Skip to main content

Parameter checklist

Work through this before running Deploy-MLOpsAccelerator.

Always

  • project_name (3 to 10 lowercase alphanumerics)
  • location, subscription_id, tenant_id
  • scenario (see scenarios)
  • iac: terraform or bicep
  • cicd: github (github_org, github_repo) or azdo (azdo_org_url, azdo_project)
  • environments: default dev, test, prod; any ordered subset of those (promotion and approvals follow the order)
  • tags

Bootstrap networking

  • use_private_networking (GitHub only): true if policy blocks public state storage; then review runners_vnet_address_space
  • runner_label: ubuntu-latest, or the label of your self-hosted runners

Workloads

  • workloads: optional list; omit for one workload at the repository root

By scenario

  • 3: allowed_outbound_fqdns
  • 4: vnet_address_space
  • 9 and 11: budget_amount, alert_emails
  • 11: existing_platform.subnet_id, .log_analytics_workspace_id, .private_dns_zone_ids (blob, file, vault, acr, aml_api, aml_notebooks)

Prerequisites

See Phase 1 - Prerequisites: tools, Azure permissions and resource providers, GitHub or Azure DevOps token, quota, and self-hosted agents for private scenarios. Test-MLOpsAcceleratorRequirement checks what it can.

Data source

  • data_source: taxi-csv (default), taxi-parquet or nyc-taxi-open (bundled MLTable samples) or adls (your own CSV in ADLS Gen2)
  • For adls: grant the AML workspace identity Storage Blob Data Reader on the account (not created by the IaC)
  • Snowflake or Azure SQL: land the data in ADLS first (ADF or Fabric), then use adls. See External data sources

Environments and cost

  • Default is three environments (dev, test, prod). Each one is a full AML workspace with its dependencies; private scenarios add networking and managed-vnet-approved adds a managed firewall per environment.
  • For a trial or small team use environments: [dev, prod]; promotion and approvals follow the order you list.