The AI Reference Implementation Baseline Pattern Module provides a secure, observable by default, scalable, and highly configurable foundation for deploying AI workloads on Azure. This pattern module integrates multiple Azure resources, following best practices and architectural standards, to deliver a comprehensive AI Reference Implementation. The goal is to accelerate the deployment of AI solutions by providing a ready-to-use infrastructure that adheres to Azure's Well-Architected Framework.
This pattern module is opinionated, meaning it comes with pre-configured defaults for security, observability, and essential AI resources. However, it remains flexible, allowing users to customize the environment to meet specific project needs by enabling or disabling various components.
The pattern module is designed to be modular and composable. The default deployment includes a minimum set of resources required to establish a secure and observable AI environment, but additional resources can be added based on project requirements. Below is a high-level architecture diagram:
This AI Reference Implementation pattern module is designed to accelerate the deployment of AI solutions on Azure, while ensuring security, compliance, flexibility, and observability. The primary objectives and functionalities of this module include:
-
Security by Default: Ensuring that all deployed resources adhere to Azure's security best practices, including network isolation, encryption, and identity management. This guarantees that AI environments are secure from the outset. For more details, see Security practices.
-
Observability: Providing out-of-the-box integration of logging, monitoring, and alerting, making the AI environment fully observable from day one. This ensures that all deployments are transparent and issues can be quickly identified and resolved. More details are available in Observability practices.
-
Modular and Flexible Design: Composed of several modules that can be individually enabled or disabled, this pattern offers a flexible architecture that can be tailored to specific project needs. This modular approach allows teams to start with a minimal setup and expand as required, ensuring the AI environment is scalable and adaptable.
-
Compliance with Azure Best Practices: Adhering to the recommendations of the Azure Well-Architected Framework, this module ensures that all resources and configurations are optimized for performance, reliability, security, and cost management.
-
Rapid Deployment and Consistency: By providing a standardized reference implementation, this module accelerates the deployment process and ensures consistency across different projects and teams. This reduces variability and guarantees that best practices are consistently applied in all AI environments.
This pattern module is designed for both data scientists and engineers who need to quickly stand up a secure, scalable AI environment on Azure. It is also suitable for organizations that require a compliant and secure environment for their AI workloads with the flexibility to customize the setup based on project-specific needs.
To deploy the AI Reference Implementation Baseline with minimal configuration:
module "ai_reference_implementation" {
source = "Azure/avm-ptn-ai-reference-implementation/azurerm"
version = "x.x.x"
resource_group_name = "<your_resource_group>"
location = "<your_location>"
}This example sets up the AI Reference Implementation with all default resources.
This module is ideal for:
- Data Scientists: Who need a secure, scalable, and integrated environment to experiment, develop, and train machine learning models.
- ML Engineers: Looking to deploy machine learning models into production with robust monitoring, scaling, and management capabilities.
- Organizations: That require a compliant and secure environment for their machine learning workloads, with the flexibility to integrate with existing Azure services.
The AI Reference Implementation Baseline Pattern Module is designed to be extended. You can add additional resources or services by integrating other Azure Verified Modules (AVM). For example, you can include additional machine learning environments, data lakes, or advanced AI services by simply integrating their respective modules and configuring them within the pattern module.
- Azure Well-Architected Framework: Azure WAF
- Azure AI Documentation: Azure AI Services
- Terraform Registry: Terraform AzureRM Provider
This module is part of the Azure Verified Modules (AVM) ecosystem, and contributions are welcome. Please follow the standard contribution guidelines if you wish to submit enhancements or report issues.
The following requirements are needed by this module:
The following resources are used by this module:
- azurerm_management_lock.this (resource)
- azurerm_public_ip.bastion_ip (resource)
- azurerm_role_assignment.this (resource)
- modtm_telemetry.telemetry (resource)
- random_uuid.telemetry (resource)
- azurerm_client_config.current (data source)
- azurerm_client_config.telemetry (data source)
- azurerm_resource_group.base (data source)
- modtm_module_source.telemetry (data source)
The following input variables are required:
Description: The location/region where the resources will be deployed.
Type: string
Description: The name of the this resource.
Type: string
Description: The resource group where the resources will be deployed.
Type: string
The following input variables are optional (have default values):
Description: The address space that is used for the Azure Bastion subnet
Type: list(string)
Default:
[
"10.1.3.0/24"
]Description: The name of the Azure Bastion resource. if not provided, a name will be generated.
Type: string
Default: ""
Description: The name of the Network Security Group for the Azure Bastion subnet. If not provided, a name will be generated.
Type: string
Default: ""
Description: The name of the Azure Container Registry. If not provided, a name will be generated.
Type: string
Default: ""
Description: This variable controls whether or not telemetry is enabled for the module.
For more information see https://aka.ms/avm/telemetryinfo.
If it is set to false, then no telemetry will be collected.
Type: bool
Default: true
Description: This creates a jumpbox if configured with jumpbox.create = true and defaults to a Windows machine.
Type:
object({
create = bool
name = optional(string, "jumpbox")
os_type = optional(string, "Windows")
size = optional(string, "Standard_D2s_v3")
zone = optional(string, "1")
image_ref = optional(object({
publisher = string
offer = string
sku = string
version = string
}), {
publisher = "microsoftwindowsdesktop"
offer = "windows-11"
sku = "win11-22h2-ent"
version = "latest"
})
})Default:
{
"create": false
}Description: The name of the Azure Key Vault. If not provided, a name will be generated.
