Get batch predictions for Gemini

Batch predictions let you send a large number of multimodal prompts in a single batch request.

For more information about the batch workflow and how to format your input data, see Get batch predictions for Gemini.

Supported Models:

Model Version
Gemini 1.5 Flash gemini-1.5-flash-002
gemini-1.5-flash-001
Gemini 1.5 Pro gemini-1.5-pro-002
gemini-1.5-pro-001
Gemini 1.0 Pro gemini-1.0-pro-001
gemini-1.0-pro-002

Example syntax

Syntax to send a batch prediction API request. This example is specific to BigQuery storage.

curl

curl -X POST 

  -H "Authorization: Bearer $(gcloud auth print-access-token)" 

  -H "Content-Type: application/json" \



https://${LOCATION}-aiplatform.googleapis.com/v1/projects/${PROJECT_ID}/locations/${LOCATION}/batchPredictionJobs 

-d '{
    "displayName": "...",
    "model": "publishers/google/models/${MODEL_ID}",
    "inputConfig": {
      "instancesFormat":"bigquery",
      "bigquerySource":{
        "inputUri" : "..."
      }
    },
    "outputConfig": {
      "predictionsFormat":"bigquery",
      "bigqueryDestination":{
        "outputUri": "..."
        }
    }
}'

Parameters

See examples for implementation details.

Body request

Parameters

displayName

A name you choose for your job.

model

The model to use for batch prediction.

inputConfig

The data format. For Gemini batch prediction, Cloud Storage and BigQuery input sources are supported.

outputConfig

The output configuration which determines model output location. Cloud Storage and BigQuery output locations are supported.

inputConfig

Parameters

instancesFormat

The prompt input format. Use jsonl for Cloud Storage or bigquery for BigQuery.

gcsSource.uris

The input source URI. This is a Cloud Storage location of the JSONL file in the form gs://bucketname/path/to/file.jsonl.

bigquerySource.inputUri

The input source URI. This is a BigQuery table URI in the form bq://project_id.dataset.table. The region of the input BigQuery dataset must be the same as the Vertex AI batch prediction job.

outputConfig

Parameters

predictionsFormat

The output format of the prediction. Use bigquery.

gcsDestination.outputUriPrefix

The Cloud Storage bucket and directory location, in the form gs://mybucket/path/to/output.

bigqueryDestination.outputUri

The BigQuery URI of the target output table, in the form bq://project_id.dataset.table. If the table doesn't already exist, then it is created for you. The region of the output BigQuery dataset must be the same as the Vertex AI batch prediction job.

Examples

Request a batch response

Batch requests for multimodal models accept Cloud Storage storage and BigQuery storage sources. To learn more, see the following:

Depending on the number of input items that you submitted, a batch generation task can take some time to complete.

REST

To create a batch prediction job, use the projects.locations.batchPredictionJobs.create method.

Cloud Storage input

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

  • LOCATION: A region that supports Gemini models.
  • PROJECT_ID: Your project ID.
  • INPUT_URI: The Cloud Storage location of your JSONL batch prediction input such as gs://bucketname/path/to/file.jsonl.
  • OUTPUT_FORMAT: To output to a BigQuery table, specify bigquery. To output to a Cloud Storage bucket, specify jsonl.
  • DESTINATION: For BigQuery, specify bigqueryDestination. For Cloud Storage, specify gcsDestination.
  • OUTPUT_URI_FIELD_NAME: For BigQuery, specify outputUri. For Cloud Storage, specify outputUriPrefix.
  • OUTPUT_URI: For BigQuery, specify the table location such as bq://myproject.mydataset.output_result. The region of the output BigQuery dataset must be the same as the Vertex AI batch prediction job. For Cloud Storage, specify the bucket and directory location such as gs://mybucket/path/to/output.

Request JSON body: 

{
  "displayName": "my-cloud-storage-batch-prediction-job",
  "model": "publishers/google/models/gemini-1.5-flash-002",
  "inputConfig": {
    "instancesFormat": "jsonl",
    "gcsSource": {
      "uris" : "INPUT_URI"
    }
  },
  "outputConfig": {
    "predictionsFormat": "OUTPUT_FORMAT",
    "DESTINATION": {
      "OUTPUT_URI_FIELD_NAME": "OUTPUT_URI"
    }
  }
}

To send your request, choose one of these options:

curl

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

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

PowerShell

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

$cred = gcloud auth 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-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/batchPredictionJobs" | Select-Object -Expand Content

You should receive a JSON response similar to the following.

