This document is relevant for: Inf1, Inf2, Trn1, Trn2, Trn3

AWS Neuron Direct Resource Allocation (DRA) on Kubernetes: Support files

This directory contains scripts, manifests, and templates supporting AWS Neuron Direct Resource Allocation (DRA) on Kubernetes. You can view and download these files from the links below.

Download the scripts and YAML files as a TAR/GZIP archive

Neuron DRA support files as .tar.gz

Preserve the directory structure when you extract the archive. The driver installation script uses this relative folder structure to find the corresponding YAML files.

Directory structure:

containers/files/
           ├── manifests/
           │   ├── clusterrole.yaml
           │   ├── clusterrolebinding.yaml
           │   ├── daemonset.yaml
           │   ├── deviceclass.yaml
           │   ├── namespace.yaml
           │   └── serviceaccount.yaml
           └── examples/
               ├── scripts/
               │   └── install-dra-driver.sh
               └── specs/
                   ├── 1x4-connected-devices.yaml
                   ├── 2-node-inference-us.yaml
                   ├── 4-node-inference-us.yaml
                   ├── all-devices.yaml
                   ├── lnc-setting-trn2.yaml
                   └── specific-driver-version.yaml

Installation Scripts

These scripts automate the deployment and configuration of the Neuron DRA driver on your Kubernetes cluster.

File Name	Description	Download
install-dra-driver.sh	Automated deployment script for the Neuron DRA driver that applies all necessary manifests and waits for successful deployment.	Download

Kubernetes Manifests

Core Kubernetes resources required to deploy and configure the Neuron DRA driver with proper RBAC permissions.

File Name	Description	Download
clusterrole.yaml	ClusterRole definition with permissions required for the Neuron DRA driver to manage device resources.	Download
clusterrolebinding.yaml	ClusterRoleBinding that associates the service account with the required cluster role permissions.	Download
daemonset.yaml	DaemonSet configuration for deploying the Neuron DRA driver on all compatible Trainium nodes.	Download
deviceclass.yaml	DeviceClass resource that defines the Neuron device class for DRA resource allocation.	Download
namespace.yaml	Namespace definition for isolating Neuron DRA driver resources within the cluster.	Download
serviceaccount.yaml	ServiceAccount configuration for the Neuron DRA driver with appropriate security context.	Download

Resource Claim Specifications

Example resource claim templates and pod specifications demonstrating different Neuron device allocation patterns for various workload requirements.

File Name	Description	Download
1x4-connected-devices.yaml	Resource claim template for allocating 4 connected Neuron devices with topology constraints for optimal performance.	Download
2-node-inference-us.yaml	Multi-node inference configuration for distributed workloads across 2 Trainium nodes.	Download
4-node-inference-us.yaml	Large-scale inference setup for distributed workloads spanning 4 Trainium nodes.	Download
all-devices.yaml	Resource claim template that allocates all available Neuron devices on a trn2.48xlarge instance.	Download
lnc-setting-trn2.yaml	Logical NeuronCore configuration template optimized for Trainium2 instances.	Download
specific-driver-version.yaml	Example configuration for requesting specific Neuron driver versions in resource claims.	Download

This document is relevant for: Inf1, Inf2, Trn1, Trn2, Trn3