Version 25.1.5 of AIsware is now available.

What's new​

BGP forwarding rules sorting fixed​

Firewall forwarding rules were not correctly sorted, causing the BGP blackhole rule (Drop ANY from Internet to BGP) to not be evaluated last. This has been fixed by adding a destination wildcard to enforce correct rule ordering.

Bond updelay fix prevents pod restarts on network events​

Leaf switch rejoin events were triggering unnecessary pod restarts due to missing bond updelay configuration. Network flaps no longer cascade into workload disruption.

Kubernetes cluster readiness accurately reported​

Clusters were incorrectly showing CREATE_COMPLETE before GPU Operator and HAMi were fully ready. Clusters now remain in BOOTSTRAPPING until fully operational, and kubeconfig is only exposed once Nova confirms the status.

Pre-upgrade health checks​

A new pre-upgrade checks role validates configuration and environment before upgrades begin, catching misconfigurations early and preventing failures mid-upgrade.

Per-GPU metrics in Grafana​

GPU dashboards now show individual GPU utilization metrics instead of an average across all GPUs in a VM, giving operators accurate per-device visibility.

Version 25.1.3 of AI Factory is now available

Overview​

This is a bugfix version

Version 25.1.2 of AI Factory is now available.

Overview​

This is a bugfix version

Version 25.1.1 of AI Factory is now available.

Overview​

This is a bugfix version

Version 25.1.0 of AI Factory is now available.

Overview​

These release notes describe the revised steps, configuration details, and changes for provisioning and managing an AI Factory cluster under the new release.