BigQuery input

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

  • LOCATION: A region that supports Gemini models.
  • PROJECT_ID: Your project ID.
  • INPUT_URI: The BigQuery table where your batch prediction input is located such as bq://myproject.mydataset.input_table. Multi-region datasets are not supported.
  • OUTPUT_FORMAT: To output to a BigQuery table, specify bigquery. To output to a Cloud Storage bucket, specify jsonl.
  • DESTINATION: For BigQuery, specify bigqueryDestination. For Cloud Storage, specify gcsDestination.
  • OUTPUT_URI_FIELD_NAME: For BigQuery, specify outputUri. For Cloud Storage, specify outputUriPrefix.
  • OUTPUT_URI: For BigQuery, specify the table location such as bq://myproject.mydataset.output_result. The region of the output BigQuery dataset must be the same as the Vertex AI batch prediction job. For Cloud Storage, specify the bucket and directory location such as gs://mybucket/path/to/output.

Request JSON body: 

{
  "displayName": "my-bigquery-batch-prediction-job",
  "model": "publishers/google/models/gemini-1.5-flash-002",
  "inputConfig": {
    "instancesFormat": "bigquery",
    "bigquerySource":{
      "inputUri" : "INPUT_URI"
    }
  },
  "outputConfig": {
    "predictionsFormat": "OUTPUT_FORMAT",
    "DESTINATION": {
      "OUTPUT_URI_FIELD_NAME": "OUTPUT_URI"
    }
  }
}

To send your request, choose one of these options:

curl

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

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

PowerShell

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

$cred = gcloud auth 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-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/batchPredictionJobs" | Select-Object -Expand Content

You should receive a JSON response similar to the following.

The response includes a unique identifier for the batch job. You can poll for the status of the batch job using the BATCH_JOB_ID until the job state is JOB_STATE_SUCCEEDED. For example:

curl -X GET \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
https://us-central1-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/us-central1/batchPredictionJobs/BATCH_JOB_ID

Python

To learn how to install or update the Vertex AI SDK for Python, see Install the Vertex AI SDK for Python. For more information, see the Python API reference documentation.

Cloud Storage input

import time
import vertexai

from vertexai.batch_prediction import BatchPredictionJob

# TODO(developer): Update and un-comment below line
# PROJECT_ID = "your-project-id"

# Initialize vertexai
vertexai.init(project=PROJECT_ID, location="us-central1")

input_uri = "gs://cloud-samples-data/batch/prompt_for_batch_gemini_predict.jsonl"

# Submit a batch prediction job with Gemini model
batch_prediction_job = BatchPredictionJob.submit(
    source_model="gemini-1.5-flash-002",
    input_dataset=input_uri,
    output_uri_prefix=output_uri,
)

# Check job status
print(f"Job resource name: {batch_prediction_job.resource_name}")
print(f"Model resource name with the job: {batch_prediction_job.model_name}")
print(f"Job state: {batch_prediction_job.state.name}")

# Refresh the job until complete
while not batch_prediction_job.has_ended:
    time.sleep(5)
    batch_prediction_job.refresh()

# Check if the job succeeds
if batch_prediction_job.has_succeeded:
    print("Job succeeded!")
else:
    print(f"Job failed: {batch_prediction_job.error}")

# Check the location of the output
print(f"Job output location: {batch_prediction_job.output_location}")

# Example response:
#  Job output location: gs://your-bucket/gen-ai-batch-prediction/prediction-model-year-month-day-hour:minute:second.12345

BigQuery input

import time
import vertexai

from vertexai.batch_prediction import BatchPredictionJob

# TODO(developer): Update and un-comment below line
# PROJECT_ID = "your-project-id"

# Initialize vertexai
vertexai.init(project=PROJECT_ID, location="us-central1")

input_uri = "bq://storage-samples.generative_ai.batch_requests_for_multimodal_input"

# Submit a batch prediction job with Gemini model
batch_prediction_job = BatchPredictionJob.submit(
    source_model="gemini-1.5-flash-002",
    input_dataset=input_uri,
    output_uri_prefix=output_uri,
)

