reservation_affinity_type (str):
Optional. The type of reservation affinity.
One of NO_RESERVATION, ANY_RESERVATION, SPECIFIC_RESERVATION,
Expand Down Expand Up @@ -1686,6 +1705,7 @@ def _deploy(
accelerator_type=accelerator_type,
accelerator_count=accelerator_count,
tpu_topology=tpu_topology,
multihost_gpu_node_count=multihost_gpu_node_count,
reservation_affinity_type=reservation_affinity_type,
reservation_affinity_key=reservation_affinity_key,
reservation_affinity_values=reservation_affinity_values,
Expand Down Expand Up @@ -1725,6 +1745,7 @@ def _deploy_call(
accelerator_type: Optional[str] = None,
accelerator_count: Optional[int] = None,
tpu_topology: Optional[str] = None,
multihost_gpu_node_count: Optional[int] = None,
reservation_affinity_type: Optional[str] = None,
reservation_affinity_key: Optional[str] = None,
reservation_affinity_values: Optional[List[str]] = None,
Expand Down Expand Up @@ -1803,6 +1824,9 @@ def _deploy_call(
tpu_topology (str):
Optional. The TPU topology to use for the DeployedModel.
Required for CloudTPU multihost deployments.
multihost_gpu_node_count (int):
Optional. The number of nodes per replica for multihost GPU DeployedModel.
Required for multihost GPU deployments.
reservation_affinity_type (str):
Optional. The type of reservation affinity.
One of NO_RESERVATION, ANY_RESERVATION, SPECIFIC_RESERVATION,
Expand Down Expand Up @@ -2030,6 +2054,9 @@ def _deploy_call(
if tpu_topology is not None:
machine_spec.tpu_topology = tpu_topology

if multihost_gpu_node_count is not None:
machine_spec.multihost_gpu_node_count = multihost_gpu_node_count

dedicated_resources.machine_spec = machine_spec
deployed_model.dedicated_resources = dedicated_resources
if fast_tryout_enabled:
Expand Down Expand Up @@ -4012,6 +4039,7 @@ def deploy(
accelerator_type: Optional[str] = None,
accelerator_count: Optional[int] = None,
tpu_topology: Optional[str] = None,
multihost_gpu_node_count: Optional[int] = None,
service_account: Optional[str] = None,
explanation_metadata: Optional[aiplatform.explain.ExplanationMetadata] = None,
explanation_parameters: Optional[
Expand Down Expand Up @@ -4089,6 +4117,9 @@ def deploy(
tpu_topology (str):
Optional. The TPU topology to use for the DeployedModel.
Required for CloudTPU multihost deployments.
multihost_gpu_node_count (int):
Optional. The number of nodes per replica for multihost GPU DeployedModel.
Required for multihost GPU deployments.
service_account (str):
The service account that the DeployedModel's container runs as. Specify the
email address of the service account. If this service account is not
Expand Down Expand Up @@ -4190,6 +4221,7 @@ def deploy(
accelerator_type=accelerator_type,
accelerator_count=accelerator_count,
tpu_topology=tpu_topology,
multihost_gpu_node_count=multihost_gpu_node_count,
reservation_affinity_type=reservation_affinity_type,
reservation_affinity_key=reservation_affinity_key,
reservation_affinity_values=reservation_affinity_values,
Expand Down Expand Up @@ -5241,6 +5273,7 @@ def deploy(
accelerator_type: Optional[str] = None,
accelerator_count: Optional[int] = None,
tpu_topology: Optional[str] = None,
multihost_gpu_node_count: Optional[int] = None,
service_account: Optional[str] = None,
explanation_metadata: Optional[aiplatform.explain.ExplanationMetadata] = None,
explanation_parameters: Optional[
Expand Down Expand Up @@ -5318,6 +5351,9 @@ def deploy(
tpu_topology (str):
Optional. The TPU topology to use for the DeployedModel.
Requireid for CloudTPU multihost deployments.
multihost_gpu_node_count (int):
Optional. The number of nodes per replica for multihost GPU DeployedModel.
Required for multihost GPU deployments.
service_account (str):
The service account that the DeployedModel's container runs as. Specify the
email address of the service account. If this service account is not
Expand Down Expand Up @@ -5462,6 +5498,7 @@ def deploy(
accelerator_type=accelerator_type,
accelerator_count=accelerator_count,
tpu_topology=tpu_topology,
multihost_gpu_node_count=multihost_gpu_node_count,
reservation_affinity_type=reservation_affinity_type,
reservation_affinity_key=reservation_affinity_key,
reservation_affinity_values=reservation_affinity_values,
Expand Down Expand Up @@ -5505,6 +5542,7 @@ def _deploy(
accelerator_type: Optional[str] = None,
accelerator_count: Optional[int] = None,
tpu_topology: Optional[str] = None,
multihost_gpu_node_count: Optional[int] = None,
reservation_affinity_type: Optional[str] = None,
reservation_affinity_key: Optional[str] = None,
reservation_affinity_values: Optional[List[str]] = None,
Expand Down Expand Up @@ -5579,6 +5617,9 @@ def _deploy(
tpu_topology (str):
Optional. The TPU topology to use for the DeployedModel.
Requireid for CloudTPU multihost deployments.
multihost_gpu_node_count (int):
Optional. The number of nodes per replica for multihost GPU DeployedModel.
Required for multihost GPU deployments.
reservation_affinity_type (str):
Optional. The type of reservation affinity.
One of NO_RESERVATION, ANY_RESERVATION, SPECIFIC_RESERVATION,
Expand Down Expand Up @@ -5713,6 +5754,7 @@ def _deploy(
accelerator_type=accelerator_type,
accelerator_count=accelerator_count,
tpu_topology=tpu_topology,
multihost_gpu_node_count=multihost_gpu_node_count,
reservation_affinity_type=reservation_affinity_type,
reservation_affinity_key=reservation_affinity_key,
reservation_affinity_values=reservation_affinity_values,
Expand Down
52 changes: 52 additions & 0 deletions tests/unit/aiplatform/test_models.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -146,6 +146,11 @@
_TEST_TPU_MACHINE_TYPE = "ct5lp-hightpu-4t"
_TEST_TPU_TOPOLOGY = "2x2"

