sagemaker

name	sagemaker
description	Manage Amazon SageMaker notebooks, training jobs, models, endpoints, and pipelines via AWS CLI.
metadata	{"openclaw":{"emoji":"🧠","requires":{"bins":["aws"]}}}

Amazon SageMaker

Use this skill for machine learning operations: managing notebook instances, training jobs, model hosting, batch transforms, endpoints, and ML pipelines.

Prerequisites

• AWS CLI v2 configured with valid credentials
• IAM permissions: sagemaker:* for full access, or scoped policies
• SageMaker execution role with access to S3, ECR, and other needed services

Common Operations

List and Inspect (Read-Only)

# List notebook instances
aws sagemaker list-notebook-instances \
  --query 'NotebookInstances[*].[NotebookInstanceName,InstanceType,NotebookInstanceStatus]' --output table

# Describe a notebook instance
aws sagemaker describe-notebook-instance --notebook-instance-name <name>

# List training jobs
aws sagemaker list-training-jobs --sort-by CreationTime --sort-order Descending \
  --query 'TrainingJobSummaries[*].[TrainingJobName,TrainingJobStatus,CreationTime]' --output table

# Describe a training job
aws sagemaker describe-training-job --training-job-name <job-name>

# List models
aws sagemaker list-models \
  --query 'Models[*].[ModelName,CreationTime]' --output table

# List endpoints
aws sagemaker list-endpoints \
  --query 'Endpoints[*].[EndpointName,EndpointStatus,CreationTime]' --output table

# Describe an endpoint
aws sagemaker describe-endpoint --endpoint-name <endpoint-name>

# List endpoint configs
aws sagemaker list-endpoint-configs \
  --query 'EndpointConfigs[*].[EndpointConfigName,CreationTime]' --output table

# List processing jobs
aws sagemaker list-processing-jobs --sort-by CreationTime --sort-order Descending \
  --query 'ProcessingJobSummaries[*].[ProcessingJobName,ProcessingJobStatus]' --output table

# List pipelines
aws sagemaker list-pipelines \
  --query 'PipelineSummaries[*].[PipelineName,PipelineStatus]' --output table

# List pipeline executions
aws sagemaker list-pipeline-executions --pipeline-name <pipeline-name> \
  --query 'PipelineExecutionSummaries[*].[PipelineExecutionArn,PipelineExecutionStatus,StartTime]' --output table

# List experiments
aws sagemaker list-experiments \
  --query 'ExperimentSummaries[*].[ExperimentName,CreationTime]' --output table

# List model packages
aws sagemaker list-model-packages --output table

# List transform jobs
aws sagemaker list-transform-jobs --sort-by CreationTime --sort-order Descending \
  --query 'TransformJobSummaries[*].[TransformJobName,TransformJobStatus]' --output table

Notebook Instances

⚠️ Cost note: Notebook instances charge per hour while running. ml.t3.medium: ~$0.05/hr. ml.p3.2xlarge (GPU): ~$3.83/hr. Stop when not in use!

# Create a notebook instance
aws sagemaker create-notebook-instance \
  --notebook-instance-name <name> \
  --instance-type ml.t3.medium \
  --role-arn <sagemaker-role-arn> \
  --volume-size-in-gb 20

# Start a notebook instance
aws sagemaker start-notebook-instance --notebook-instance-name <name>

# Stop a notebook instance
aws sagemaker stop-notebook-instance --notebook-instance-name <name>

# Get presigned URL to access notebook
aws sagemaker create-presigned-notebook-instance-url \
  --notebook-instance-name <name> \
  --query 'AuthorizedUrl' --output text

Training Jobs

# Create a training job
aws sagemaker create-training-job \
  --training-job-name <job-name> \
  --algorithm-specification '{
    "TrainingImage": "<ecr-image-uri>",
    "TrainingInputMode": "File"
  }' \
  --role-arn <sagemaker-role-arn> \
  --input-data-config '[{
    "ChannelName": "train",
    "DataSource": {"S3DataSource": {"S3DataType": "S3Prefix", "S3Uri": "s3://<bucket>/train/"}}
  }]' \
  --output-data-config '{"S3OutputPath": "s3://<bucket>/output/"}' \
  --resource-config '{
    "InstanceType": "ml.m5.xlarge",
    "InstanceCount": 1,
    "VolumeSizeInGB": 50
  }' \
  --stopping-condition '{"MaxRuntimeInSeconds": 3600}' \
  --hyper-parameters '{"epochs": "10", "batch_size": "32"}'

# Stop a training job
aws sagemaker stop-training-job --training-job-name <job-name>

# Get training job logs
aws logs get-log-events \
  --log-group-name /aws/sagemaker/TrainingJobs \
  --log-stream-name <job-name>/algo-1 \
  --query 'events[*].message' --output text

