anoop-ml / sagemaker_pytorch_scatter Goto Github PK

Sample to bootstrap PyTorch Scatter in SageMaker

Pre-reqs

• Setup Machine to build the SageMaker Training container (make sure your drive has enough space for docker images)
  • Option 1 [easy to setup] - Launch a SageMaker notebook on ml.p3.2xlarge and use terminal in JupyterLab
  • Option 2 - Launch a EC2 instance on p3.2xlarge with Deep Learning AMI (Amazon Linux 2) (tested against Version 51.0, ami-021c79f126312c652)

To Run

• Prepare machine to enable GPU access during build (this repo was run in a SageMaker notebook on ml.p3.2xlarge)
  • You need to enable GPU access during docker build
  • Run cd utils;sh setup.sh
• Create a Custom Container image to be used in SageMaker Training
  • Run cd container; sh build_and_push.sh
• Run the notebook to trigger SageMaker Training

Tested on SageMaker notebook on ml.p3.2xlarge and us-east-1