AWS::SageMaker::EndpointConfig ProductionVariant

Specifies a model that you want to host and the resources to deploy for hosting it. If you are deploying multiple models, tell Amazon SageMaker how to distribute traffic among the models by specifying the InitialVariantWeight objects.

Syntax

To declare this entity in your CloudFormation template, use the following syntax:

JSON


{
  "CapacityReservationConfig" : CapacityReservationConfig,
  "ContainerStartupHealthCheckTimeoutInSeconds" : Integer,
  "CoreDumpConfig" : CoreDumpConfig,
  "EnableSSMAccess" : Boolean,
  "InferenceAmiVersion" : String,
  "InitialInstanceCount" : Integer,
  "InitialVariantWeight" : Number,
  "InstancePools" : [ InstancePools, ... ],
  "InstanceType" : String,
  "ManagedInstanceScaling" : ManagedInstanceScaling,
  "ModelDataDownloadTimeoutInSeconds" : Integer,
  "ModelName" : String,
  "RoutingConfig" : RoutingConfig,
  "ServerlessConfig" : ServerlessConfig,
  "VariantInstanceProvisionTimeoutInSeconds" : Integer,
  "VariantName" : String,
  "VolumeSizeInGB" : Integer
}

YAML


  CapacityReservationConfig: 
    CapacityReservationConfig
  ContainerStartupHealthCheckTimeoutInSeconds: Integer
  CoreDumpConfig: 
    CoreDumpConfig
  EnableSSMAccess: Boolean
  InferenceAmiVersion: String
  InitialInstanceCount: Integer
  InitialVariantWeight: Number
  InstancePools: 
    - InstancePools
  InstanceType: String
  ManagedInstanceScaling: 
    ManagedInstanceScaling
  ModelDataDownloadTimeoutInSeconds: Integer
  ModelName: String
  RoutingConfig: 
    RoutingConfig
  ServerlessConfig: 
    ServerlessConfig
  VariantInstanceProvisionTimeoutInSeconds: Integer
  VariantName: String
  VolumeSizeInGB: Integer

Properties

CapacityReservationConfig

Settings for the capacity reservation for the compute instances that SageMaker AI reserves for an endpoint.

Required: No

Type: CapacityReservationConfig

Update requires: Replacement

ContainerStartupHealthCheckTimeoutInSeconds

The timeout value, in seconds, for your inference container to pass health check by SageMaker Hosting. For more information about health check, see How Your Container Should Respond to Health Check (Ping) Requests.

Required: No

Type: Integer

Minimum: 60

Maximum: 3600

Update requires: Replacement

CoreDumpConfig

Specifies configuration for a core dump from the model container when the process crashes.

Required: No

Type: CoreDumpConfig

Update requires: Replacement

EnableSSMAccess

You can use this parameter to turn on native AWS Systems Manager (SSM) access for a production variant behind an endpoint. By default, SSM access is disabled for all production variants behind an endpoint. You can turn on or turn off SSM access for a production variant behind an existing endpoint by creating a new endpoint configuration and calling UpdateEndpoint.

Required: No

Type: Boolean

Update requires: Replacement

InferenceAmiVersion

Specifies an option from a collection of preconfigured Amazon Machine Image (AMI) images. Each image is configured by AWS with a set of software and driver versions. AWS optimizes these configurations for different machine learning workloads.

By selecting an AMI version, you can ensure that your inference environment is compatible with specific software requirements, such as CUDA driver versions, Linux kernel versions, or AWS Neuron driver versions.

