hami
v2.8.0AI & MLJan 20, 2026
Release v2.8.0 adds capabilities and metrics, corrects multiple defects, and updates dependencies. The `nvidia-mig-parted` upgrade addresses security issues.
Action needed (1)
securityThe
nvidia-mig-parteddependency, upgraded tov0.12. 2 HAMi v2.8.0 upgrades the
nvidia-mig-parteddependency tov0.to address security issues.12. 2
All 22 other recorded changesfixes 12 · additions 5 · value changes 4 · constraints 1
fixes (12)
- Fix: Calculation error for quotas
- Fix: vXPU feature may not working properly on P800 node
- Fix scheduler allocate incorrect mig instance
- Fix: failed clusterrolebinding when change release name or chart name
- fix: check pod nil in
ReleaseNodeLock - Fix: After removing the device plugin from the gpu node, it can still
- Fix concurrent map iteration and map write fatal error.
- Fix concurrent map read write fatal error.
- Update device-numa acquisition logic
- Improved support for iluvatar GPUs
- Improve: Replace
StrategicMergePatchTypebyMergePatchType - optimize schedule failure event
additions (5)
- HAMi DRA is now supported
- Enable leader select among multiple schedulers.
- Add hami_build_info metrics
- version print
- Watch and hot reload the updated certificate
value changes (4)
- Sync with k8s-device-plugin from nvidia v0.18.0
- Update HAMi-core to fix vllm-related issues
- add podInfos in DeviceUsage to enhance scheduling decision
- Update go version to v1.25.5
constraints (1)
- Support CDI mode on NVIDIA devices.
Add hami to your stack
A weekly email arrives when a release needs action. Like the security patches in this release.