Version updates

[marosset]

Purpose

• ...

Does this introduce a breaking change?

[ ] Yes
[x] No

Pull Request Type

What kind of change does this Pull Request introduce?

This PR updates versions of various components (Azure APIs, Ray, KubeRay, etc)

[ ] Bugfix
[ ] Feature
[ ] Code style update (formatting, local variables)
[ ] Refactoring (no functional changes, no api changes)
[ ] Documentation content changes
[x] Other... Please describe: Chore

How to Test

• Get the code

git clone [repo-address]
cd [repo-name]
git checkout [branch-name]

• Test the code

# Deploy an AKS cluster (aks-classic example)
cd aks-classic
tofu init && tofu apply

# Get cluster credentials
az aks get-credentials --resource-group <rg-name> --name <cluster-name>

# Install KubeRay operator (should install v1.6.0)
helm repo add kuberay https://ray-project.github.io/kuberay-helm/
helm install kuberay-operator kuberay/kuberay-operator --version 1.6.0 --namespace kuberay --create-namespace

# Download and run the Ray pytorch-mnist job with Ray 2.54.1
curl -LO https://raw.githubusercontent.com/ray-project/kuberay/master/ray-operator/config/samples/pytorch-mnist/ray-job.pytorch-mnist.yaml
sed -i 's/2\.52\.0/2.54.1/g' ray-job.pytorch-mnist.yaml
kubectl apply -f ray-job.pytorch-mnist.yaml

# Monitor job status (takes ~8-10 minutes for pip install + training)
kubectl get rayjob rayjob-pytorch-mnist -w

What to Check

Verify that the following are valid

• KubeRay operator pod is Running in the kuberay namespace with version 1.6.0
• Ray job completes with status SUCCEEDED
• Training accuracy is ~87% on Fashion MNIST (consistent with previous versions)
• Helm chart versions in all deploy.sh scripts reference 1.6.0 (not 1.1.1 or 1.4.2)
• Ray image references use rayproject/ray:2.54.1 (not ray-ml which is discontinued)
  rayVersion fields match 2.54.1 in all YAML manifests
• AKS API version is 2026-01-01 (GA, not preview) in main.tf files
• Kubernetes version defaults are 1.33 in all variables.tf files

Other Information

[mboersma]

lgtm, I checked out this branch and tested the aks-classic variant locally as per the PR instructions. I verified the "What to Check" bullet points.

Training Accuracy passes: 'accuracy': 0.8716

👍🏻