Vertex AI AutoML Tabular model — Batch Prediction fails immediately with INTERNAL (code 13)

Google Cloud Build with AI Custom ML & MLOps
,
1

Hi,

I have an AutoML Tabular regression model in us-central1 that used to run Batch Prediction successfully against a BigQuery source (instancesFormat: bigquery). Since yesterday, every batch job fails within ~20–40 seconds with:

  • state: JOB_STATE_FAILED

  • error.code: 13

  • error.message: INTERNAL

The job input/output config hasn’t changed (same model ID/version, same BigQuery input table and output dataset). I also tried lowering replicas (2 → 1) and running on only 10 rows in a separate probe table — still the same INTERNAL error. Billing is enabled on the project.

Has anyone seen AutoML batch jobs suddenly fail with generic INTERNAL with no worker logs? Any known outages or workarounds (e.g., re-export model, new training run)?

Thanks.