Repositories list

• uccl

  Public
  UCCL is an efficient communication library for GPUs, covering collectives, P2P (e.g., KV cache transfer, RL weight transfer), and EP (e.g., GPU-driven)

  ainetworkinghpc+ 12

  ainetworkinghpcamdgpucollectivecudap2pnvidiabroadcom+ 5

  C++

  •

  Apache License 2.0

  •160160 forks•1.4k1.4k stars•5858 issues•77 pull requests•Updated Jul 8, 2026Jul 8, 2026

• CommBench

  Public
  Can LLMs Write Correct and Efficient GPU Communication Code?

  Python

  •22 forks•4444 stars•00 issues•11 pull request•Updated Jul 7, 2026Jul 7, 2026

• rdmatop

  Public
  htop-like TUI for real-time RDMA network monitoring.

  linuxrustnetworking+ 6

  linuxrustnetworkingmonitoringhpctuirdmainfinibandefa

  Rust

  •

  Apache License 2.0

  •55 forks•7171 stars•11 issue•00 pull requests•Updated Jul 7, 2026Jul 7, 2026

• mKernel

  Public
  mKernel: fast multi-node, multi-GPU fused kernels

  Cuda

  •

  MIT License

  •2424 forks•249249 stars•11 issue•22 pull requests•Updated Jun 21, 2026Jun 21, 2026

• uccl-project.github.io

  Public
  Astro

  •

  MIT License

  •33 forks•33 stars•11 issue•11 pull request•Updated Jun 14, 2026Jun 14, 2026

• nixl

  Public
  NVIDIA Inference Xfer Library (NIXL)

  C++

  •

  Apache License 2.0

  •366366 forks•00 stars•00 issues•00 pull requests•Updated Nov 24, 2025Nov 24, 2025

• vllm

  Public
  A high-throughput and memory-efficient inference and serving engine for LLMs

  Python

  •

  Apache License 2.0

  •19k19k forks•00 stars•00 issues•11 pull request•Updated Nov 3, 2025Nov 3, 2025

• ray-uccl

  Public
  Ray is an AI compute engine. Ray consists of a core distributed runtime and a set of AI Libraries for accelerating ML workloads.

  Python

  •

  Apache License 2.0

  •7.8k7.8k forks•00 stars•00 issues•00 pull requests•Updated Jul 22, 2025Jul 22, 2025

• nccl

  Public
  Optimized primitives for collective multi-GPU communication

  C++

  •

  Other

  •1.3k1.3k forks•00 stars•00 issues•00 pull requests•Updated Jul 4, 2025Jul 4, 2025