HAMi Project

HAMi website, join the community https://discord.gg/Amhy7XmbNq

A transparent, in-container GPU resource controller that enforces memory and compute limits by intercepting CUDA calls without application or driver changes.

Kubernetes DRA Driver for Ascend NPUs, supporting soft NPU slicing with hami-vnpu-core

Fast migration from Device Plugin to DRA for HAMi users.

Heterogeneous GPU Sharing on Kubernetes

Kubernetes device plugin for Ascend NPUs, supporting hard and soft NPU slicing with HAMi and Volcano.

Device-plugin for volcano vgpu which support hard resource isolation

HAMi GPU virtualization demo on GKE: vLLM + YOLOv8 multi-workload inference with fractional A100 sharing, Terraform IaC

GPU memory isolation for KAI-Scheduler GPU-sharing workloads, powered by HAMi-core.

Kubernetes device plugin for AMD GPUs, supporting soft vGPU with HAMi.