Getting started

This Getting Started Guide provides the beginning point to start developing and deploying your ML inference applications, whether you are a first time user or if you are looking for specific topic documentation.

Setup Neuron environment

A typical workflow with the Neuron SDK will be to compile trained ML models on a compute instance (compilation instance) and then distribute the artifacts to a fleet of inf1 instances (deployment instance) , for execution and deployment.

image

Note

AWS Deep Learning AMI (DLAMI) is the recommended AMI to use with Neuron SDK.

Step 1 - Compilation Instance

It is recommended to choose c5.4xlarge or larger for compilation instance, however the user can choose to compile and deploy on the same instance, when choosing the same instance for compilation and deployment it is recommend to use an inf1.6xlarge instance or larger.

  1. Launch compilation instance with DLAMI , see Deep Learning AMI for more information, If you choose other AMI launch EC2 instance and choose your AMI of choice.

  2. Install Neuron SDK

Step 2 - Deployment Instance

Deployment instance is the inf1 instance chosen to deploy and execute the user trained model.

  1. Launch inf1 instance with DLAMI , see Deep Learning AMI for more information, If you choose other AMI launch an inf1 instance and choose your AMI of choice.

  2. Install Neuron SDK

Start with ML Framework

Start with Tensorflow

  1. Install Neuron Tensorflow

  2. Run tensorflow-getting-started

  3. Visit TensorFlow Neuron for more resources.

Start with MXNet

  1. Install Neuron MXNet

  2. Run Getting Started with MXNet (ResNet50)

  3. Visit MXNet Neuron for more resources.

Run Tutorials & Examples

Learn Neuron Fundamentals

Get familiar with Neuron fundamentals and tools:

Performance optimization

The following steps are recommended for you to build highly optimized Neuron applications:

  1. Get familiar with Neuron fundamentals and tools:

  2. Learn how to optimize your application by reviewing the HowTo guides at Performance Optimization .