Models and Endpoints

# Create a model
aws sagemaker create-model \
  --model-name <model-name> \
  --primary-container '{
    "Image": "<ecr-image-uri>",
    "ModelDataUrl": "s3://<bucket>/output/<job-name>/output/model.tar.gz"
  }' \
  --execution-role-arn <sagemaker-role-arn>

# Create an endpoint config
aws sagemaker create-endpoint-config \
  --endpoint-config-name <config-name> \
  --production-variants '[{
    "VariantName": "primary",
    "ModelName": "<model-name>",
    "InstanceType": "ml.m5.large",
    "InitialInstanceCount": 1,
    "InitialVariantWeight": 1.0
  }]'

# Create an endpoint
aws sagemaker create-endpoint \
  --endpoint-name <endpoint-name> \
  --endpoint-config-name <config-name>

# Invoke an endpoint (for testing)
aws sagemaker-runtime invoke-endpoint \
  --endpoint-name <endpoint-name> \
  --content-type application/json \
  --body '{"instances": [{"features": [1.0, 2.0, 3.0]}]}' \
  /dev/stdout

# Update an endpoint (blue/green)
aws sagemaker update-endpoint \
  --endpoint-name <endpoint-name> \
  --endpoint-config-name <new-config-name>

Batch Transform

# Create a transform job
aws sagemaker create-transform-job \
  --transform-job-name <job-name> \
  --model-name <model-name> \
  --transform-input '{
    "DataSource": {"S3DataSource": {"S3DataType": "S3Prefix", "S3Uri": "s3://<bucket>/input/"}},
    "ContentType": "text/csv"
  }' \
  --transform-output '{"S3OutputPath": "s3://<bucket>/predictions/"}' \
  --transform-resources '{"InstanceType": "ml.m5.xlarge", "InstanceCount": 1}'

Delete / Destructive

🛑 DESTRUCTIVE — Always confirm with the user before executing any of these commands.

# Delete an endpoint (stops billing)
aws sagemaker delete-endpoint --endpoint-name <endpoint-name>

# Delete an endpoint config
aws sagemaker delete-endpoint-config --endpoint-config-name <config-name>

# Delete a model
aws sagemaker delete-model --model-name <model-name>

# Delete a notebook instance (must be stopped first)
aws sagemaker delete-notebook-instance --notebook-instance-name <name>

Safety Rules

1. NEVER delete endpoints or models without explicit user confirmation.
2. NEVER expose or log AWS credentials or model data.
3. ALWAYS stop notebook instances when not in use — they charge per hour.
4. ALWAYS set MaxRuntimeInSeconds on training jobs to prevent runaway costs.
5. WARN about GPU instance costs before creating training jobs or endpoints.
6. WARN that endpoints charge continuously until deleted.

Best Practices

• Always set stopping conditions on training jobs.
• Use spot instances for training to save 60-90%.
• Stop notebook instances when not in use.
• Use SageMaker Pipelines for reproducible ML workflows.
• Monitor endpoint invocation metrics to right-size instances.

Common Patterns

Pattern: End-to-End ML Workflow

# 1. Train
aws sagemaker create-training-job --training-job-name my-model-v1 ...

# 2. Wait for completion
aws sagemaker wait training-job-completed-or-stopped --training-job-name my-model-v1

# 3. Create model from trained artifacts
MODEL_DATA=$(aws sagemaker describe-training-job --training-job-name my-model-v1 \
  --query 'ModelArtifacts.S3ModelArtifacts' --output text)
aws sagemaker create-model --model-name my-model-v1 \
  --primary-container "{\"Image\":\"<image>\",\"ModelDataUrl\":\"$MODEL_DATA\"}" \
  --execution-role-arn <role>

# 4. Deploy
aws sagemaker create-endpoint-config --endpoint-config-name my-config-v1 ...
aws sagemaker create-endpoint --endpoint-name my-endpoint --endpoint-config-name my-config-v1

Pattern: Check Running Costs

echo "=== Running Notebooks ==="
aws sagemaker list-notebook-instances --status-equals InService \
  --query 'NotebookInstances[*].[NotebookInstanceName,InstanceType]' --output table

echo "=== Active Endpoints ==="
aws sagemaker list-endpoints --status-equals InService \
  --query 'Endpoints[*].[EndpointName]' --output table

Troubleshooting

Error	Cause	Fix
ResourceNotFound	Resource doesn't exist	Verify name with list commands
Training job failed	Algorithm error or resource issue	Check CloudWatch logs for the training job
Endpoint creation stuck	Instance provisioning issue	Check describe-endpoint for failure reason
ModelError on invoke	Model loading or inference error	Check endpoint CloudWatch logs
ThrottlingException	API rate limit	Reduce request rate; use exponential backoff
High endpoint latency	Instance undersized	Scale up instance type or add auto-scaling