Create training pipelines

Training pipelines let you perform custom machine learning (ML) training and automatically create a Model resource based on your training output.

Before you create a pipeline

Before you create a training pipeline on Vertex AI, you need to create a Python training application or a custom container to define the training code and dependencies you want to run on Vertex AI. If you create a Python training application using TensorFlow, scikit-learn, or XGBoost, you can use our pre-built containers to run your code. If you're not sure which of these options to choose, refer to the training code requirements to learn more.

Training pipeline options

A training pipeline encapsulates training jobs with additional steps. This guide explains two different training pipelines:

  • Launch a custom job and upload the resulting model to Vertex AI
  • Launch a hyperparameter tuning job and upload the resulting model to Vertex AI

Additionally, you can use Vertex AI datasets (managed datasets) in your training pipeline. Learn more about configuring your training pipeline to use a managed dataset.

What a custom job includes

When you create a custom job, you specify settings that Vertex AI needs to run your training code, including:

Within the worker pool(s), you can specify the following settings:

If you want to create a standalone custom job outside of a Vertex AI training pipeline, refer to the guide on custom jobs.

Configure your pipeline to use a managed dataset

Within your training pipeline, you can configure your custom training job or hyperparameter tuning job to use a managed dataset. Managed datasets let you manage your datasets with your training applications and models.

To use a managed dataset in your training pipeline:

  1. Create your dataset.
  2. Update your training application to use a managed dataset. For more information, see how Vertex AI passes your dataset to your training application.
  3. Specify a managed dataset when you create your training pipeline. For example, if you create your training pipeline using the REST API, specify the dataset settings in the inputDataConfig section.

    You must create the training pipeline in the same region where you created the dataset.

To learn more, refer to the API reference on TrainingPipeline.

Configure distributed training

Within your training pipeline, you can configure your custom training job or hyperparameter tuning job for distributed training by specifying multiple worker pools.

All the examples on this page show single-replica training jobs with one worker pool. To modify them for distributed training:

  • Use your first worker pool to configure your primary replica, and set the replica count to 1.
  • Add more worker pools to configure worker replicas, parameter server replicas, or evaluator replicas, if your machine learning framework supports these additional cluster tasks for distributed training.

Learn more about using distributed training.

Custom job and model upload

This training pipeline encapsulates a custom job with an added convenience step that makes it easier to deploy your model to Vertex AI after training. This training pipeline does two main things:

  1. The training pipeline creates a custom job resource. The custom job runs the training application using the computing resources that you specify.

  2. After the custom job completes, the training pipeline finds the model artifacts that your training application creates in the output directory you specified for your Cloud Storage bucket. It uses these artifacts to create a model resource, which sets you up for model deployment.

    The total file size of the model artifacts must be 10 GB or less.

There are two different ways to set the location for your model artifacts:

If you specify both baseOutputDirectory and modelToUpload.artifactUri, Vertex AI uses modelToUpload.artifactUri.

To create this type of training pipeline:

Console

  1. In the Cloud Console, in the Vertex AI section, go to the Training pipelines page.

    Go to Training pipelines

  2. Click Create to open the Train new model pane.

  3. On the Choose training method step, specify the following settings:

    1. If you want to use a managed dataset for training, then specify a Dataset and an Annotation set.

      Otherwise, in the Dataset drop-down list, select No managed dataset.

    2. Select Custom training (advanced).

    Click Continue.

  4. On the Define your model step, enter a name of your choice, MODEL_NAME, for your model. Click Continue.

  5. On the Training container step, specify the following settings:

    1. Select whether to use a Pre-built container or a Custom container for training.

    2. Depending on your choice, do one of the following:

    3. In the Model output directory field, specify the Cloud Storage URI of a directory in a bucket that you have access to. The directory does not need to exist yet.

      This value gets passed to Vertex AI in the baseOutputDirectory API field, which sets several environment variables that your training application can access when it runs.

      At the end of training, Vertex AI looks for model artifacts in a subdirectory of this URI in order to create a Model. (This subdirectory is available to your training code as the AIP_MODEL_DIR environment variable.)

      When you do not use hyperparameter tuning, Vertex AI expects to find model artifacts in BASE_OUTPUT_DIRECTORY/model/.

    4. In the Arguments field, you may optionally specify arguments for Vertex AI to use when it starts running your training code. The behavior of these arguments differs depending on what type of container you are using:

    Click Continue.

  6. On the Hyperparameter tuning step, make sure that the Enable hyperparameter tuning checkbox is not selected. Click Continue.

