Hacktoberfest 2026: los issues que los mantenedores marcaron para octubre, abiertos y aptos para principiantes. Explorar issues de Hacktoberfest

Vertex batch embeddings (gemini-embedding-001): no way to attach billing labels; job and row labels never reach Cloud Billing

Abierto
  • ¿Algo no está bien?
#3,053 1 comentario 0 reacciones 1 asignado Ver en GitHub

Los mantenedores suelen responder en 1 día

@Venkaiahbabuneelam ya está trabajando en esto.

Desde el 5/10/2026.

Evaluación

Este issue todavía no se ha evaluado.

Descripción

priority: p3 type: question

Billing labels on Vertex AI batch embedding jobs never reach Cloud Billing

Environment: google-genai 2.22.0, Python 3.13, Vertex AI (vertexai=True), us-central1, model gemini-embedding-001

What you're trying to do

We split our Vertex AI costs by custom metadata labels.

  • Real-time calls work: generateContent with GenerateContentConfig(labels=...), and gemini-embedding-001 embed_content (which the SDK sends to :predict) with labels added through http_options.extra_body. Both labels show up in Cloud Billing.
  • Batch embeddings don't: for the same model through client.batches.create(...), no label has ever appeared in billing, whether set on the job or in the JSONL rows.

Questions:

  1. Is there a supported way to attach billing labels to a Vertex batch embedding job?
  2. Batch rows accept request.labels, while every other unknown key fails the row. Is that intended, and should it reach billing?
  3. Could CreateBatchJobConfig and EmbedContentConfig get a labels field, as GenerateContentConfig and GenerateImagesConfig have? Could BatchJob keep labels when reading a job back? Today it drops them.
What code you've already tried
1. Real-time embedding: works

The label shows in billing.

await client.aio.models.embed_content(
    model='gemini-embedding-001',
    contents='...',
    config=types.EmbedContentConfig(
        task_type='RETRIEVAL_DOCUMENT',
        output_dimensionality=768,
        http_options=types.HttpOptions(extra_body={'labels': {'testing': 'rt_predict'}}),
    ),
)

Request as sent (intercepted):

POST https://us-central1-aiplatform.googleapis.com/v1beta1/projects/PROJECT/locations/us-central1/publishers/google/models/gemini-embedding-001:predict
{"instances": [{"content": "...", "task_type": "RETRIEVAL_DOCUMENT"}],
 "parameters": {"outputDimensionality": 768},
 "labels": {"testing": "rt_predict"}}
-> 200

Note: PredictRequest.labels is documented as "for Imagen billing usage only", but these labels do reach billing for this model.

2. Batch job labels: the job stores them, billing doesn't show them

CreateBatchJobConfig has no labels field, so we add them through extra_body:

await client.aio.batches.create(
    model='gemini-embedding-001',
    src=types.BatchJobSource(gcs_uri=['gs://BUCKET/input.jsonl'], format='jsonl'),
    config=types.CreateBatchJobConfig(
        display_name='label_test',
        dest='gs://BUCKET/outputs/',
        http_options=types.HttpOptions(extra_body={'labels': {'testing': 'j_labels'}}),
    ),
)

The job is created, a REST GET on the job returns "labels": {"testing": "j_labels"}, and the rows embed fine. No testing label appears on the cost.

3. Per-row labels

We tried every placement we could think of, one job per placement. Rows look like this:

{"key": "1", "request": {"content": {"parts": [{"text": "..."}]}, "output_dimensionality": 768, "task_type": "RETRIEVAL_DOCUMENT", "labels": {"testing": "r_request"}}}
Label placement in the JSONL line Result
request.labels row embeds fine (accepted), not in billing
request.content.labels row error: no such field: 'labels' (Content)
request.content.parts[0].labels row error: no such field: 'labels' (Part)
request.embed_content_config.labels / request.embedContentConfig.labels row error: no such field: 'labels' (EmbedContentConfig)
request.metadata.labels row error: no such field: 'metadata'
request.request_labels row error: no such field: 'request_labels'
labels object beside request job fails (see errors below)
labels as a JSON string beside request accepted as a pass-through column, not in billing
flat request.content string (:predict-style) row error: can't parse into EmbedContentRequest

Job-level placements:

Placement on the job Result
labels stored on the job, not in billing
modelParameters.labels stored, not in billing
inputConfig.labels, outputConfig.labels, instanceConfig.labels 400 at creation
4. Real-time embedContent with labels

This is the request type batch rows are parsed into. It is rejected on both v1 and v1beta1:

POST .../publishers/google/models/gemini-embedding-001:embedContent
{"content": {"parts": [{"text": "..."}]}, "taskType": "RETRIEVAL_DOCUMENT", "outputDimensionality": 768, "labels": {"testing": "rt_embed_content"}}
-> 400
Any error messages you're getting

No error where it matters: request.labels and job labels are accepted and the jobs succeed, but the labels never reach Cloud Billing.

Errors from the other placements:

Failed to parse JSON into proto: google.cloud.aiplatform.master.EmbedContentRequest with status: invalid JSON in google.cloud.aiplatform.master.EmbedContentRequest @ content: message google.cloud.aiplatform.master.Content, near 1:22 (offset 21): no such field: 'labels'

Failed to parse JSON into proto: google.cloud.aiplatform.master.EmbedContentRequest with status: invalid JSON in google.cloud.aiplatform.master.EmbedContentRequest, near 1:86 (offset 85): no such field: 'metadata'

The column or property "labels" in the specified input data is of unsupported type. Supported types for this column are: [STRING, INTEGER, FLOAT, BOOLEAN, TIMESTAMP, DATE, DATETIME, NUMERIC].

400 INVALID_ARGUMENT. Invalid JSON payload received. Unknown name "labels" at 'batch_prediction_job.input_config': Cannot find field.

400 INVALID_ARGUMENT. Invalid JSON payload received. Unknown name "labels": Cannot find field.   (real-time :embedContent, v1 and v1beta1)
Lenguaje dominante
Python
Estrellas
4k
Forks
1k
Merge medio
1 d 17 h
PR fusionados (30 d)
54

Preparar el entorno

Primeros pasos

  1. Lee el issue completo y luego la guía de contribución del proyecto.
  2. Comenta en el issue que vas a ocuparte — evita que dos personas hagan lo mismo.
  3. Haz un fork del repositorio y trabaja en una rama.
  4. Abre un pull request que haga referencia al número del issue.

Más de googleapis/python-genai

Todos los issues de googleapis/python-genai

Issues similares

Más issues de Python

Recibe los nuevos issues en tu correo

Un resumen breve de issues de GitHub para principiantes.