Open in tt-awesome →

tt-operator

official
Python · Apache-2.0 · 5⭐ ·

The official umbrella Helm chart for running Tenstorrent workloads on Kubernetes. One `helm install` from the OCI registry (`oci://ghcr.io/tenstorrent/helm/tt-operator`) brings up the whole stack as subcharts: Node Feature Discovery to label nodes that carry a Tenstorrent PCI device, `tt-k8s-driver-manager` to own the lifecycle of `tt-kmd`, firmware and `tt-smi` on every node, `tt-fabric-manager` for inter-card and inter-host topology, a Dynamic Resource Allocation driver that publishes cards as `ResourceSlices` (Kubernetes 1.33+), `tt-telemetry` with a Prometheus endpoint, and JobSet plus a PMIx-injecting webhook for multi-node training. Every subchart can be switched off independently, image pins forward to each component, and a `kind` dev loop with fake-labelled nodes lets you work on the controllers without hardware. Releases ship a component version matrix; v0.3.0 (Sep 2026) added a fail-fast check for DRA on older clusters.

LATEST v0.3.0 2026-09-16T16:11:43Z Release notes ↗
4 previous releases
v0.2.0 2026-08-01T04:23:17Z
v0.1.0 2026-07-01T02:16:36Z
v0.0.5 2026-06-30T17:00:07Z
v0.0.4 2026-06-26T21:13:52Z
See all releases on GitHub ↗
kubernetes helm operator orchestration cloud-native dra tt-kmd telemetry multi-node
wormhole blackhole