  7. On the Compute and pricing step, specify the following settings:

    1. In the Region drop-down list, select a region that supports custom training.

    2. In the Worker pool 0 section, specify compute resources to use for training.

      If you specify accelerators, make sure the type of accelerator that you choose is available in your selected region.

      If you want to perform distributed training, then click Add more worker pools and specify an additional set of compute resources for each additional worker pool that you want.

    Click Continue.

  8. On the Prediction container step, specify the following settings:

    1. Select whether to use a Pre-built container or a Custom container to serve predictions from your trained model.

    2. Depending on your choice, do one of the following:

    3. The Model directory field contains the value that you previously set in the Model output directory field of the Training container step. Changing either of these fields has the same effect. See the previous instruction for more information about this field.

    4. Leave the fields in the Predict schemata section blank.

  9. Click Start training to start the custom training pipeline.

REST & CMD LINE

Use the following code sample to create a training pipeline using the create method of the trainingPipeline resource.

Before using any of the request data, make the following replacements:

  • LOCATION_ID: The region where the training code is run and the Model is stored.
  • PROJECT_ID: Your project ID.
  • TRAINING_PIPELINE_NAME: Required. A display name for the trainingPipeline.
  • If your training application uses a Vertex AI dataset, specify the following:
    • DATASET_ID: The ID of the dataset.
    • ANNOTATIONS_FILTER: Filters the dataset by the annotations that you specify.
    • ANNOTATION_SCHEMA_URI: Filters the dataset by the specified annotation schema URI.
    • Use one of the following options to specify how data items are split into training, validation, and test sets.
      • To split the dataset based on fractions defining the size of each set, specify the following:
        • TRAINING_FRACTION: The fraction of the dataset to use to train your model.
        • VALIDATION_FRACTION: The fraction of the dataset to use to validate your model.
        • TEST_FRACTION: The fraction of the dataset to use to evaluate your model.
      • To split the dataset based on filters, specify the following:
        • TRAINING_FILTER: Filters the dataset to data items to use for training your model.
        • VALIDATION_FILTER: Filters the dataset to data items to use for validating your model.
        • TEST_FILTER: Filters the dataset to data items to use for evaluating your model.
      • To use a predefined split, specify the following:
        • PREDEFINED_SPLIT_KEY: The name of the column to use to split the dataset. Acceptable values in this column include `training`, `validation`, and `test`.
      • To split the dataset based on the timestamp on the dataitems, specify the following:
        • TIMESTAMP_TRAINING_FRACTION: The fraction of the dataset to use to train your model.
        • TIMESTAMP_VALIDATION_FRACTION: The fraction of the dataset to use to validate your model.
        • TIMESTAMP_TEST_FRACTION: The fraction of the dataset to use to evaluate your model.
        • TIMESTAMP_SPLIT_KEY: The name of the timestamp column to use to split the dataset.
    • OUTPUT_URI_PREFIX: The Cloud Storage location where Vertex AI exports your training dataset, once it has been split into training, validation, and test sets.
  • Define the custom training job:
    • MACHINE_TYPE: The type of the machine. Refer to available machine types for training.
    • ACCELERATOR_TYPE: (Optional.) The type of accelerator to attach to each trial.
    • ACCELERATOR_COUNT: (Optional.) The number of accelerators to attach to each trial.
    • REPLICA_COUNT: The number of worker replicas to use for each trial.
    • If your training application runs in a custom container, specify the following:
      • CUSTOM_CONTAINER_IMAGE_URI: The URI of a container image in Artifact Registry, Container Registry, or Docker Hub that is to be run on each worker replica.
      • CUSTOM_CONTAINER_COMMAND: (Optional.) The command to be invoked when the container is started. This command overrides the container's default entrypoint.
      • CUSTOM_CONTAINER_ARGS: (Optional.) The arguments to be passed when starting the container.
    • If your training application is a Python package that runs in a pre-built container, specify the following:
      • PYTHON_PACKAGE_EXECUTOR_IMAGE_URI: The URI of the container image that runs the provided Python package. Refer to the available pre-built containers for training.
      • PYTHON_PACKAGE_URIS: The 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.
      • PYTHON_MODULE: The Python module name to run after installing the packages.
      • PYTHON_PACKAGE_ARGS: (Optional.) Command-line arguments to be passed to the Python module.
    • TIMEOUT: (Optional.) The maximum running time for the job.
  • MODEL_NAME: A display name for the model uploaded (created) by the TrainingPipeline.
  • MODEL_DESCRIPTION: A description for the model.
  • IMAGE_URI: The URI of the container image to use for running predictions. For example, us-docker.pkg.dev/vertex-ai/prediction/tf2-cpu.2-1:latest. Use pre-built containers or custom containers.
  • modelToUpload.labels: Any set of key-value pairs to organize your models. For example:
    • "env": "prod"
    • "tier": "backend"
  • Specify the LABEL_NAME and LABEL_VALUE for any labels that you want to apply to this training pipeline.

