All pull requests

Pull requests

• Optimize rank-major combine and replay synchronization

  #913·Mahdieh Ghazi (mahdiehghazim) opened Sep 23, 2026Contributor·

  ·
  1

• [MSCCLPP-EP] Python interface for the C++ expert-parallel runtime

  #912·Binyang Li (Binyang2014) opened Sep 22, 2026Contributor·

  ·
  1

• PortChannel: add optional GPUNetIO backend via upstream submodule

  #910·Qinghua Zhou (seagater) opened Sep 22, 2026Contributor·

  ·
  20

• [MSCCLPP-EP] C++ throughput expert-parallel runtime

  #903·Binyang Li (Binyang2014) opened Sep 10, 2026Contributor·

  ·
  82

• [MSCCLPP-EP] Add C++ latency expert-parallel runtime

  #895·Binyang Li (Binyang2014) opened Sep 3, 2026Contributor·

  ·
  34

• add dsl algorithms support to bench collective

  #887·RJ Souza (rj-souza) opened Aug 25, 2026Contributor·

  ·
  27

• [Not ready] MSCCLPP EP implementation

  #852·Binyang Li (Binyang2014) opened Jul 23, 2026Contributor·

  ·
  39

• Add switch-channel NVLS cross-rank barrier

  #843·RJ Souza (rj-souza) opened Jul 16, 2026Contributor·

  ·
  21

• Extending DSL Executor CI tests for multi node envinronment

  #812·Caio Rocha (caiocbr) opened May 22, 2026Contributor·

  ·
  2

• Introduce GPUBufferPool for symmetric memory support

  #798·Binyang Li (Binyang2014) opened May 6, 2026Contributor·

  ·
  18

• Migrate MoE dispatch / combine primitives into MSCCL++

  #796·Qinghua Zhou (seagater) opened May 4, 2026Contributor·

  ·
  26

• Add FP8 conversion unit tests

  #789·Binyang Li (Binyang2014) opened Apr 17, 2026Contributor·

  ·

• Support TBG inside Pipeline

  #783·Caio Rocha (caiocbr) opened Apr 11, 2026Contributor·

  ·
  10

• TorchComms Integration for MSCCL++

  #771·Michael Beebe (michael-beebe) opened Apr 5, 2026·

  ·
  20

• Unique QP per channel and env-controlled GID index for executor

  #762·Binyang Li (Binyang2014) opened Mar 9, 2026Contributor·

  ·
  10

• Align the data element counts for Allreduce nvls kernel

  #697·Qinghua Zhou (seagater) opened Dec 1, 2025Contributor·

  ·
  9

• Alltoallv with dynamic execution plan template

  #629·Qinghua Zhou (seagater) opened Sep 9, 2025Contributor·

  ·

• Track Intra Rank Data Dependency and Automatically Sync Thread in the same Rank

  #627·Caio Rocha (caiocbr) opened Sep 8, 2025Contributor·

  ·

• Fifo test with data transfer between multiple GPUs

  #596·Qinghua Zhou (seagater) opened Aug 5, 2025Contributor·

  ·

• Try to extend the connection

  #578·Binyang Li (Binyang2014) opened Jul 23, 2025Contributor·

  ·

• Fifo test with multiple gpus

  #573·Qinghua Zhou (seagater) opened Jul 22, 2025Contributor·

  ·
  33

• Add indirect connection support

  #571·Zhaoyang (Cyrus) Hong (hfhongzy) opened Jul 19, 2025·

  ·
  4

• Exchange recvbuff at the beginning of the CUDA kernel call.

  #464·SreevatsaAnantharamu opened Feb 19, 2025Contributor·

  ·
  1

• A two-stage copy design with scratch buffer

  #445·SreevatsaAnantharamu opened Jan 8, 2025Contributor·

  ·
  1

• Support Simple protocol plans using scratch buffer

  #371·Ziyue Yang (yzygitzh) opened Oct 22, 2024Contributor·

  ·
  6