https://pgabriel-01.github.io/mlops-v3/options/index.htmlEvery 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.Hugoen-ushttps://pgabriel-01.github.io/mlops-v3/options/01-two-environments/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/options/01-two-environments/index.htmlenvironments: [dev, prod] Any ordered subset of dev, test, prod; promotion and approvals follow the order. Each environment is a full workspace, so this roughly cuts the fixed cost by a third (see scenarios). Back to Options.https://pgabriel-01.github.io/mlops-v3/options/02-terraform-or-bicep/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/options/02-terraform-or-bicep/index.htmliac: bicep # or terraform Both build on Azure Verified Modules and produce the same resources. The bootstrap (repository, identity and state storage) is always Terraform regardless of this setting. Back to Options.https://pgabriel-01.github.io/mlops-v3/options/03-github-or-azure-devops/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/options/03-github-or-azure-devops/index.htmlcicd: github github_org: contoso github_repo: mlops-demo # or cicd: azdo azdo_org_url: https://dev.azure.com/contoso azdo_project: mlops Azure DevOps pipelines are generated under pipelines/ and must be registered manually (see Run). Back to Options.https://pgabriel-01.github.io/mlops-v3/options/04-your-own-data-adls/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/options/04-your-own-data-adls/index.htmldata_source: adls adls: account_name: stcontosodata container: raw path: taxi/trips.csv # a CSV with the same columns as the taxi sample The datastore uses identity-based access; grant the workspace identity Storage Blob Data Reader (Grant-MLOpsStorageAccess does this). Snowflake and Azure SQL: land the data in ADLS first, see data sources. The adls source registers one file as a uri_file data asset, not an MLTable like the bundled samples. MLTable support for ADLS is deferred (see Known issues).https://pgabriel-01.github.io/mlops-v3/options/05-outbound-allow-list/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/options/05-outbound-allow-list/index.htmlScenario 3 only. The default allows pypi.org, files.pythonhosted.org, *.anaconda.com and conda.anaconda.org. Replace it with exactly what your environment needs: allowed_outbound_fqdns: - pypi.org - files.pythonhosted.org - internal-mirror.contoso.com Fewer hosts is safer, but training jobs that install packages fail if a host is missing. Back to Options.https://pgabriel-01.github.io/mlops-v3/options/06-vnet-address-space/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/options/06-vnet-address-space/index.htmlScenario 4 only. Pick a range that does not overlap your corporate or hub networks: vnet_address_space: 10.20.0.0/16 Back to Options.https://pgabriel-01.github.io/mlops-v3/options/07-existing-platform/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/options/07-existing-platform/index.htmlScenario 11 only. Supply IDs from your platform team; no network, DNS zones or Log Analytics are created: existing_platform: subnet_id: /subscriptions/.../subnets/snet-aml log_analytics_workspace_id: /subscriptions/.../workspaces/log-platform private_dns_zone_ids: blob: /subscriptions/.../privateDnsZones/privatelink.blob.core.windows.net # file, vault, acr, aml_api, aml_notebooks Back to Options.https://pgabriel-01.github.io/mlops-v3/options/08-budget-alerts/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/options/08-budget-alerts/index.htmlbudget_amount: 500 # USD per month, per environment resource group alert_emails: [mlops-team@contoso.com] Back to Options.https://pgabriel-01.github.io/mlops-v3/options/09-tags/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/options/09-tags/index.htmltags: costcenter: "1234" owner: mlops-team Applied to every resource group and resource. Back to Options.https://pgabriel-01.github.io/mlops-v3/options/10-compute-sizes/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/options/10-compute-sizes/index.htmlSet the VM size and the maximum node count in inputs.yaml, then regenerate. Check quota for the size in your region first. compute_vm_size: Standard_DS3_v2 compute_max_nodes: 4 The training cluster always scales down to 0 nodes. Instance counts for endpoints are not inputs; edit the generated files and commit your change after each -Upgrade review: What Setting Default Training cluster compute_vm_size and compute_max_nodes Standard_DS3_v2, 0 to 4 nodes Online endpoint compute_vm_size for the size; instance_count in mlops/azureml/online-deployment.yml Standard_DS3_v2, 1 Batch endpoint instance_count in mlops/azureml/batch-deployment.yml 1 Back to Options.https://pgabriel-01.github.io/mlops-v3/options/11-switch-scenario/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/options/11-switch-scenario/index.htmlChange scenario and regenerate. Adding serving, a registry, a feature store or governance is a small change. Changing the network mode of an existing workspace is a bigger one and has not been tested; for anything beyond a trial, plan on a new workspace. Back to Options.https://pgabriel-01.github.io/mlops-v3/options/12-policy-mode/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/options/12-policy-mode/index.htmlpolicy_mode: audit # audit, deny or none Built-in Azure Policy guardrails on each environment resource group. audit (default): report non-compliance only. deny: block non-compliant changes. Try it in dev first. none: assign nothing. This is the default for scenario 11, where the platform team owns policy. See Azure Policy for the policies, the reasoning and the FAQ. Back to Options.https://pgabriel-01.github.io/mlops-v3/options/13-csv-or-parquet/index.htmlMon, 01 Jan 0001 00:00:00 +0000https://pgabriel-01.github.io/mlops-v3/options/13-csv-or-parquet/index.htmlChoose the bundled sample format with data_source: data_source: taxi-parquet # or taxi-csv (default) taxi-csv and taxi-parquet hold the same NYC taxi rows and register as MLTable data assets, so the pipeline code is identical. For real parquet, set data_source: nyc-taxi-open. It is a 2,000-row sample of the public Azure Open Datasets yellow taxi data (January 2015) with its original column names such as tpepPickupDateTime and fareAmount. prep.py cleans it and maps it to the model features. To train on more rows, copy parquet files from https://azureopendatastorage.blob.core.windows.net/nyctlc/yellow/ (anonymous read) into your ADLS container and read them with the adls source.