chainer-sagemaker-tools

This repository is a collection of tools to run SageMaker jobs.

It contains

• CLIs : Some command line tools to use SageMaker easily. See below guidelines.
• sage_extensions : Some Extensionf for Chainer Trainer.

Installation

$ git clone https://github.com/tn1031/chainer-sagemaker-tools.git
$ cd chainer-sagemaker-tools
$ pip install .

When installing to the ML instances, since installing to them is based on the contents of requirements.txt , it is necessary to contain the below line on requirements.txt.

git+https://github.com/tn1031/chainer-sagemaker-tools.git

Then put the file in the source_dir .

Run SageMaker training jobs

smtrain is a command line tool to run SageMaker training jobs.

Usage

$ smtrain <job_name> <path_to_setting> [-p <aws_profile_name>]

• job_name - Training job name. It must be unique in the same AWS account.
• path_to_setting - Path to the setting file. The format of this file is described in here.
• aws_profile_name - The name of profile that are stored in ~/.aws/config .

Deploy trained model

smdeploy is a command line tool to deploy.

Usage

$ smdeploy <endpoint_name> <path_to_setting> [-p <aws_profile_name>]

• endpoint_name - Endpoint name.
• path_to_setting - Path to the setting file. The format of this file is described in here.
• aws_profile_name - The name of profile that are stored in ~/.aws/config .

Run batch inference

smbatch is a command line tool to run batch inference.

Usage

$ smbatch <model_name> <path_to_setting> [-p <aws_profile_name>]

• model_name - Model name which used for inference.
• path_to_setting - Path to the setting file. The format of this file is described in here.
• aws_profile_name - The name of profile that are stored in ~/.aws/config .