Find file History
Permalink
Type Name Latest commit message Commit time
..
Failed to load latest commit information.
source Remove secrets Jan 11, 2019
.gitignore Fix ignore and reqs Nov 26, 2018
README.md Improvements to sagemaker Jan 11, 2019
requirements.txt Made cifar10 use sagemaker Nov 24, 2018
train.ipynb Fix ignore and reqs Nov 26, 2018
train.py Improvements to sagemaker Jan 11, 2019

README.md

PyTorch Cifar10 SageMaker Sweep

This example uses AWS SageMaker to launch a hyperparameter sweep. train.py uses the AWS SageMaker api to launch a sweep.

If you run this script outside of a SageMaker notebook instance, you'll need to download the aws cli and run aws configure and then set the SAGEMAKER_ROLE environment variable to an AWS role name that has access to SageMaker, you can find your roles here.

To authenticate with W&B, we call wandb.sagemaker_auth(path="source") in train.py. This will look for W&B credentials in the current environment and pass them to sagemaker. Calling wandb login MY_API_KEY on the machine you're running from will ensure credentials get passed.