HTTP method and URL:

POST https://LOCATION_ID-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION_ID/trainingPipelines

Request JSON body:

{
  "displayName": "TRAINING_PIPELINE_NAME",
  "inputDataConfig": {
    "datasetId": DATASET_ID,
    "annotationsFilter": ANNOTATIONS_FILTER,
    "annotationSchemaUri": ANNOTATION_SCHEMA_URI,

    // Union field split can be only one of the following:
    "fractionSplit": {
      "trainingFraction": TRAINING_FRACTION,
      "validationFraction": VALIDATION_FRACTION,
      "testFraction": TEST_FRACTION
    },
    "filterSplit": {
      "trainingFilter": TRAINING_FILTER,
      "validationFilter": VALIDATION_FILTER,
      "testFilter": TEST_FILTER
    },
    "predefinedSplit": {
      "key": PREDEFINED_SPLIT_KEY
    },
    "timestampSplit": {
      "trainingFraction": TIMESTAMP_TRAINING_FRACTION,
      "validationFraction": TIMESTAMP_VALIDATION_FRACTION,
      "testFraction": TIMESTAMP_TEST_FRACTION,
      "key": TIMESTAMP_SPLIT_KEY
    }
    // End of list of possible types for union field split.
    "gcsDestination": {
      "outputUriPrefix": OUTPUT_URI_PREFIX
    }
  },
  "trainingTaskDefinition": "gs://google-cloud-aiplatform/schema/trainingjob/definition/custom_task_1.0.0.yaml",
  "trainingTaskInputs": {
    "workerPoolSpecs": [
        {
          "machineSpec": {
            "machineType": MACHINE_TYPE,
            "acceleratorType": ACCELERATOR_TYPE,
            "acceleratorCount": ACCELERATOR_COUNT
          },
          "replicaCount": REPLICA_COUNT,

          // Union field task can be only one of the following:
          "containerSpec": {
            "imageUri": CUSTOM_CONTAINER_IMAGE_URI,
            "command": [
              CUSTOM_CONTAINER_COMMAND
            ],
            "args": [
              CUSTOM_CONTAINER_ARGS
            ]
          },
          "pythonPackageSpec": {
            "executorImageUri": PYTHON_PACKAGE_EXECUTOR_IMAGE_URI,
            "packageUris": [
              PYTHON_PACKAGE_URIS
            ],
            "pythonModule": PYTHON_MODULE,
            "args": [
              PYTHON_PACKAGE_ARGS
            ]
          }
          // End of list of possible types for union field task.
        }
      ],
      "scheduling": {
        "TIMEOUT": TIMEOUT
      }
    }
  },
  "modelToUpload": {
    "displayName": "MODEL_NAME",
    "predictSchemata": {},
    "containerSpec": {
      "imageUri": "IMAGE_URI"
    }
  },
  "labels": {
    LABEL_NAME_1": LABEL_VALUE_1,
    LABEL_NAME_2": LABEL_VALUE_2
  }
}

To send your request, choose one of these options:

curl

Save the request body in a file called request.json, and execute the following command:

curl -X POST \
-H "Authorization: Bearer "$(gcloud auth application-default print-access-token) \
-H "Content-Type: application/json; charset=utf-8" \
-d @request.json \
"https://LOCATION_ID-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION_ID/trainingPipelines"

PowerShell

Save the request body in a file called request.json, and execute the following command:

$cred = gcloud auth application-default print-access-token
$headers = @{ "Authorization" = "Bearer $cred" }

Invoke-WebRequest `
-Method POST `
-Headers $headers `
-ContentType: "application/json; charset=utf-8" `
-InFile request.json `
-Uri "https://LOCATION_ID-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION_ID/trainingPipelines" | Select-Object -Expand Content

The response contains information about specifications as well as the TRAININGPIPELINE_ID.

