All pull requests

Pull requests

• Fix barrier offset bug in zero-token expert notification path

  #212·dajiaohuang opened Sep 25, 2026·

  ·
  1

• fix(amd): honor the documented stream=None default in rocshmem_barrier_all_on_stream

  #211·Linxiushen opened Sep 21, 2026·

  ·
  1

• [NVIDIA] Add single-node sequence-parallel MLA decode

  #210·0z5a opened Sep 19, 2026·

  ·

• fix: use int64 row offsets in persistent 2D copy

  #209·Levius-Fubuki opened Sep 4, 2026·

  ·
  1

  5

• fix: preserve IPv4 bootstrap when detection is unavailable

  #206·Kotlinliu opened Aug 30, 2026·

  ·
  3

• fix: respect TORCHRUN override in launch script

  #205·Kotlinliu opened Aug 30, 2026·

  ·
  2

• fix(nvidia): promote LL all-to-all v2 buffer offsets to int64

  #204·zkyue opened Aug 29, 2026·

  ·
  1

  1

• Fix non-power-of-two head dimensions in GQA decode

  #202·Skystellan opened Aug 27, 2026·

  ·
  1

  2

• docs: fix LittleKernel test paths

  #197·dajiaohuang opened Aug 24, 2026·

  ·

• docs: correct README links and paths

  #196·dajiaohuang opened Aug 24, 2026·

  ·

• fix: detect repository root in setup

  #193·dajiaohuang opened Aug 24, 2026·

  ·
  1

• docs: remove stale aggregate test commands

  #191·dajiaohuang opened Aug 24, 2026·

  ·
  1

• docs: fix LittleKernel scalar type table

  #189·dajiaohuang opened Aug 24, 2026·

  ·
  1

• fix: accept output path in AMD libdevice generator

  #187·dajiaohuang opened Aug 24, 2026·

  ·
  1

• Fix swiglu backward writing every row's gradient to row 0

  #185·rootkiller6788 opened Aug 21, 2026·

  ·
  1

• Zero-initialize the 1-D reduce_tensor output

  #184·truong-v opened Aug 12, 2026·

  ·
  1

  2

• Mask the histogram reduction in calc_gather_scatter_index_v2_kernel

  #183·truong-v opened Aug 12, 2026·

  ·
  1

  2

• Gemm-allreduce loop optimization

  #170·erieaton-amd opened May 13, 2026Collaborator·

  ·
  1

• Port the low latency allgather kernels to AMD

  #162·erieaton-amd opened Feb 25, 2026Collaborator·

  ·

• Fix AMD allgather

  #161·erieaton-amd opened Feb 18, 2026Collaborator·

  ·
  1

• [AMD] Enable CUDA graph capture with rocshmem_hipmodule_init API

  #158·drprajap opened Feb 10, 2026Contributor·

  ·
  2

• fix 06-inter-node-reduce-scatter.py tutorial bug

  #130·AkiRusProd opened Oct 10, 2025·

  ·
  2

• update amd build doc

  #25·yiakwy-xpu-ml-framework-team opened May 16, 2025·

  ·
  22