CustomJobSpec

Represents the spec of a CustomJob.

Fields
persistentResourceId string

Optional. The id of the PersistentResource in the same Project and Location which to run

If this is specified, the job will be run on existing machines held by the PersistentResource instead of on-demand short-live machines. The network and CMEK configs on the job should be consistent with those on the PersistentResource, otherwise, the job will be rejected.

workerPoolSpecs[] object (WorkerPoolSpec)

Required. The spec of the worker pools including machine type and Docker image. All worker pools except the first one are optional and can be skipped by providing an empty value.

scheduling object (Scheduling)

Scheduling options for a CustomJob.

serviceAccount string

Specifies the service account for workload run-as account. Users submitting jobs must have act-as permission on this run-as account. If unspecified, the Agent Platform Custom code service Agent for the CustomJob's project is used.

network string

Optional. The full name of the Compute Engine network to which the Job should be peered. For example, projects/12345/global/networks/myVPC. Format is of the form projects/{project}/global/networks/{network}. Where {project} is a project number, as in 12345, and {network} is a network name.

To specify this field, you must have already configured VPC Network Peering for Agent Platform.

If this field is left unspecified, the job is not peered with any network.

reservedIpRanges[] string

Optional. A list of names for the reserved ip ranges under the VPC network that can be used for this job.

If set, we will deploy the job within the provided ip ranges. Otherwise, the job will be deployed to any ip ranges under the provided VPC network.

Example: ['vertex-ai-ip-range'].

pscInterfaceConfig object (PscInterfaceConfig)

Optional. Configuration for PSC-I for CustomJob.

baseOutputDirectory object (GcsDestination)

The Cloud Storage location to store the output of this CustomJob or HyperparameterTuningJob. For HyperparameterTuningJob, the baseOutputDirectory of each child CustomJob backing a Trial is set to a subdirectory of name id under its parent HyperparameterTuningJob's baseOutputDirectory.

The following Agent Platform environment variables will be passed to containers or python modules when this field is set:

For CustomJob:

  • AIP_MODEL_DIR = <baseOutputDirectory>/model/
  • AIP_CHECKPOINT_DIR = <baseOutputDirectory>/checkpoints/
  • AIP_TENSORBOARD_LOG_DIR = <baseOutputDirectory>/logs/

For CustomJob backing a Trial of HyperparameterTuningJob:

  • AIP_MODEL_DIR = <baseOutputDirectory>/<trial_id>/model/
  • AIP_CHECKPOINT_DIR = <baseOutputDirectory>/<trial_id>/checkpoints/
  • AIP_TENSORBOARD_LOG_DIR = <baseOutputDirectory>/<trial_id>/logs/
protectedArtifactLocationId string

The id of the location to store protected artifacts. e.g. us-central1. Populate only when the location is different than CustomJob location. List of supported locations: https://cloud.google.com/vertex-ai/docs/general/locations

tensorboard string

Optional. The name of a Agent Platform Tensorboard resource to which this CustomJob will upload Tensorboard logs. Format: projects/{project}/locations/{location}/tensorboards/{tensorboard}

enableWebAccess boolean

Optional. Whether you want Agent Platform to enable interactive shell access to training containers.

If set to true, you can access interactive shells at the URIs given by CustomJob.web_access_uris or Trial.web_access_uris (within HyperparameterTuningJob.trials).

enableDashboardAccess boolean

Optional. Whether you want Agent Platform to enable access to the customized dashboard in training chief container.

If set to true, you can access the dashboard at the URIs given by CustomJob.web_access_uris or Trial.web_access_uris (within HyperparameterTuningJob.trials).

experiment string

Optional. The Experiment associated with this job. Format: projects/{project}/locations/{location}/metadataStores/{metadataStores}/contexts/{experiment-name}

experimentRun string

Optional. The Experiment Run associated with this job. Format: projects/{project}/locations/{location}/metadataStores/{metadataStores}/contexts/{experiment-name}-{experiment-run-name}

models[] string

Optional. The name of the Model resources for which to generate a mapping to artifact URIs. Applicable only to some of the Google-provided custom jobs. Format: projects/{project}/locations/{location}/models/{model}

In order to retrieve a specific version of the model, also provide the version id or version alias. Example: projects/{project}/locations/{location}/models/{model}@2 or projects/{project}/locations/{location}/models/{model}@golden If no version id or alias is specified, the "default" version will be returned. The "default" version alias is created for the first version of the model, and can be moved to other versions later on. There will be exactly one default version.

JSON representation 
{
  "persistentResourceId": string,
  "workerPoolSpecs": [
    {
      object (WorkerPoolSpec)
    }
  ],
  "scheduling": {
    object (Scheduling)
  },
  "serviceAccount": string,
  "network": string,
  "reservedIpRanges": [
    string
  ],
  "pscInterfaceConfig": {
    object (PscInterfaceConfig)
  },
  "baseOutputDirectory": {
    object (GcsDestination)
  },
  "protectedArtifactLocationId": string,
  "tensorboard": string,
  "enableWebAccess": boolean,
  "enableDashboardAccess": boolean,
  "experiment": string,
  "experimentRun": string,
  "models": [
    string
  ]
}

WorkerPoolSpec

Represents the spec of a worker pool in a job.

