An open-source, single-cluster observability UI for heterogeneous accelerators managed by HAMi
HAMi-WebUI visualizes accelerator inventory, allocation state, and available vendor telemetry across Kubernetes nodes. It provides:
- Cluster overview: Review detected accelerator capacity, allocation, and available utilization signals.
- Node inventory: Inspect accelerator nodes, their devices, and resource state.
- Accelerator visibility: Inspect per-device allocation and vendor telemetry.
- Workload visibility: Inspect current HAMi Pod/container assignments and the monitoring data available for their devices.
Current provider integrations include NVIDIA GPUs, Huawei Ascend 910B/310P, Hygon DCUs, Cambricon MLUs, and MetaX GPUs/sGPUs. Metric coverage depends on the HAMi integration and exporter available for each device type.
HAMi-WebUI is read-only. It does not schedule resources, create or mutate workloads, change nodes, provide built-in authentication or user RBAC, or aggregate multiple clusters. Deploy one instance per cluster and protect it with an authenticated Ingress or proxy unless it is reachable only from a trusted network.
If you're interested in contributing to HAMi-WebUI:
- Start by reading the Contributing guide.
- Set up your local environment by following our Developer guide.
- Explore good first issues.
HAMi-WebUI is discussed in the HAMi community. See the canonical HAMi community channels for current meetings, Discord, Slack, and the mailing list.
HAMi-WebUI is distributed under Apache-2.0. For details, see LICENSE.