https://pgabriel-01.github.io/mlops-v3/faq/index.htmlFrequently asked questions, grouped by topic. Cleanup: remove what the accelerator created. Permissions: who needs which role, and why. Customization: change generated output safely. Miscellaneous: everything else.Hugoen-ushttps://pgabriel-01.github.io/mlops-v3/faq/cleanup/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/faq/cleanup/index.htmlHow do I remove everything? Destroy the workload resources first, one environment at a time: terraform destroy in infra/ (Terraform), or delete the environment resource groups (Bicep). Remove the bootstrap resources (identity, state storage, repository and environments): Remove-MLOpsAccelerator -InputFile inputs.yaml -ProjectPath ./my-mlops It does not touch the workload resource groups, which is why step 1 comes first.https://pgabriel-01.github.io/mlops-v3/faq/permissions/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/faq/permissions/index.htmlWhat does the person running the bootstrap need? Owner on the subscription, or Contributor plus Role Based Access Control Administrator (or User Access Administrator). No Microsoft Entra directory permissions are required, because the identity is an Azure managed identity and not an app registration. See Azure subscriptions and permissions. What does the pipeline identity get? Contributor, Role Based Access Control Administrator and Resource Policy Contributor at subscription scope, and Storage Blob Data Contributor on the state account. Scope them down for production if your organization requires it.https://pgabriel-01.github.io/mlops-v3/faq/customization/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/faq/customization/index.htmlHow do I change the generated output? Prefer an option in inputs.yaml and regenerate. If you edit generated files, keep those changes small and in separate commits so -Upgrade diffs stay readable. See the upgrade guide. Can I run more than one model? Yes. Set workloads: [taxi, churn]. All workloads share the one platform, and each gets workloads/<name>/ and its own train and deploy pipelines.https://pgabriel-01.github.io/mlops-v3/faq/misc/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/faq/misc/index.htmlHas this been applied to live Azure? Partly. Scenarios 1 to 12 are applied live on Terraform and Bicep. The GitHub bootstrap apply and the Bicep pipeline are verified live. Private runners, Azure DevOps and endpoint deployment are not verified live. See known issues. Can I train a model in the tests? The accelerator’s own tests never train a model; they render, validate and plan only. Training runs in your generated ml-train pipeline.