Skip to main content

GCP Ai Platform Batch Prediction Job

A Batch Prediction Job in Google Cloud’s AI Platform (now Vertex AI) is an offline inference task that applies an existing machine-learning model to a large, static dataset and writes the results to Cloud Storage or BigQuery. Unlike online prediction, it is optimised for high-throughput, cost-effective processing of many instances that do not require real-time responses. The job encapsulates configuration such as the source of the input instances, the destination for the predictions, machine types, acceleration, encryption settings and the service account under which the workload runs.
Official documentation: https://cloud.google.com/vertex-ai/docs/reference/rest/v1/projects.locations.batchPredictionJobs#BatchPredictionJob

Supported Methods​

  • GET: Get a gcp-ai-platform-batch-prediction-job by its "locations|batchPredictionJobs"
  • LIST
  • SEARCH: Search Batch Prediction Jobs within a location. Use the location name e.g., 'us-central1'

gcp-big-query-table​

A Batch Prediction Job can write its output directly to a BigQuery table when bigqueryDestination is specified. Overmind links the job to the table it populates, allowing you to see which datasets are affected by each prediction run.

gcp-iam-service-account​

The job executes under a service account (either the Vertex AI Service Agent or a user-supplied account). This link shows the principal that holds the permissions to read the model, access the input data and write the results.

gcp-cloud-kms-crypto-key​

If customer-managed encryption keys are used, the job references a KMS CryptoKey to encrypt output artefacts in BigQuery or Cloud Storage. The link highlights the key that secures the predictions.

gcp-ai-platform-model​

A Batch Prediction Job must point to an existing AI Platform / Vertex AI Model resource, specifying which version to apply to the input data. Overmind links the job to this model so you can trace how a model is being used in production workflows.

gcp-storage-bucket​

Input instances are usually read from Cloud Storage URIs, and output files can be written back to a bucket when gcsDestination is chosen. The link indicates the buckets that supply data to or receive results from the job.