# Check job status
print(f"Job resource name: {batch_prediction_job.resource_name}")
print(f"Model resource name with the job: {batch_prediction_job.model_name}")
print(f"Job state: {batch_prediction_job.state.name}")

# Refresh the job until complete
while not batch_prediction_job.has_ended:
    time.sleep(5)
    batch_prediction_job.refresh()

# Check if the job succeeds
if batch_prediction_job.has_succeeded:
    print("Job succeeded!")
else:
    print(f"Job failed: {batch_prediction_job.error}")

# Check the location of the output
print(f"Job output location: {batch_prediction_job.output_location}")

# Example response:
#  Job output location: bq://Project-ID/gen-ai-batch-prediction/predictions-model-year-month-day-hour:minute:second.12345

Node.js

Before trying this sample, follow the Node.js setup instructions in the Vertex AI quickstart using client libraries. For more information, see the Vertex AI Node.js API reference documentation.

To authenticate to Vertex AI, set up Application Default Credentials. For more information, see Set up authentication for a local development environment.

Cloud Storage input

// Import the aiplatform library
const aiplatformLib = require('@google-cloud/aiplatform');
const aiplatform = aiplatformLib.protos.google.cloud.aiplatform.v1;

/**
 * TODO(developer):  Uncomment/update these variables before running the sample.
 */
// projectId = 'YOUR_PROJECT_ID';
// URI of the output folder in Google Cloud Storage.
// E.g. "gs://[BUCKET]/[OUTPUT]"
// outputUri = 'gs://my-bucket';

// URI of the input file in Google Cloud Storage.
// E.g. "gs://[BUCKET]/[DATASET].jsonl"
// Or try:
// "gs://cloud-samples-data/generative-ai/batch/gemini_multimodal_batch_predict.jsonl"
// for a batch prediction that uses audio, video, and an image.
const inputUri =
  'gs://cloud-samples-data/generative-ai/batch/batch_requests_for_multimodal_input.jsonl';
const location = 'us-central1';
const parent = `projects/${projectId}/locations/${location}`;
const modelName = `${parent}/publishers/google/models/gemini-1.5-flash-002`;

// Specify the location of the api endpoint.
const clientOptions = {
  apiEndpoint: `${location}-aiplatform.googleapis.com`,
};

// Instantiate the client.
const jobServiceClient = new aiplatformLib.JobServiceClient(clientOptions);

// Create a Gemini batch prediction job using Google Cloud Storage input and output buckets.
async function create_batch_prediction_gemini_gcs() {
  const gcsSource = new aiplatform.GcsSource({
    uris: [inputUri],
  });

  const inputConfig = new aiplatform.BatchPredictionJob.InputConfig({
    gcsSource: gcsSource,
    instancesFormat: 'jsonl',
  });

  const gcsDestination = new aiplatform.GcsDestination({
    outputUriPrefix: outputUri,
  });

  const outputConfig = new aiplatform.BatchPredictionJob.OutputConfig({
    gcsDestination: gcsDestination,
    predictionsFormat: 'jsonl',
  });

  const batchPredictionJob = new aiplatform.BatchPredictionJob({
    displayName: 'Batch predict with Gemini - GCS',
    model: modelName,
    inputConfig: inputConfig,
    outputConfig: outputConfig,
  });

  const request = {
    parent: parent,
    batchPredictionJob,
  };

  // Create batch prediction job request
  const [response] = await jobServiceClient.createBatchPredictionJob(request);
  console.log('Response name: ', response.name);
  // Example response:
  // Response name: projects/<project>/locations/us-central1/batchPredictionJobs/<job-id>
}

await create_batch_prediction_gemini_gcs();

BigQuery input

// Import the aiplatform library
const aiplatformLib = require('@google-cloud/aiplatform');
const aiplatform = aiplatformLib.protos.google.cloud.aiplatform.v1;

/**
 * TODO(developer):  Uncomment/update these variables before running the sample.
 */
// projectId = 'YOUR_PROJECT_ID';
// URI of the output BigQuery table.
// E.g. "bq://[PROJECT].[DATASET].[TABLE]"
// outputUri = 'bq://projectid.dataset.table';