Java

import com.google.cloud.aiplatform.v1.LocationName;
import com.google.cloud.aiplatform.v1.Model;
import com.google.cloud.aiplatform.v1.ModelContainerSpec;
import com.google.cloud.aiplatform.v1.PipelineServiceClient;
import com.google.cloud.aiplatform.v1.PipelineServiceSettings;
import com.google.cloud.aiplatform.v1.TrainingPipeline;
import com.google.gson.JsonArray;
import com.google.gson.JsonObject;
import com.google.protobuf.Value;
import com.google.protobuf.util.JsonFormat;
import java.io.IOException;

public class CreateTrainingPipelineCustomJobSample {

  public static void main(String[] args) throws IOException {
    // TODO(developer): Replace these variables before running the sample.
    String project = "PROJECT";
    String displayName = "DISPLAY_NAME";
    String modelDisplayName = "MODEL_DISPLAY_NAME";
    String containerImageUri = "CONTAINER_IMAGE_URI";
    String baseOutputDirectoryPrefix = "BASE_OUTPUT_DIRECTORY_PREFIX";
    createTrainingPipelineCustomJobSample(
        project, displayName, modelDisplayName, containerImageUri, baseOutputDirectoryPrefix);
  }

  static void createTrainingPipelineCustomJobSample(
      String project,
      String displayName,
      String modelDisplayName,
      String containerImageUri,
      String baseOutputDirectoryPrefix)
      throws IOException {
    PipelineServiceSettings settings =
        PipelineServiceSettings.newBuilder()
            .setEndpoint("us-central1-aiplatform.googleapis.com:443")
            .build();
    String location = "us-central1";

    // Initialize client that will be used to send requests. This client only needs to be created
    // once, and can be reused for multiple requests. After completing all of your requests, call
    // the "close" method on the client to safely clean up any remaining background resources.
    try (PipelineServiceClient client = PipelineServiceClient.create(settings)) {
      JsonObject jsonMachineSpec = new JsonObject();
      jsonMachineSpec.addProperty("machineType", "n1-standard-4");

      JsonArray jsonArgs = new JsonArray();
      jsonArgs.add("--model_dir=$(AIP_MODEL_DIR)");

      // A working docker image can be found at
      // gs://cloud-samples-data/ai-platform/mnist_tfrecord/custom_job
      JsonObject jsonContainerSpec = new JsonObject();
      jsonContainerSpec.addProperty("imageUri", containerImageUri);
      jsonContainerSpec.add("args", jsonArgs);

      JsonObject jsonJsonWorkerPoolSpec0 = new JsonObject();
      jsonJsonWorkerPoolSpec0.addProperty("replicaCount", 1);
      jsonJsonWorkerPoolSpec0.add("machineSpec", jsonMachineSpec);
      jsonJsonWorkerPoolSpec0.add("containerSpec", jsonContainerSpec);

      JsonArray jsonWorkerPoolSpecs = new JsonArray();
      jsonWorkerPoolSpecs.add(jsonJsonWorkerPoolSpec0);

      JsonObject jsonBaseOutputDirectory = new JsonObject();
      // The GCS location for outputs must be accessible by the project's AI Platform
      // service account.
      jsonBaseOutputDirectory.addProperty("output_uri_prefix", baseOutputDirectoryPrefix);

      JsonObject jsonTrainingTaskInputs = new JsonObject();
      jsonTrainingTaskInputs.add("workerPoolSpecs", jsonWorkerPoolSpecs);
      jsonTrainingTaskInputs.add("baseOutputDirectory", jsonBaseOutputDirectory);

      Value.Builder trainingTaskInputsBuilder = Value.newBuilder();
      JsonFormat.parser().merge(jsonTrainingTaskInputs.toString(), trainingTaskInputsBuilder);
      Value trainingTaskInputs = trainingTaskInputsBuilder.build();
      String trainingTaskDefinition =
          "gs://google-cloud-aiplatform/schema/trainingjob/definition/custom_task_1.0.0.yaml";
      String imageUri = "gcr.io/cloud-aiplatform/prediction/tf-cpu.1-15:latest";
      ModelContainerSpec containerSpec =
          ModelContainerSpec.newBuilder().setImageUri(imageUri).build();
      Model modelToUpload =
          Model.newBuilder()
              .setDisplayName(modelDisplayName)
              .setContainerSpec(containerSpec)
              .build();
      TrainingPipeline trainingPipeline =
          TrainingPipeline.newBuilder()
              .setDisplayName(displayName)
              .setTrainingTaskDefinition(trainingTaskDefinition)
              .setTrainingTaskInputs(trainingTaskInputs)
              .setModelToUpload(modelToUpload)
              .build();
      LocationName parent = LocationName.of(project, location);
      TrainingPipeline response = client.createTrainingPipeline(parent, trainingPipeline);
      System.out.format("response: %s\n", response);
      System.out.format("Name: %s\n", response.getName());
    }
  }
}

Python

The following examples show how to use the Vertex SDK for Python to create a custom training pipeline. Choose whether you plan to use a custom container or a pre-built container for training:

Pre-built container

When you use the Vertex SDK for Python to create a training pipeline that runs your Python code in a pre-built container, you can provide your training code in one of the following ways:

  • Specify the URI of a Python source distribution package in Cloud Storage.