Type: string
Default: ""
Description: Controls the Resource Lock configuration for this resource. The following properties can be specified:
kind- (Required) The type of lock. Possible values are\"CanNotDelete\"and\"ReadOnly\".name- (Optional) The name of the lock. If not specified, a name will be generated based on thekindvalue. Changing this forces the creation of a new resource.
Type:
object({
kind = string
name = optional(string, null)
})Default: null
Description: The name of the Log Analytics Workspace. If not provided, a name will be generated.
Type: string
Default: ""
Description: The name of the Azure Machine Learning Workspace. If not provided, a name will be generated.
Type: string
Default: ""
Description: The name of the Network Security Group for the private endpoints subnet. If not provided, a name will be generated.
Type: string
Default: ""
Description: The address space that is used for the private endpoints subnet
Type: list(string)
Default:
[
"10.1.2.0/24"
]Description: A map of role assignments to create on the . The map key is deliberately arbitrary to avoid issues where map keys maybe unknown at plan time.
role_definition_id_or_name- The ID or name of the role definition to assign to the principal.principal_id- The ID of the principal to assign the role to.description- (Optional) The description of the role assignment.skip_service_principal_aad_check- (Optional) If set to true, skips the Azure Active Directory check for the service principal in the tenant. Defaults to false.condition- (Optional) The condition which will be used to scope the role assignment.condition_version- (Optional) The version of the condition syntax. Leave asnullif you are not using a condition, if you are then valid values are '2.0'.delegated_managed_identity_resource_id- (Optional) The delegated Azure Resource Id which contains a Managed Identity. Changing this forces a new resource to be created. This field is only used in cross-tenant scenario.principal_type- (Optional) The type of theprincipal_id. Possible values areUser,GroupandServicePrincipal. It is necessary to explicitly set this attribute when creating role assignments if the principal creating the assignment is constrained by ABAC rules that filters on the PrincipalType attribute.
Note: only set
skip_service_principal_aad_checkto true if you are assigning a role to a service principal.
Type:
map(object({
role_definition_id_or_name = string
principal_id = string
description = optional(string, null)
skip_service_principal_aad_check = optional(bool, false)
condition = optional(string, null)
condition_version = optional(string, null)
delegated_managed_identity_resource_id = optional(string, null)
principal_type = optional(string, null)
}))Default: {}
Description: The name of the Azure Storage Account. If not provided, a name will be generated.
Type: string
Default: ""
Description: A map of tags to add to all resources
Type: map(string)
Default: null
Description: The address space that is used for the virtual machines subnet
Type: list(string)
Default:
[
"10.1.1.0/24"
]Description: The name of the Virtual Network. If not provided, a name will be generated.
Type: string
Default: ""
Description: The name of the Network Security Group for the virtual machines subnet. If not provided, a name will be generated.
Type: string
Default: ""
Description: The address space that is used the virtual network
Type: list(string)
Default:
[
"10.1.0.0/16"
]The following outputs are exported:
Description: This is the full output for the resource.
Description: The Azure resource id of the resource.
The following Modules are called:
Source: Azure/avm-res-machinelearningservices-workspace/azurerm
Version: 0.1.1
Source: Azure/avm-res-containerregistry-registry/azurerm
Version: ~> 0.2
Source: Azure/avm-res-network-bastionhost/azurerm
Version: 0.3.0
Source: Azure/avm-res-network-networksecuritygroup/azurerm
Version: ~> 0.2.0
Source: Azure/avm-res-compute-virtualmachine/azurerm
Version: 0.15.1
Source: Azure/avm-res-keyvault-vault/azurerm
Version: ~> 0.5
Source: Azure/avm-res-operationalinsights-workspace/azurerm
Version: ~> 0.1
Source: Azure/avm-res-network-networksecuritygroup/azurerm
Version: ~> 0.2.0
Source: Azure/avm-res-network-privatednszone/azurerm
Version: ~> 0.1.1
Source: Azure/avm-res-network-privatednszone/azurerm
Version: ~> 0.1.1
Source: Azure/avm-res-network-privatednszone/azurerm
Version: ~> 0.1.1
Source: Azure/avm-res-network-privatednszone/azurerm
Version: ~> 0.1.1
Source: Azure/avm-res-storage-storageaccount/azurerm
Version: 0.2.1
Source: Azure/avm-res-network-virtualnetwork/azurerm
Version: ~> 0.2.0
Source: Azure/avm-res-network-networksecuritygroup/azurerm
Version: ~> 0.2.0
The software may collect information about you and your use of the software and send it to Microsoft. Microsoft may use this information to provide services and improve our products and services. You may turn off the telemetry as described in the repository. There are also some features in the software that may enable you and Microsoft to collect data from users of your applications. If you use these features, you must comply with applicable law, including providing appropriate notices to users of your applications together with a copy of Microsoft’s privacy statement. Our privacy statement is located at https://go.microsoft.com/fwlink/?LinkID=824704. You can learn more about data collection and use in the help documentation and our privacy statement. Your use of the software operates as your consent to these practices.