// URI of the multimodal input BigQuery table.
// E.g. "bq://[PROJECT].[DATASET].[TABLE]"
const inputUri =
  'bq://storage-samples.generative_ai.batch_requests_for_multimodal_input';
const location = 'us-central1';
const parent = `projects/${projectId}/locations/${location}`;
const modelName = `${parent}/publishers/google/models/gemini-1.5-flash-002`;

// Specify the location of the api endpoint.
const clientOptions = {
  apiEndpoint: `${location}-aiplatform.googleapis.com`,
};

// Instantiate the client.
const jobServiceClient = new aiplatformLib.JobServiceClient(clientOptions);

// Create a Gemini batch prediction job using BigQuery input and output datasets.
async function create_batch_prediction_gemini_bq() {
  const bqSource = new aiplatform.BigQuerySource({
    inputUri: inputUri,
  });

  const inputConfig = new aiplatform.BatchPredictionJob.InputConfig({
    bigquerySource: bqSource,
    instancesFormat: 'bigquery',
  });

  const bqDestination = new aiplatform.BigQueryDestination({
    outputUri: outputUri,
  });

  const outputConfig = new aiplatform.BatchPredictionJob.OutputConfig({
    bigqueryDestination: bqDestination,
    predictionsFormat: 'bigquery',
  });

  const batchPredictionJob = new aiplatform.BatchPredictionJob({
    displayName: 'Batch predict with Gemini - BigQuery',
    model: modelName, // Add model parameters per request in the input BigQuery table.
    inputConfig: inputConfig,
    outputConfig: outputConfig,
  });

  const request = {
    parent: parent,
    batchPredictionJob,
  };

  // Create batch prediction job request
  const [response] = await jobServiceClient.createBatchPredictionJob(request);
  console.log('Response name: ', response.name);
  // Example response:
  // Response name: projects/<project>/locations/us-central1/batchPredictionJobs/<job-id>
}

await create_batch_prediction_gemini_bq();

Java

Before trying this sample, follow the Java setup instructions in the Vertex AI quickstart using client libraries. For more information, see the Vertex AI Java API reference documentation.

To authenticate to Vertex AI, set up Application Default Credentials. For more information, see Set up authentication for a local development environment.

Cloud Storage input

import com.google.cloud.aiplatform.v1.BatchPredictionJob;
import com.google.cloud.aiplatform.v1.GcsDestination;
import com.google.cloud.aiplatform.v1.GcsSource;
import com.google.cloud.aiplatform.v1.JobServiceClient;
import com.google.cloud.aiplatform.v1.JobServiceSettings;
import com.google.cloud.aiplatform.v1.LocationName;
import java.io.IOException;

public class CreateBatchPredictionGeminiJobSample {

  public static void main(String[] args) throws IOException {
    // TODO(developer): Update these variables before running the sample.
    String project = "PROJECT_ID";
    String gcsDestinationOutputUriPrefix = "gs://MY_BUCKET/";

    createBatchPredictionGeminiJobSample(project, gcsDestinationOutputUriPrefix);
  }

  // Create a batch prediction job using a JSONL input file and output URI, both in Cloud
  // Storage.
  public static BatchPredictionJob createBatchPredictionGeminiJobSample(
      String project, String gcsDestinationOutputUriPrefix) throws IOException {
    String location = "us-central1";
    JobServiceSettings settings =
        JobServiceSettings.newBuilder()
            .setEndpoint(String.format("%s-aiplatform.googleapis.com:443", location))
            .build();

    // Initialize client that will be used to send requests. This client only needs to be created
    // once, and can be reused for multiple requests.
    try (JobServiceClient client = JobServiceClient.create(settings)) {
      GcsSource gcsSource =
          GcsSource.newBuilder()
              .addUris(
                  "gs://cloud-samples-data/generative-ai/batch/"
                      + "batch_requests_for_multimodal_input.jsonl")
              // Or try
              // "gs://cloud-samples-data/generative-ai/batch/gemini_multimodal_batch_predict.jsonl"
              // for a batch prediction that uses audio, video, and an image.
              .build();
      BatchPredictionJob.InputConfig inputConfig =
          BatchPredictionJob.InputConfig.newBuilder()
              .setInstancesFormat("jsonl")
              .setGcsSource(gcsSource)
              .build();
      GcsDestination gcsDestination =
          GcsDestination.newBuilder().setOutputUriPrefix(gcsDestinationOutputUriPrefix).build();
      BatchPredictionJob.OutputConfig outputConfig =
          BatchPredictionJob.OutputConfig.newBuilder()
              .setPredictionsFormat("jsonl")
              .setGcsDestination(gcsDestination)
              .build();
      String modelName =
          String.format(
              "projects/%s/locations/%s/publishers/google/models/%s",
              project, location, "gemini-1.5-flash-002");