The AMI version names, and their configurations, are the following:

al2-ami-sagemaker-inference-gpu-2

Accelerator: GPU
NVIDIA driver version: 535
CUDA version: 12.2

al2-ami-sagemaker-inference-gpu-2-1

Accelerator: GPU
NVIDIA driver version: 535
CUDA version: 12.2
NVIDIA Container Toolkit with disabled CUDA-compat mounting

al2-ami-sagemaker-inference-gpu-3-1

Accelerator: GPU
NVIDIA driver version: 550
CUDA version: 12.4
NVIDIA Container Toolkit with disabled CUDA-compat mounting

al2023-ami-sagemaker-inference-gpu-4-1

Accelerator: GPU
NVIDIA driver version: 580
CUDA version: 13.0
NVIDIA Container Toolkit with disabled CUDA-compat mounting

al2-ami-sagemaker-inference-neuron-2

Accelerator: Inferentia2 and Trainium
Neuron driver version: 2.19

Required: No

Type: String

Allowed values: al2-ami-sagemaker-inference-gpu-2 | al2-ami-sagemaker-inference-gpu-2-1 | al2-ami-sagemaker-inference-gpu-3-1 | al2-ami-sagemaker-inference-neuron-2 | al2023-ami-sagemaker-inference-gpu-4-1

Update requires: Replacement

InitialInstanceCount

Number of instances to launch initially.

Required: No

Type: Integer

Minimum: 1

Update requires: Replacement

InitialVariantWeight

Determines initial traffic distribution among all of the models that you specify in the endpoint configuration. The traffic to a production variant is determined by the ratio of the VariantWeight to the sum of all VariantWeight values across all ProductionVariants. If unspecified, it defaults to 1.0.

Required: No

Type: Number

Update requires: Replacement

InstancePools

A list of instance pools for the production variant. Each instance pool specifies an instance type and its priority for provisioning. Use instance pools to configure heterogeneous endpoints that deploy models across multiple instance types.

Required: No

Type: Array of InstancePools

Minimum: 1

Maximum: 5

Update requires: Replacement

InstanceType

The ML compute instance type.

Required: No

Type: String

Update requires: Replacement

ManagedInstanceScaling

Settings that control the range in the number of instances that the endpoint provisions as it scales up or down to accommodate traffic.

Required: No

Type: ManagedInstanceScaling

Update requires: Replacement

ModelDataDownloadTimeoutInSeconds

The timeout value, in seconds, to download and extract the model that you want to host from Amazon S3 to the individual inference instance associated with this production variant.

Required: No

Type: Integer

Minimum: 60

Maximum: 3600

Update requires: Replacement

ModelName

The name of the model that you want to host. This is the name that you specified when creating the model.

Required: No

Type: String

Pattern: [a-zA-Z0-9]([\-a-zA-Z0-9]*[a-zA-Z0-9])?

Minimum: 0

Maximum: 63

Update requires: Replacement

RoutingConfig

Settings that control how the endpoint routes incoming traffic to the instances that the endpoint hosts.

Required: No

Type: RoutingConfig

Update requires: Replacement

ServerlessConfig

The serverless configuration for an endpoint. Specifies a serverless endpoint configuration instead of an instance-based endpoint configuration.

Required: No

Type: ServerlessConfig

Update requires: Replacement

VariantInstanceProvisionTimeoutInSeconds

The timeout value, in seconds, for provisioning instances for the production variant. When SageMaker encounters an insufficient capacity error while provisioning instances, it retries with the next instance pool (if configured) or waits until the timeout expires. This timeout applies only to capacity provisioning and does not include the time for model download or container startup.

Valid values: 300 to 3600.

Required: No

Type: Integer

Minimum: 300

Maximum: 3600

Update requires: Replacement

VariantName

The name of the production variant.

Required: Yes

Type: String

Pattern: [a-zA-Z0-9](-*[a-zA-Z0-9]){0,62}

Minimum: 0

Maximum: 63

Update requires: Replacement

VolumeSizeInGB

The size, in GB, of the ML storage volume attached to individual inference instance associated with the production variant. Currently only Amazon EBS gp2 storage volumes are supported.

Required: No

Type: Integer

Minimum: 1

Maximum: 512

Update requires: Replacement

Warning Javascript is disabled or is unavailable in your browser.

To use the Amazon Web Services Documentation, Javascript must be enabled. Please refer to your browser's Help pages for instructions.

Document Conventions

PrefixAwareRoutingConfig

RoutingConfig