_TEST_GPU_MACHINE_TYPE = "a3-highgpu-8g"
_TEST_GPU_ACCELERATOR_TYPE = "NVIDIA_TESLA_A100"
_TEST_GPU_ACCELERATOR_COUNT = 8
_TEST_MULTIHOST_GPU_NODE_COUNT = 2

_TEST_BATCH_SIZE = 16

_TEST_PIPELINE_RESOURCE_NAME = (
Expand Down Expand Up @@ -2239,6 +2244,53 @@ def test_deploy_no_endpoint_with_tpu_topology(self, deploy_model_mock, sync):
timeout=None,
)

@pytest.mark.usefixtures(
"get_endpoint_mock", "get_model_mock", "create_endpoint_mock"
)
@pytest.mark.parametrize("sync", [True, False])
def test_deploy_no_endpoint_with_multihost_gpu_node_count(self, deploy_model_mock, sync):
test_model = models.Model(_TEST_ID)
test_model._gca_resource.supported_deployment_resources_types.append(
aiplatform.gapic.Model.DeploymentResourcesType.DEDICATED_RESOURCES
)
test_endpoint = test_model.deploy(
machine_type=_TEST_GPU_MACHINE_TYPE,
accelerator_type=_TEST_GPU_ACCELERATOR_TYPE,
accelerator_count=_TEST_GPU_ACCELERATOR_COUNT,
multihost_gpu_node_count=_TEST_MULTIHOST_GPU_NODE_COUNT,
sync=sync,
deploy_request_timeout=None,
)

if not sync:
test_endpoint.wait()

expected_machine_spec = gca_machine_resources.MachineSpec(
machine_type=_TEST_GPU_MACHINE_TYPE,
accelerator_type=_TEST_GPU_ACCELERATOR_COUNT,
accelerator_count=_TEST_GPU_ACCELERATOR_COUNT,
multihost_gpu_node_count=_TEST_MULTIHOST_GPU_NODE_COUNT,
)
expected_dedicated_resources = gca_machine_resources.DedicatedResources(
machine_spec=expected_machine_spec,
min_replica_count=1,
max_replica_count=1,
spot=False,
)
expected_deployed_model = gca_endpoint.DeployedModel(
dedicated_resources=expected_dedicated_resources,
model=test_model.resource_name,
display_name=None,
)
deploy_model_mock.assert_called_once_with(
endpoint=test_endpoint.resource_name,
deployed_model=expected_deployed_model,
traffic_split={"0": 100},
metadata=(),
timeout=None,
)


@pytest.mark.usefixtures(
"get_endpoint_mock", "get_model_mock", "create_endpoint_mock"
)
Expand Down