    (This option is also available when you create a training pipeline without using the Vertex SDK for Python.)

  • Specify the path to a Python script on your local machine. Before it creates a training pipeline, the Vertex SDK for Python packages your script as a source distribution and uploads it to the Cloud Storage bucket of your choice.

    (This option is only available when you use the Vertex SDK for Python.)

To see a code sample for each of these options, select the corresponding tab:

Package

The following sample uses the CustomPythonPackageTrainingJob class.

def create_training_pipeline_custom_package_job_sample(
    project: str,
    location: str,
    staging_bucket: str,
    display_name: str,
    python_package_gcs_uri: str,
    python_module_name: str,
    container_uri: str,
    model_serving_container_image_uri: str,
    dataset_id: Optional[str] = None,
    model_display_name: Optional[str] = None,
    args: Optional[List[Union[str, float, int]]] = None,
    replica_count: int = 1,
    machine_type: str = "n1-standard-4",
    accelerator_type: str = "ACCELERATOR_TYPE_UNSPECIFIED",
    accelerator_count: int = 0,
    training_fraction_split: float = 0.8,
    validation_fraction_split: float = 0.1,
    test_fraction_split: float = 0.1,
    sync: bool = True,
):
    aiplatform.init(project=project, location=location, staging_bucket=staging_bucket)

    job = aiplatform.CustomPythonPackageTrainingJob(
        display_name=display_name,
        python_package_gcs_uri=python_package_gcs_uri,
        python_module_name=python_module_name,
        container_uri=container_uri,
        model_serving_container_image_uri=model_serving_container_image_uri,
    )

    # This example uses an ImageDataset, but you can use another type
    dataset = aiplatform.ImageDataset(dataset_id) if dataset_id else None

    model = job.run(
        dataset=dataset,
        model_display_name=model_display_name,
        args=args,
        replica_count=replica_count,
        machine_type=machine_type,
        accelerator_type=accelerator_type,
        accelerator_count=accelerator_count,
        training_fraction_split=training_fraction_split,
        validation_fraction_split=validation_fraction_split,
        test_fraction_split=test_fraction_split,
        sync=sync,
    )

    model.wait()

    print(model.display_name)
    print(model.resource_name)
    print(model.uri)
    return model

Script

The following sample uses the CustomTrainingJob class.

def create_training_pipeline_custom_job_sample(
    project: str,
    location: str,
    staging_bucket: str,
    display_name: str,
    script_path: str,
    container_uri: str,
    model_serving_container_image_uri: str,
    dataset_id: Optional[str] = None,
    model_display_name: Optional[str] = None,
    args: Optional[List[Union[str, float, int]]] = None,
    replica_count: int = 0,
    machine_type: str = "n1-standard-4",
    accelerator_type: str = "ACCELERATOR_TYPE_UNSPECIFIED",
    accelerator_count: int = 0,
    training_fraction_split: float = 0.8,
    validation_fraction_split: float = 0.1,
    test_fraction_split: float = 0.1,
    sync: bool = True,
):
    aiplatform.init(project=project, location=location, staging_bucket=staging_bucket)

    job = aiplatform.CustomTrainingJob(
        display_name=display_name,
        script_path=script_path,
        container_uri=container_uri,
        model_serving_container_image_uri=model_serving_container_image_uri,
    )

    # This example uses an ImageDataset, but you can use another type
    dataset = aiplatform.ImageDataset(dataset_id) if dataset_id else None

    model = job.run(
        dataset=dataset,
        model_display_name=model_display_name,
        args=args,
        replica_count=replica_count,
        machine_type=machine_type,
        accelerator_type=accelerator_type,
        accelerator_count=accelerator_count,
        training_fraction_split=training_fraction_split,
        validation_fraction_split=validation_fraction_split,
        test_fraction_split=test_fraction_split,
        sync=sync,
    )

    model.wait()

    print(model.display_name)
    print(model.resource_name)
    print(model.uri)
    return model

Custom container

The following sample uses the CustomContainerTrainingJob class.