Fields
machineSpec object (MachineSpec)

Optional. Immutable. The specification of a single machine.

replicaCount string (int64 format)

Optional. The number of worker replicas to use for this worker pool.

nfsMounts[] object (NfsMount)

Optional. List of NFS mount spec.

lustreMounts[] object (LustreMount)

Optional. List of Lustre mounts.

diskSpec object (DiskSpec)

Disk spec.

task Union type
The custom task to be executed in this worker pool. task can be only one of the following:
containerSpec object (ContainerSpec)

The custom container task.

pythonPackageSpec object (PythonPackageSpec)

The Python packaged task.

JSON representation 
{
  "machineSpec": {
    object (MachineSpec)
  },
  "replicaCount": string,
  "nfsMounts": [
    {
      object (NfsMount)
    }
  ],
  "lustreMounts": [
    {
      object (LustreMount)
    }
  ],
  "diskSpec": {
    object (DiskSpec)
  },

  // task
  "containerSpec": {
    object (ContainerSpec)
  },
  "pythonPackageSpec": {
    object (PythonPackageSpec)
  }
  // Union type
}

PythonPackageSpec

The spec of a Python packaged code.

Fields
executorImageUri string

Required. The URI of a container image in Artifact Registry that will run the provided Python package. Agent Platform provides a wide range of executor images with pre-installed packages to meet users' various use cases. See the list of pre-built containers for training. You must use an image from this list.

packageUris[] string

Required. The Google Cloud Storage location of the Python package files which are the training program and its dependent packages. The maximum number of package URIs is 100.

pythonModule string

Required. The Python module name to run after installing the packages.

args[] string

Command line arguments to be passed to the Python task.

env[] object (EnvVar)

Environment variables to be passed to the python module. Maximum limit is 100.

JSON representation 
{
  "executorImageUri": string,
  "packageUris": [
    string
  ],
  "pythonModule": string,
  "args": [
    string
  ],
  "env": [
    {
      object (EnvVar)
    }
  ]
}

MachineSpec

Specification of a single machine.

Fields
machineType string

Immutable. The type of the machine.

See the list of machine types supported for prediction

See the list of machine types supported for custom training.

For DeployedModel this field is optional, and the default value is n1-standard-2. For BatchPredictionJob or as part of WorkerPoolSpec this field is required.

acceleratorType enum (AcceleratorType)

Immutable. The type of accelerator(s) that may be attached to the machine as per acceleratorCount.

acceleratorCount integer

The number of accelerators to attach to the machine.

For accelerator optimized machine types (https://cloud.google.com/compute/docs/accelerator-optimized-machines), One may set the acceleratorCount from 1 to N for machine with N GPUs. If acceleratorCount is less than or equal to N / 2, Vertex will co-schedule the replicas of the model into the same VM to save cost.

For example, if the machine type is a3-highgpu-8g, which has 8 H100 GPUs, one can set acceleratorCount to 1 to 8. If acceleratorCount is 1, 2, 3, or 4, Vertex will co-schedule 8, 4, 2, or 2 replicas of the model into the same VM to save cost.

When co-scheduling, CPU, memory and storage on the VM will be distributed to replicas on the VM. For example, one can expect a co-scheduled replica requesting 2 GPUs out of a 8-GPU VM will receive 25% of the CPU, memory and storage of the VM.

Note that the feature is not compatible with [multihostGpuNodeCount][]. When multihostGpuNodeCount is set, the co-scheduling will not be enabled.

gpuPartitionSize string

Optional. Immutable. The Nvidia GPU partition size.

When specified, the requested accelerators will be partitioned into smaller GPU partitions. For example, if the request is for 8 units of NVIDIA A100 GPUs, and gpuPartitionSize="1g.10gb", the service will create 8 * 7 = 56 partitioned MIG instances.

The partition size must be a value supported by the requested accelerator. Refer to Nvidia GPU Partitioning for the available partition sizes.

If set, the acceleratorCount should be set to 1.

tpuTopology string

Immutable. The topology of the TPUs. Corresponds to the TPU topologies available from GKE. (Example: tpuTopology: "2x2x1").

reservationAffinity object (ReservationAffinity)

Optional. Immutable. Configuration controlling how this resource pool consumes reservation.

JSON representation 
{
  "machineType": string,
  "acceleratorType": enum (AcceleratorType),
  "acceleratorCount": integer,
  "gpuPartitionSize": string,
  "tpuTopology": string,
  "reservationAffinity": {
    object (ReservationAffinity)
  }
}

ReservationAffinity

A ReservationAffinity can be used to configure a Agent Platform resource (e.g., a DeployedModel) to draw its Compute Engine resources from a Shared Reservation, or exclusively from on-demand capacity.

