Sagemaker training jobs with arm64 (Graviton) processor

Question

I don't see any instance type selection with graviton arch64 in Sagemaker training job ? I checked with even bringing my own algorithm of ECR container and native sagemaker algorithm.

Answer

Hi,

There are plenty of Graviton-based Sagemaker instances: see https://aws.amazon.com/about-aws/whats-new/2022/10/amazon-sagemaker-adds-new-graviton-based-instances-model-deployment/

```
Amazon SageMaker expands access to eight new Graviton2 and Graviton3-based machine learning (ML) 
instance families so that customers have more options for optimizing their cost and performance when 
deploying their ML models on SageMaker. Now, customers can use ml.c7g, ml.m6g, ml.m6gd, ml.c6g, 
ml.c6gd, ml.c6gn, ml.r6g, and ml.r6gd for Real-time and Asynchronous Inference model deployment 
options.
```

Best,

Didier

Answer

You will have to Create an endpoint configuration for the type of instance you want. You can refer this [Blog](https://aws.amazon.com/blogs/machine-learning/run-machine-learning-inference-workloads-on-aws-graviton-based-instances-with-amazon-sagemaker/) page as it highlights information around running workloads on AWS Graviton with Amazon Sagemaker.

Sagemaker training jobs with arm64 (Graviton) processor

Relevant content