def create_training_pipeline_custom_container_job_sample(
    project: str,
    location: str,
    staging_bucket: str,
    display_name: str,
    container_uri: str,
    model_serving_container_image_uri: str,
    dataset_id: Optional[str] = None,
    model_display_name: Optional[str] = None,
    args: Optional[List[Union[str, float, int]]] = None,
    replica_count: int = 1,
    machine_type: str = "n1-standard-4",
    accelerator_type: str = "ACCELERATOR_TYPE_UNSPECIFIED",
    accelerator_count: int = 0,
    training_fraction_split: float = 0.8,
    validation_fraction_split: float = 0.1,
    test_fraction_split: float = 0.1,
    sync: bool = True,
):
    aiplatform.init(project=project, location=location, staging_bucket=staging_bucket)

    job = aiplatform.CustomContainerTrainingJob(
        display_name=display_name,
        container_uri=container_uri,
        model_serving_container_image_uri=model_serving_container_image_uri,
    )

    # This example uses an ImageDataset, but you can use another type
    dataset = aiplatform.ImageDataset(dataset_id) if dataset_id else None

    model = job.run(
        dataset=dataset,
        model_display_name=model_display_name,
        args=args,
        replica_count=replica_count,
        machine_type=machine_type,
        accelerator_type=accelerator_type,
        accelerator_count=accelerator_count,
        training_fraction_split=training_fraction_split,
        validation_fraction_split=validation_fraction_split,
        test_fraction_split=test_fraction_split,
        sync=sync,
    )

    model.wait()

    print(model.display_name)
    print(model.resource_name)
    print(model.uri)
    return model

Hyperparameter tuning job and model upload

This training pipeline encapsulates a hyperparameter tuning job with an added convenience step that makes it easier to deploy your model to Vertex AI after training. This training pipeline does two main things:

  1. The training pipeline creates a hyperparameter tuning job resource. The hyperparameter tuning job creates multiple trials. For each trial, a custom job runs your training application using the computing resources and hyperparameters that you specify.

  2. After the hyperparameter tuning job completes, the training pipeline finds the model artifacts from the best trial, within the output directory (baseOutputDirectory) you specified for your Cloud Storage bucket. The training pipeline uses these artifacts to create a model resource, which sets you up for model deployment.

For this training pipeline, you must specify a baseOutputDirectory where Vertex AI searches for the model artifacts from the best trial.

Hyperparameter tuning jobs have additional settings to configure. Learn more about the settings for a HyperparameterTuningJob.

REST & CMD LINE

Use the following code sample to create a training pipeline using the create method of the trainingPipeline resource.

Before using any of the request data, make the following replacements:

  • LOCATION_ID: Your project's region.
  • PROJECT_ID: Your project ID.
  • TRAINING_PIPELINE_NAME: Required. A display name for the trainingPipeline.
  • If your training application uses a Vertex AI dataset, specify the following:
    • DATASET_ID: The ID of the dataset.
    • ANNOTATIONS_FILTER: Filters the dataset by the annotations that you specify.
    • ANNOTATION_SCHEMA_URI: Filters the dataset by the specified annotation schema URI.
    • Use one of the following options to specify how data items are split into training, validation, and test sets.
      • To split the dataset based on fractions defining the size of each set, specify the following:
        • TRAINING_FRACTION: The fraction of the dataset to use to train your model.
        • VALIDATION_FRACTION: The fraction of the dataset to use to validate your model.
        • TEST_FRACTION: The fraction of the dataset to use to evaluate your model.
      • To split the dataset based on filters, specify the following:
        • TRAINING_FILTER: Filters the dataset to data items to use for training your model.
        • VALIDATION_FILTER: Filters the dataset to data items to use for validating your model.
        • TEST_FILTER: Filters the dataset to data items to use for evaluating your model.
      • To use a predefined split, specify the following:
        • PREDEFINED_SPLIT_KEY: The name of the column to use to split the dataset. Acceptable values in this column include `training`, `validation`, and `test`.
      • To split the dataset based on the timestamp on the dataitems, specify the following:
        • TIMESTAMP_TRAINING_FRACTION: The fraction of the dataset to use to train your model.
        • TIMESTAMP_VALIDATION_FRACTION: The fraction of the dataset to use to validate your model.
        • TIMESTAMP_TEST_FRACTION: The fraction of the dataset to use to evaluate your model.
        • TIMESTAMP_SPLIT_KEY: The name of the timestamp column to use to split the dataset.
    • OUTPUT_URI_PREFIX: The Cloud Storage location where Vertex AI exports your training dataset, once it has been split into training, validation, and test sets.
  • Specify your hyperparameter tuning job:
    • Specify your metrics:
      • METRIC_ID: The name of this metric.
      • METRIC_GOAL: The goal of this metric. Can be MAXIMIZE or MINIMIZE.
    • Specify your hyperparameters:
      • PARAMETER_ID: The name of this hyperparameter.
      • PARAMETER_SCALE: (Optional.) How the parameter should be scaled. Leave unset for CATEGORICAL parameters. Can be UNIT_LINEAR_SCALE, UNIT_LOG_SCALE, UNIT_REVERSE_LOG_SCALE, or SCALE_TYPE_UNSPECIFIED
      • If this hyperparameter's type is DOUBLE, specify the minimum (DOUBLE_MIN_VALUE) and maximum (DOUBLE_MAX_VALUE) values for this hyperparameter.
      • If this hyperparameter's type is INTEGER, specify the minimum (INTEGER_MIN_VALUE) and maximum (INTEGER_MAX_VALUE) values for this hyperparameter.
      • If this hyperparameter's type is CATEGORICAL, specify the acceptable values (CATEGORICAL_VALUES) as an array of strings.
      • If this hyperparameter's type is DISCRETE, specify the acceptable values (DISCRETE_VALUES) as an array of numbers.
    • ALGORITHM: (Optional.) The search algorithm to use in this hyperparameter tuning job. Can be ALGORITHM_UNSPECIFIED, GRID_SEARCH, or RANDOM_SEARCH.
    • MAX_TRIAL_COUNT: The maximum number of trials to run in this job.
    • PARALLEL_TRIAL_COUNT: The maximum number of trials that can run in parallel.
    • MAX_FAILED_TRIAL_COUNT: The number of jobs that can fail before the hyperparameter tuning job fails.
    • Define the trial custom training job:
      • MACHINE_TYPE: The type of the machine. Refer to the available machine types for training.
      • ACCELERATOR_TYPE: (Optional.) The type of accelerator to attach to each trial.
      • ACCELERATOR_COUNT: (Optional.) The number of accelerators to attach to each trial.
      • REPLICA_COUNT: The number of worker replicas to use for each trial.
      • If your training application runs in a custom container, specify the following:
        • CUSTOM_CONTAINER_IMAGE_URI: The URI of a container image in Artifact Registry, Container Registry, or Docker Hub that is to be run on each worker replica.
        • CUSTOM_CONTAINER_COMMAND: (Optional.) The command to be invoked when the container is started. This command overrides the container's default entrypoint.
        • CUSTOM_CONTAINER_ARGS: (Optional.) The arguments to be passed when starting the container.
      • If your training application is a Python package that runs in a pre-built container, specify the following:
        • PYTHON_PACKAGE_EXECUTOR_IMAGE_URI: The URI of the container image that runs the provided Python package. Refer to the available pre-built containers for training.
        • PYTHON_PACKAGE_URIS: The 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.
        • PYTHON_MODULE: The Python module name to run after installing the packages.
        • PYTHON_PACKAGE_ARGS: (Optional.) Command-line arguments to be passed to the Python module.
    • Learn about job scheduling options.
    • TIMEOUT: (Optional.) The maximum running time for each trial.
    • Specify the LABEL_NAME and LABEL_VALUE for any labels that you want to apply to this hyperparameter tuning job.
  • MODEL_NAME: A display name for the model uploaded (created) by the TrainingPipeline.
  • MODEL_DESCRIPTION: Optional. A description for the model.
  • PREDICTION_IMAGE_URI: Required. Specify one of the two following options:
  • modelToUpload.labels: Optional. Any set of key-value pairs to organize your models. For example:
    • "env": "prod"
    • "tier": "backend"
  • Specify the LABEL_NAME and LABEL_VALUE for any labels that you want to apply to this training pipeline.

HTTP method and URL:

POST https://LOCATION_ID-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION_ID/trainingPipelines

Request JSON body:

{
  "displayName": "TRAINING_PIPELINE_NAME",
  "inputDataConfig": {
    "datasetId": DATASET_ID,
    "annotationsFilter": ANNOTATIONS_FILTER,
    "annotationSchemaUri": ANNOTATION_SCHEMA_URI,

    // Union field split can be only one of the following:
    "fractionSplit": {
      "trainingFraction": TRAINING_FRACTION,
      "validationFraction": VALIDATION_FRACTION,
      "testFraction": TEST_FRACTION
    },
    "filterSplit": {
      "trainingFilter": TRAINING_FILTER,
      "validationFilter": VALIDATION_FILTER,
      "testFilter": TEST_FILTER
    },
    "predefinedSplit": {
      "key": PREDEFINED_SPLIT_KEY
    },
    "timestampSplit": {
      "trainingFraction": TIMESTAMP_TRAINING_FRACTION,
      "validationFraction": TIMESTAMP_VALIDATION_FRACTION,
      "testFraction": TIMESTAMP_TEST_FRACTION,
      "key": TIMESTAMP_SPLIT_KEY
    }
    // End of list of possible types for union field split.
    "gcsDestination": {
      "outputUriPrefix": OUTPUT_URI_PREFIX
    }
  },
  "trainingTaskDefinition": "gs://google-cloud-aiplatform/schema/trainingjob/definition/hyperparameter_tuning_task_1.0.0.yaml",
  "trainingTaskInputs": {
    "studySpec": {
    "metrics": [
      {
        "metricId": METRIC_ID,
        "goal": METRIC_GOAL
      }
    ],
    "parameters": [
      {
        "parameterId": PARAMETER_ID,
        "scaleType": PARAMETER_SCALE,

