Skip to content
View mayooot's full-sized avatar
🤔
Woodstock jamming
🤔
Woodstock jamming
  • Beijing

Block or report mayooot

Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
mayooot/README.md

Hi, I'm Ming 👋

I build AI infrastructure on Kubernetes — currently working on multi-billion-scale image vector retrieval (Milvus / FAISS) and AI platform (training & inference) at Megvii.

What I work on

  • ☸️ Kubernetes internals: operators, GPU scheduling (HAMi / Volcano / Kueue / Device Plugin)
  • 🔍 Vector search at scale: index selection, recall/latency/memory benchmarking, capacity planning
  • 🖥️ GPU infrastructure: NCCL tuning, DCU/NPU enablement, container-level GPU management — earlier built a multi-tenant GPU cloud (A100/H800, per-second billing) serving external users
  • 🏗️ Multi-tenant K8s clusters (VCluster) — previously at DataCanvas
  • 📊 Measuring before believing — most of my work starts with a benchmark

Upstream contributions kubernetes/kubernetes · golang/go · apple/containerization · vcluster · aibrix · HAMi · kueue · all PRs →

Writing: Zhihu column (Chinese)

📫 harrymingh@gmail.com

Pinned Loading

  1. gpu-docker-api gpu-docker-api Public

    Easier than K8s to lift and lower the gpu number of docker container and scale capacity size of volume.

    Go 82 15

  2. detect-gpu detect-gpu Public

    Detect-GPU is an http server that detect the host for NVIDIA GPU info.

    Go 21 1

  3. build-nccl-tests-with-pytorch build-nccl-tests-with-pytorch Public

    Build NCCL-Tests and configure SSHD in PyTorch container to help you test NCCL faster!

    Cuda 13

  4. alfred-json-format alfred-json-format Public

    🤘An Alfred workflow that formatting a copied json string to the pasteboard using jq.

    14 2

  5. kubernetes/kubernetes kubernetes/kubernetes Public

    Production-Grade Container Scheduling and Management

    Go 126k 44k

  6. loft-sh/vcluster loft-sh/vcluster Public

    vCluster creates tenant clusters: fully isolated environments delivered as managed Kubernetes, or as the foundation for Slurm, Ray, Run:ai and inference clusters. Each gets its own API server, CRDs…

    Go 11.3k 597