      BatchPredictionJob batchPredictionJob =
          BatchPredictionJob.newBuilder()
              .setDisplayName("my-display-name")
              .setModel(modelName) // Add model parameters per request in the input jsonl file.
              .setInputConfig(inputConfig)
              .setOutputConfig(outputConfig)
              .build();

      LocationName parent = LocationName.of(project, location);
      BatchPredictionJob response = client.createBatchPredictionJob(parent, batchPredictionJob);
      System.out.format("\tName: %s\n", response.getName());
      // Example response:
      //   Name: projects/<project>/locations/us-central1/batchPredictionJobs/<job-id>
      return response;
    }
  }
}

BigQuery input

import com.google.cloud.aiplatform.v1.BatchPredictionJob;
import com.google.cloud.aiplatform.v1.BigQueryDestination;
import com.google.cloud.aiplatform.v1.BigQuerySource;
import com.google.cloud.aiplatform.v1.JobServiceClient;
import com.google.cloud.aiplatform.v1.JobServiceSettings;
import com.google.cloud.aiplatform.v1.LocationName;
import java.io.IOException;

public class CreateBatchPredictionGeminiBigqueryJobSample {

  public static void main(String[] args) throws IOException {
    // TODO(developer): Update these variables before running the sample.
    String project = "PROJECT_ID";
    String bigqueryDestinationOutputUri = "bq://PROJECT_ID.MY_DATASET.MY_TABLE";

    createBatchPredictionGeminiBigqueryJobSample(project, bigqueryDestinationOutputUri);
  }

  // Create a batch prediction job using BigQuery input and output datasets.
  public static BatchPredictionJob createBatchPredictionGeminiBigqueryJobSample(
      String project, String bigqueryDestinationOutputUri) throws IOException {
    String location = "us-central1";
    JobServiceSettings settings =
        JobServiceSettings.newBuilder()
            .setEndpoint(String.format("%s-aiplatform.googleapis.com:443", location))
            .build();

    // Initialize client that will be used to send requests. This client only needs to be created
    // once, and can be reused for multiple requests.
    try (JobServiceClient client = JobServiceClient.create(settings)) {
      BigQuerySource bigquerySource =
          BigQuerySource.newBuilder()
              .setInputUri("bq://storage-samples.generative_ai.batch_requests_for_multimodal_input")
              .build();
      BatchPredictionJob.InputConfig inputConfig =
          BatchPredictionJob.InputConfig.newBuilder()
              .setInstancesFormat("bigquery")
              .setBigquerySource(bigquerySource)
              .build();
      BigQueryDestination bigqueryDestination =
          BigQueryDestination.newBuilder().setOutputUri(bigqueryDestinationOutputUri).build();
      BatchPredictionJob.OutputConfig outputConfig =
          BatchPredictionJob.OutputConfig.newBuilder()
              .setPredictionsFormat("bigquery")
              .setBigqueryDestination(bigqueryDestination)
              .build();
      String modelName =
          String.format(
              "projects/%s/locations/%s/publishers/google/models/%s",
              project, location, "gemini-1.5-flash-002");

      BatchPredictionJob batchPredictionJob =
          BatchPredictionJob.newBuilder()
              .setDisplayName("my-display-name")
              .setModel(modelName) // Add model parameters per request in the input BigQuery table.
              .setInputConfig(inputConfig)
              .setOutputConfig(outputConfig)
              .build();

      LocationName parent = LocationName.of(project, location);
      BatchPredictionJob response = client.createBatchPredictionJob(parent, batchPredictionJob);
      System.out.format("\tName: %s\n", response.getName());
      // Example response:
      //   Name: projects/<project>/locations/us-central1/batchPredictionJobs/<job-id>
      return response;
    }
  }
}

Retrieve batch output

When a batch prediction task completes, the output is stored in the Cloud Storage bucket or the BigQuery table that you specified in your request.

What's next