        // Union field parameter_value_spec can be only one of the following:
        "doubleValueSpec": {
            "minValue": DOUBLE_MIN_VALUE,
            "maxValue": DOUBLE_MAX_VALUE
        },
        "integerValueSpec": {
            "minValue": INTEGER_MIN_VALUE,
            "maxValue": INTEGER_MAX_VALUE
        },
        "categoricalValueSpec": {
            "values": [
              CATEGORICAL_VALUES
            ]
        },
        "discreteValueSpec": {
            "values": [
              DISCRETE_VALUES
            ]
        }
        // End of list of possible types for union field parameter_value_spec.
      }
    ],
    "ALGORITHM": ALGORITHM
  },
  "maxTrialCount": MAX_TRIAL_COUNT,
  "parallelTrialCount": PARALLEL_TRIAL_COUNT,
  "maxFailedTrialCount": MAX_FAILED_TRIAL_COUNT,
  "trialJobSpec": {
      "workerPoolSpecs": [
        {
          "machineSpec": {
            "machineType": MACHINE_TYPE,
            "acceleratorType": ACCELERATOR_TYPE,
            "acceleratorCount": ACCELERATOR_COUNT
          },
          "replicaCount": REPLICA_COUNT,

          // Union field task can be only one of the following:
          "containerSpec": {
            "imageUri": CUSTOM_CONTAINER_IMAGE_URI,
            "command": [
              CUSTOM_CONTAINER_COMMAND
            ],
            "args": [
              CUSTOM_CONTAINER_ARGS
            ]
          },
          "pythonPackageSpec": {
            "executorImageUri": PYTHON_PACKAGE_EXECUTOR_IMAGE_URI,
            "packageUris": [
              PYTHON_PACKAGE_URIS
            ],
            "pythonModule": PYTHON_MODULE,
            "args": [
              PYTHON_PACKAGE_ARGS
            ]
          }
          // End of list of possible types for union field task.
        }
      ],
      "scheduling": {
        "TIMEOUT": TIMEOUT
      }
    },
    "labels": {
      LABEL_NAME_1": LABEL_VALUE_1,
      LABEL_NAME_2": LABEL_VALUE_2
    }
  },
  "modelToUpload": {
    "displayName": "MODEL_NAME",
    "description": "MODEL_DESCRIPTION",
    "predictSchemata": {},
    "containerSpec": {
      "imageUri": "PREDICTION_IMAGE_URI"
    }
  },
  "labels": {
    LABEL_NAME_1": LABEL_VALUE_1,
    LABEL_NAME_2": LABEL_VALUE_2
  }
}

To send your request, choose one of these options:

curl

Save the request body in a file called request.json, and execute the following command:

curl -X POST \
-H "Authorization: Bearer "$(gcloud auth application-default print-access-token) \
-H "Content-Type: application/json; charset=utf-8" \
-d @request.json \
"https://LOCATION_ID-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION_ID/trainingPipelines"

PowerShell

Save the request body in a file called request.json, and execute the following command:

$cred = gcloud auth application-default print-access-token
$headers = @{ "Authorization" = "Bearer $cred" }

Invoke-WebRequest `
-Method POST `
-Headers $headers `
-ContentType: "application/json; charset=utf-8" `
-InFile request.json `
-Uri "https://LOCATION_ID-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION_ID/trainingPipelines" | Select-Object -Expand Content

The response contains information about specifications as well as the TRAININGPIPELINE_ID.

Monitor training

To view training logs, do the following:

  1. In the Cloud Console, in the Vertex AI section, go to the Training page.

    Go to the Training page

  2. Click the name of your job to go to the custom job page.

  3. Click View logs.

You can also use an interactive shell to inspect your training containers while the training pipeline is running.

View your trained model

When the custom training pipeline completes, you can find the trained model in the Cloud Console, in the Vertex AI section, on the Models page.

Go to the Models page

What's next