A Kubernetes device plugin that lets HAMi share AMD GPUs between pods. It limits each pod to its share of the GPU's compute units and memory.
- STATUS.md: what was checked on a real GPU, the problems found and fixed, and what is left.
- ONBOARDING.md: how the project works, for someone new.
The library that limits memory inside pods lives in amd-hami-core; STATUS.md and ONBOARDING.md cover it too. Install steps and settings are in docs, and the changes of each release in CHANGELOG.md.
The same sharing through Kubernetes DRA is amd-dra-driver. A node runs one of the two: the chart keeps this plugin off nodes labelled hami.io/amd-mode=dra.