What is Project-HAMi?

Project-HAMi is a Heterogeneous AI Computing Virtualization Middleware for Kubernetes. It allows for the management, sharing, and optimization of heterogeneous devices like GPUs and NPUs.

What problem does HAMi solve?

HAMi addresses challenges in managing diverse AI hardware, such as resource fragmentation, scheduling inefficiencies, application silos, and overall management complexity.

What are the key features of HAMi?

Key features include device virtualization (sharing and isolation), a unified management interface, topology-aware scheduling, and multi-orchestrator support.

Which devices are supported by HAMi?

HAMi supports a wide range of devices, including Nvidia GPUs, Cambricon MLUs, Hygon DCUs, Iluvatar GPUs, Mthreads GPUs, Ascend NPUs, and Metax GPUs.

How do I install HAMi?

You can install HAMi using Helm after labeling your GPU nodes and meeting the necessary prerequisites.

What is device virtualization in HAMi?

Device virtualization includes device sharing, which allows multiple pods to share a single physical device, and device resource isolation, which ensures dedicated access to allocated resources.

How does HAMi improve resource utilization?

HAMi improves resource utilization by allowing fine-grained allocation of device resources, enabling multiple pods to share devices, and optimizing scheduling based on device topology.

What is the relationship between HAMi and UBOS?

HAMi focuses on efficient hardware resource management, while UBOS is a full-stack AI Agent Development Platform. Together, they enable optimized AI infrastructure and faster AI Agent development.

How can I contribute to Project-HAMi?

You can contribute by reporting issues, submitting code patches, improving documentation, and participating in community discussions.

Where can I find documentation for HAMi?

Documentation can be found in the HAMi repository on GitHub, including guides on installation, configuration, and usage.