Options

Every option is a change to inputs.yaml followed by a regenerate, or a small edit to the generated files. This page shows which.

How to change an option

Edit inputs.yaml.
Validate: Test-MLOpsAcceleratorInput -InputFile inputs.yaml.
Regenerate over an existing project (commit first, then review the diff):

Deploy-MLOpsAccelerator -InputFile inputs.yaml -OutputPath ./my-mlops -Upgrade

Add -Deploy only if you also need to re-run the bootstrap (new repository, environments or identity). Changes to the workload and infrastructure are rolled out by the pipelines after you merge to main.

Warning

Regenerating overwrites files. Anything you edited by hand in the generated project is replaced, so keep local edits small and review the diff.

Options

All inputs

Input	Type	Default	Notes
`project_name`	string	required	3 to 10 lowercase alphanumerics, used in resource names
`location`	string	required	Azure region, for example `eastus2`
`subscription_id`, `tenant_id`	string	required
`iac`	`terraform` or `bicep`	required
`cicd`	`github` or `azdo`	required
`scenario`	string	required	See scenarios
`data_source`	`taxi-csv`, `taxi-parquet`, `nyc-taxi-open` or `adls`	`taxi-csv`
`environments`	list	`[dev, test, prod]`	Ordered subset
`tags`	map	`{}`
`github_org`, `github_repo`	string		`cicd: github`
`azdo_org_url`, `azdo_project`	string		`cicd: azdo`
`workload_type`	`tabular`	`tabular`	Kind of workload pack. `nlp` and `vision` are planned. The old name `workload` still works and prints a deprecation warning
`vnet_address_space`	CIDR	`10.20.0.0/16`	Scenario 4
`allowed_outbound_fqdns`	list	PyPI and conda hosts	Scenario 3
`policy_mode`	`audit`, `deny` or `none`	`audit` (`none` for scenario 11)	Azure Policy
`compute_vm_size`	string	`Standard_DS3_v2`	Training cluster and online endpoint VM size
`compute_max_nodes`	integer	`4`	Training cluster maximum nodes
`workloads`	list	Single workload	Workload names (not types) sharing the platform; reserved names such as `src`, `data` and `mlops` are rejected; each gets `workloads/<name>/` and its own pipelines
`use_private_networking`	bool	`false`	GitHub only, rejected with Azure DevOps: deploy self-hosted ACI runners in a VNet and make the state storage private-endpoint only
`runners_vnet_address_space`	string	`10.250.0.0/24`	Runner VNet address space
`runner_label`	string	`ubuntu-latest`	GitHub `runs-on` label for generated workflows. Must be a self-hosted label when `use_private_networking` is true
`approvers`	list	`[]`	GitHub usernames that must approve deployments to non-dev environments
`name_suffix`	string	derived	exactly 5 lowercase alphanumerics; override to redeploy after a destroy (soft-delete holds names)
`keyvault_retention_days`	integer	`7`	Key Vault soft-delete retention (7 to 90 days, fixed at creation)
`budget_amount`	number	`500`	Scenarios 9 and 11
`alert_emails`	list	`[]`	Scenarios 9 and 11
`existing_platform`	object		Scenario 11
`adls`	object		`data_source: adls`

The authoritative definition is schema/inputs.schema.json.

Validation rules

The inputs are checked against the schema and a set of semantic rules before anything is generated.

rag, existing_platform and adls are accepted only with rag-llmops, existing-platform and data_source: adls.
use_private_networking: true is rejected with cicd: azdo and needs a self-hosted runner_label. Labels that start with ubuntu-, windows- or macos-, or end with -latest, are rejected.
Environments are sorted into dev, test, prod order whatever order you write. dev is optional. Registry promotion scenarios need at least two environments. A warning is printed when prod is first, or when a non-dev environment has no approvers on GitHub.
vnet_address_space (scenario 4) must be a network address in private space (10.0.0.0/8, 172.16.0.0/12 or 192.168.0.0/16) with a prefix from /16 to /18. runners_vnet_address_space needs a prefix from /16 to /27. The two must not overlap.
name_suffix is exactly 5 lowercase letters or digits.