Fields
reservationAffinityType enum (Type)

Required. Specifies the reservation affinity type.

key string

Optional. Corresponds to the label key of a reservation resource. To target a SPECIFIC_RESERVATION by name, use compute.googleapis.com/reservation-name as the key and specify the name of your reservation as its value.

values[] string

Optional. Corresponds to the label values of a reservation resource. This must be the full resource name of the reservation or reservation block.

JSON representation 
{
  "reservationAffinityType": enum (Type),
  "key": string,
  "values": [
    string
  ]
}

NfsMount

Represents a mount configuration for Network File System (NFS) to mount.

Fields
server string

Required. IP address of the NFS server.

path string

Required. Source path exported from NFS server. Has to start with '/', and combined with the ip address, it indicates the source mount path in the form of server:path

mountPoint string

Required. Destination mount path. The NFS will be mounted for the user under /mnt/nfs/

JSON representation 
{
  "server": string,
  "path": string,
  "mountPoint": string
}

LustreMount

Represents a mount configuration for Lustre file system.

Fields
instanceIp string

Required. IP address of the Lustre instance.

volumeHandle string

Required. The unique identifier of the Lustre volume.

filesystem string

Required. The name of the Lustre filesystem.

mountPoint string

Required. Destination mount path. The Lustre file system will be mounted for the user under /mnt/lustre/

JSON representation 
{
  "instanceIp": string,
  "volumeHandle": string,
  "filesystem": string,
  "mountPoint": string
}

DiskSpec

Represents the spec of disk options.

Fields
bootDiskType string

type of the boot disk. For non-A3U machines, the default value is "pd-ssd", for A3U machines, the default value is "hyperdisk-balanced". Valid values: "pd-ssd" (Persistent Disk Solid state Drive), "pd-standard" (Persistent Disk Hard Disk Drive) or "hyperdisk-balanced".

bootDiskSizeGb integer

Size in GB of the boot disk (default is 100GB).

JSON representation 
{
  "bootDiskType": string,
  "bootDiskSizeGb": integer
}

Scheduling

All parameters related to queuing and scheduling of custom jobs.

Fields
timeout string (Duration format)

Optional. The maximum job running time. The default is 7 days.

A duration in seconds with up to nine fractional digits, ending with 's'. Example: "3.5s".

restartJobOnWorkerRestart boolean

Optional. Restarts the entire CustomJob if a worker gets restarted. This feature can be used by distributed training jobs that are not resilient to workers leaving and joining a job.

strategy enum (Strategy)

Optional. This determines which type of scheduling strategy to use.

disableRetries boolean

Optional. Indicates if the job should retry for internal errors after the job starts running. If true, overrides Scheduling.restart_job_on_worker_restart to false.

maxWaitDuration string (Duration format)

Optional. This is the maximum duration that a job will wait for the requested resources to be provisioned if the scheduling strategy is set to [Strategy.DWS_FLEX_START]. If set to 0, the job will wait indefinitely. The default is 24 hours.

A duration in seconds with up to nine fractional digits, ending with 's'. Example: "3.5s".

JSON representation 
{
  "timeout": string,
  "restartJobOnWorkerRestart": boolean,
  "strategy": enum (Strategy),
  "disableRetries": boolean,
  "maxWaitDuration": string
}

PscInterfaceConfig

Configuration for PSC-I.

Fields
networkAttachment string

Optional. The name of the Compute Engine network attachment to attach to the resource within the region and user project. To specify this field, you must have already created a network attachment. This field is only used for resources using PSC-I.

dnsPeeringConfigs[] object (DnsPeeringConfig)

Optional. DNS peering configurations. When specified, Agent Platform will attempt to configure DNS peering zones in the tenant project VPC to resolve the specified domains using the target network's Cloud DNS. The user must grant the dns.peer role to the Agent Platform service Agent on the target project.

JSON representation 
{
  "networkAttachment": string,
  "dnsPeeringConfigs": [
    {
      object (DnsPeeringConfig)
    }
  ]
}

DnsPeeringConfig

DNS peering configuration. These configurations are used to create DNS peering zones in the Vertex tenant project VPC, enabling resolution of records within the specified domain hosted in the target network's Cloud DNS.

Fields
domain string

Required. The DNS name suffix of the zone being peered to, e.g., "my-internal-domain.corp.". Must end with a dot.

targetProject string

Required. The project id hosting the Cloud DNS managed zone that contains the 'domain'. The Agent Platform service Agent requires the dns.peer role on this project.

targetNetwork string

Required. The VPC network name in the targetProject where the DNS zone specified by 'domain' is visible.

JSON representation 
{
  "domain": string,
  "targetProject": string,
  "targetNetwork": string
}

GcsDestination

The Google Cloud Storage location where the output is to be written to.

Fields
outputUriPrefix string

Required. Google Cloud Storage URI to output directory. If the uri doesn't end with '/', a '/' will be automatically appended. The directory is created if it doesn't exist.

JSON representation 
{
  "outputUriPrefix": string
}