Skip to content

Commit

Permalink
Merge pull request #303 from aws-samples/readme-fixes
Browse files Browse the repository at this point in the history
README changes
  • Loading branch information
perifaws authored May 3, 2024
2 parents 3a7008d + 5d07c0e commit 5a6aa82
Showing 1 changed file with 2 additions and 2 deletions.
4 changes: 2 additions & 2 deletions 1.architectures/4.amazon-eks/README.md
Original file line number Diff line number Diff line change
@@ -28,8 +28,8 @@ The following example cluster configurations for distributed training are provid
* [**`eks-g4dn.yaml`**](./eks-g4dn.yaml): Cluster with a nodegroup of 2 * `g4dn.8xlarge` instances, created in a new VPC. This example shows that when a VPC is not specified, one is created for the cluster. The manifest can work without any modifications, however if you wish to change the cluster name, API version, region, availability zones, etc. you can modify the file before using it to create the cluster.
* [**`eks-p4de-odcr-vpc.yaml`**](./eks-p4de-odcr-vpc.yaml): Cluster using an existing VPC with a nodegroup of 2 * `p4de.24xlarge` instances from an existing on-demand capacity reservation (ODCR). This is the most common configuration for distributed training workloads.Edit the file to specify vpc id, subnets, and capacityReservationID. Please note that the subnet of the nodeGroup should match the one of the capacity reservation.
* [**`eks-p4de-odcr.yaml`**](./eks-p4de-odcr.yaml): Cluster with 2 * `p4de.24xlarge` instances from an existing ODCR. A new VPC will be created for this cluster. This configuration is useful for distributed training when no VPC is already available. Note that you would have to match the AZ of your ODCR in the nodegroup section of the manifest. Nodegroups in this and previous examples are fully-managed and can be accessed via the EKS console. If you are using an instance type that is not yet supported in managed nodegroups by EKS, you can define a nodegroup in a self-manged nodegroup section as shown at the end of this example.
* [**`eks-p5-odcr.yaml`**](./eks-p4de-odcr-vpc.yaml): Cluster with 1 * `p5.48xlarge` instances from an existing ODCR and an existing VPC. Note that you would have to match the AZ of your ODCR in the nodegroup section of the manifest. Nodegroups in this and previous examples are fully-managed and can be accessed via the EKS console. If you are using an instance type that is not yet supported in managed nodegroups by EKS, you can define a nodegroup in a self-manged nodegroup by using the `eks-p5-capacity-block.yaml` template.
* [**`eks-p5-capacity-block.yaml`**](./eks-p4de-odcr.yaml): This deploys a cluster without a node group, allowing you to create an unmanaged node group for Capacity Blocks for ML. See the section [Capacity Block](#4-capacity-blocks) for further detail.
* [**`eks-p5-odcr.yaml`**](./eks-p5-odcr.yaml): Cluster with 1 * `p5.48xlarge` instances from an existing ODCR and an existing VPC. Note that you would have to match the AZ of your ODCR in the nodegroup section of the manifest. Nodegroups in this and previous examples are fully-managed and can be accessed via the EKS console. If you are using an instance type that is not yet supported in managed nodegroups by EKS, you can define a nodegroup in a self-manged nodegroup by using the `eks-p5-capacity-block.yaml` template.
* [**`eks-p5-capacity-block.yaml`**](./eks-p5-capacity-block.yaml): This deploys a cluster without a node group, allowing you to create an unmanaged node group for Capacity Blocks for ML. See the section [Capacity Block](#4-capacity-blocks) for further detail.

## 3. Cluster creation

0 comments on commit 5a6aa82

Please sign in to comment.