What is HAMi?
HAMi (Heterogeneous AI Computing Virtualization Middleware) formerly known as k8s-vGPU-scheduler, is an "all-in-one" chart designed to manage Heterogeneous AI Computing Devices in a k8s cluster. It can provide the ability to share Heterogeneous AI devices and provide resource isolation among tasks.
HAMi is committed to improving the utilization rate of heterogeneous computing devices in Kubernetes clusters and providing a unified multiplexing interface for different types of heterogeneous devices.
Why HAMi
Kubernetes Native API Compatible
Zero change upgrade: compatible with default behaviour from Kubernetes.
Open and Neutral
Jointly initiated by Internet, finance, manufacturing, cloud providers, etc. Target for open governance with CNCF
Avoid Vendor Lock-in
Integration with mainstream cloud providers; Not tied to proprietary vendor orchestration
Resource Isolation
Provides hard isolation of resources within containers, task in containers can't use resources that exceed their quota
Supports a variety of heterogeneous computing devices
Provides device-sharing on GPU, MLU, NPU from a variety of manufacturers
Unified Management
Unified monitoring system, Configurable scheduling policies(binpack,spread,etc...)
