gcloud alpha compute tpus create - create a new Cloud TPU
gcloud alpha compute tpus create (TPU : --zone=ZONE) --version=VERSION [--accelerator-type=ACCELERATOR_TYPE; default="v2-8"] [--async] [--description=DESCRIPTION] [--network=NETWORK; default="default"] [--preemptible] [--range=RANGE] [--reserved] [--use-service-networking] [--model-base-path=MODEL_BASE_PATH --model-name=MODEL_NAME | --model-config-file=MODEL_CONFIG_FILE --platform-config-file=PLATFORM_CONFIG_FILE] [GCLOUD_WIDE_FLAG ...]
(ALPHA) Create a new Cloud TPU.
The following command creates a TPU with ID my-tpu in the default user project, network and compute/region (with other defaults supplied by API):
$ gcloud alpha compute tpus create my-tpu
The following command creates a TPU with ID my-tpu with explicit values for all required and optional parameters:
$ gcloud alpha compute tpus create my-tpu --zone=us-central1-a \ --range='10.240.0.0/29' --accelerator-type='v2-8' \ --network=my-tf-network --description='My TF Node' \ --version='1.1'
- Tpu resource - The Cloud TPU you want to create. The arguments in this group
can be used to specify the attributes of this resource. (NOTE) Some attributes are not given arguments in this group but can be set in other ways. To set the project attribute:
- —
provide the argument tpu on the command line with a fully specified name;
- —
provide the argument --project on the command line;
- —
set the property core/project.
This must be specified.
- TPU
ID of the tpu or fully qualified identifier for the tpu. To set the tpu attribute:
provide the argument tpu on the command line.
This positional argument must be specified if any of the other arguments in this group are specified.
- --zone=ZONE
The compute/zone of the Cloud TPU.
If not specified, will use default compute/zone.
To set the zone attribute:
provide the argument tpu on the command line with a fully specified name;
provide the argument --zone on the command line;
set the property compute/zone.
- --version=VERSION
TensorFlow version for the TPU, such as 1.14. For a list of available TensorFlow versions please see https://www.tensorflow.org/versions/.
- --accelerator-type=ACCELERATOR_TYPE; default="v2-8"
TPU accelerator type for the TPU. If not specified, this defaults to v2-8.
For a list of available accelerator types run:
gcloud alpha compute tpus accelerator-types list
- --async
Return immediately, without waiting for the operation in progress to complete.
- --description=DESCRIPTION
Specifies a text description of the TPU.
- --network=NETWORK; default="default"
Specifies the network that this TPU will be a part of.
- --preemptible
If provided, the TPU will be preemptible and time-limited. It may be preempted to free up resources for standard TPUs, and will only be able to run for a limited amount of time.
Preemptible TPUs cannot be restarted.
- --range=RANGE
CIDR Range for the TPU.
The IP range that the TPU will select an IP address from. Must be in CIDR notation and a /29 range, for example 192.168.0.0/29. Errors will occur if the CIDR range has already been used for a currently existing TPU, the CIDR range conflicts with any networks in the user's provided network, or the provided network is peered with another network that is using that CIDR range.
- --reserved
When specified, will attempt to create the TPU node under reservations made in the current project. The reservations can be made separately but used in aggregated form. i.e., the user can make a reservation of 128 V2 TPUs and later on make another reservation of 128 V2 TPUs then creates a v2-256 TPU instance. If there exists no reservation or not sufficient amount of reserved cores under the project, the request will fail due to lack of capacity.
- --use-service-networking
If provided, the TPU will be configured via the Service Networking (SN) API instead of using a CIDR range. The Service Networking API should be enabled on the project before creating the TPU.
For more information on Service Networking see https://cloud.google.com/service-infrastructure/docs/service-networking/getting-started.
- At most one of these can be specified:
- --model-base-path=MODEL_BASE_PATH
Base path to exported saved model. This flag can be used instead of '--model-config-file' directly to specify the exported saved model's base path (excluding timestamp), whereas '--model-config-file' points to a Google Cloud Storage path to a ModelServerConfig proto.
- --model-name=MODEL_NAME
Model name for tensorflow serving to serve to incoming requests. If not provided, 'serving_default' will be used.
- --model-config-file=MODEL_CONFIG_FILE
Google Cloud Storage URI of the configuration file for models to serve. For example: 'gs://my_bucket/path/to/config.pbtxt'.
The contents of the model configuration file should look something like the following:
model_config_list: { config: { name: "Model1", base_path: "gs://my_bucket/path/to/model1", model_platform: "tensorflow" }, config: { name: "Model2", base_path: "gs://my_other_bucket/path/to/model2", model_platform: "tensorflow" }, }
- --platform-config-file=PLATFORM_CONFIG_FILE
Google Cloud Storage URI of configuration file for platform requirements. For example: 'gs://my_bucket/path/to/platform.pbtxt'.
The contents of the platform configuration file should look something like the following:
platform_configs { key: "tensorflow" value { source_adapter_config { [type.googleapis.com/tensorflow.serving.SavedModelBundleSourceAdapterConfig] { legacy_config { saved_model_tags: "tpu" saved_model_tags: "serve" batching_parameters { max_batch_size { value: 8 } batch_timeout_micros { value: 50000 } max_enqueued_batches { value: 10000 } num_batch_threads { value: 8 } } } } } } }
These flags are available to all commands: --access-token-file, --account, --billing-project, --configuration, --flags-file, --flatten, --format, --help, --impersonate-service-account, --log-http, --project, --quiet, --trace-token, --user-output-enabled, --verbosity.
Run $ gcloud help for details.
This command uses the tpu/v1alpha1 API. The full documentation for this API can be found at: https://cloud.google.com/tpu/
This command is currently in alpha and might change without notice. If this command fails with API permission errors despite specifying the correct project, you might be trying to access an API with an invitation-only early access allowlist. These variants are also available:
$ gcloud compute tpus create
$ gcloud beta compute tpus create