Documentation
Search…
Launch with Amazon SageMaker

Beta product in active development

Interested in Launch? Reach out to your account team to talk about joining the customer pilot program for W&B Launch.
Pilot customers need to use AWS EKS or SageMaker to qualify for the beta program. We ultimately plan to support additional platforms.
Sagemaker is Amazon's ML development platform. Using Launch you can build and execute training jobs on Sagemaker.

SageMaker quickstart

Requirements

  • The system should be configured with AWS by running the command: aws configure
  • Create an ECR repository and give access to the AWS user associated with the machine that will run launch.
  • Create a Sagemaker execution role with access to pull images from the ECR Repository, and if you intend on outputting data to S3, access to push to the desired S3 bucket.
  • Make sure you have the latest version of wandb and you have Launch enabled by adding instant replay to your W&B profile.
You should now be able to run wandb launch <URI> --resource sagemaker --resource-args <JSON args>, or select the GCP Vertex resource option from the web UI.

Resource args reference

The arguments below should be supplied as a JSON dictionary under the key sagemaker and passed in via --resource-args or via the web UI. Arguments in bold are required.
Name
Type
Description
EcrRepoName
string
An ECR repository that wandb launch can send the built container image to.
region
string
The AWS region to use, can override the default found in ~/.aws/config
profile
string
the local aws profile configuration to use, defaults to default
OutputDataConfig
dict
configuration indicating where to send training job output data
ResourceConfig
dict
configuration indicating the resource to use to run the training job
StoppingCondition
dict
dictionary indicating the conditions which will cause the training job to stop
RoleArn
string
the AWS role used to execute training jobs on Sagemaker.
Sagemaker training jobs are highly configurable and additional resource arguments can be provided. For an exhaustive list of all possible arguments see Sagemaker API reference.
Note that if the AlgorithmSpecification argument is provided and contains aTrainingImage key the value of this key will be used without a build step.
Copy link
Outline
SageMaker quickstart
Requirements
Resource args reference