Getting Started Guide Launch EC2 with Habana¶
This document provides guides and instructions for how to set up a Habana Deep Learning AMI on Amazon EC2 services, and provides release notes for the Habana image.
On the Amazon Web Services (AWS) platform Habana makes available different VMIs, known within the AWS ecosystem as an Amazon Machine Image (AMI). These are Deep Learning optimized AMIs for AWS instances with Habana Training Gaudi Cards.
There are two core types of Habana AMIs:
Deep Learning AMI (DLAMI)
Preconfigured with Habana SynapseAI software and AWS’s modified machine learning frameworks TensorFlow and PyTorch.
Deep Learning Base AMI
A lightweight AMI preconfigured with Habana SynapseAI software and Docker engine.
To see a list of available Habana AMIs refer to AWS Habana MarketPlace
AWS supplies Habana’s Gaudi through dl1 instance types. Combining a Habana AMI and a dl1 instance type provides a platform for accelerated machine learning and development
For those familiar with the AWS platform, the process of launching the instance is as simple as logging in, selecting the Habana AMI of choice, configuring settings as needed, then launching the Instance. After launching the instance, you can SSH into it and start building a host of AI applications in deep learning, machine learning and data science by leveraging the Gaudi Hardware to achieve optimal accelerated training and development.
These instructions assume the following:
You have an AWS account - https://aws.amazon.com
Perform these preliminary setup steps before creating a EC2 Instance.
Dl1 Availability Zones¶
DL1 instance types are available in US East (N. Virginia) and US West (Oregon) regions. Within those regions, dl1s are only supported in specific availability zones. Follow the steps below to find out the supported availability zones:
Open the Amazon EC2 console at https://console.aws.amazon.com/ec2/v2/home.
In the navigation bar at the top of the screen, the current Region is displayed (for example, US East (Ohio)). Choose either US East (N. Virginia) or US West (Oregon) region.
Select Instance Types.
Search for dl1.
Click on the dl1.24xlarge instance type.
Find Availability Zone in the Networking Section and keep note of the listed zone. Please Note that Availability Zone will be different depending on the current region specified.
This ensures a level of network security to prevent unwanted intruders.
Create a Virtual Private Cloud (Get started with Amazon VPC)
Create a Security Group for VPC (Create a Security Group)
Follow Authorize inbound traffic for your Linux instances to get access to the EC2 Instance
Create an EC2 Instance¶
Follow the below step-by-step instructions to launch a Habana EC2 Instance. For more information on launching EC2 instances refer to Get Started with Amazon EC2 Linux.
Initiate Instance Launch¶
Open the Amazon EC2 console at https://console.aws.amazon.com/ec2/.
From the Amazon EC2 console dashboard, choose Launch instance.
Choose an Amazon Machine Image (AMI)¶
Choose a Habana AMI to launch:
Habana Deep Learning Base AMI - Lightweight and training can be done through Docker.
Deep Learning AMI (DLAMI) Habana - Contains machine learning frameworks allowing for local training.
Habana Deep Learning Base AMI¶
This is a lightweight AMI that can be leveraged as a base for custom developed AMIs. It contains the essentials setup to run and train on Gaudi to achieve accelerated deep learning training.
The major difference between the Base AMI and DLAMI, is that the Base AMI does not have any Machine Learning Frameworks such as TensorFlow or PyTorch installed. To perform training, one can develop on Habana Prebuilt Containers or AWS Deep Learning Containers, see Pull Prebuilt Containers or install the desired Machine Learning framework, see Install Native Frameworks.
Search for “Habana Deep Learning Base AMI” and choose the desired operating system. These are located in the AWS Marketplace AMIs.
Deep Learning AMI Habana¶
This is AWS’s constructed Habana AMI that comes preconfigured with Habana SynapseAI software and AWS’s modified machine learning frameworks TensorFlow and PyTorch. Deep Learning training can be performed directly on the instance without the use of Docker.
Search for “Deep Learning AMI Habana” and choose the desired Machine Learning Framework, SynapseAI version and Operating System (These are located in the Community AMIs section).
AMI Names will look like “Deep Learning AMI Habana [TensorFlow, PyTorch] * SynapseAI *.
Choose Instance Type¶
Choose the dl1.24xlarge instance type to run on Habana Gaudi.
Configure Instance Details¶
Choose valid Network and Subnet for secure connections.
Ensure that the dl1.24xlarge is available in the selected subnet region.
Modify storage to suit your needs. dl1.24xlarge comes allocated with four local nvme each 1 TB. However, these are local to the allocated dl1 and will not persist on stopping and starting the instance.
Configure Security Group¶
Choose desired security group for your EC2 instance.
Verify your configuration is correct and proceed to Launch.
Choose/Create a key pair and Launch Instance.
You have now launched an EC2 Instance.
Connect to your Instance¶
Once you establish an ssh connection, run the below command to verify the dl1.24xlarge instance is working:
Now you have the power of Habana’s Gaudi for AI applications.
Pull Habana Docker Images¶
Follow the instructions outlined in Installation Guide to to train on Habana TensorFlow or PyTorch Containers.
Refer to TensorFlow User Guide for instructions on developing with TensorFow.
Refer to PyTorch User Guide for instructions on developing with PyTorch.
Release Notes for Habana Amazon Machine Images on AWS¶
Habana Deep Learning Base AMI provides a foundational platform for deep learning on Amazon EC2 instances with Habana® Gaudi® and Docker. The Habana Gaudi processor is designed to maximize training throughput and efficiency, while providing developers with optimized software and tools that scale to many workloads and systems.
This AMI is suitable for deploying your own custom deep learning environment at scale.
The Habana Deep Learning Base AMI is provided at no additional charge to Amazon EC2 users.
Below are the core components of Habana Deep Learning Base AMI:
Containerization platforms including Docker and habanalabs-container-runtime to run Gaudi accelerated Docker containers.
For an in depth guide of getting started with Gaudi follow the guides located at https://developer.habana.ai/resources/getting-started-with-gaudi/
Refer to the Release Notes of the Gaudi Architecture and Software.
Habana Base Deep Learning AMI Packages¶
Packages installed in the Habana Base AMI.