All pull requests

Pull requests

• Fix Triton norm MXFP4

  ci-level 1
  #747·matthiasdiener opened Sep 30, 2026Contributor·

  ·
  5

  

• FlyDSL persistent work-stealing MXFP8 grouped GEMM (gfx950)

  ci-level 1
  #746·aris134 opened Sep 29, 2026Contributor·

  ·
  

• extend MXFP4 recipe to GroupedLinear and a8w4 (Triton grouped GEMM kernels)

  ci-level 1
  #745·matthiasdiener opened Sep 28, 2026Contributor·

  ·
  30

  

• Replace the gfx950 blockwise FP8 running-scale accumulator

  ci-level 3
  #744·wangye805 opened Sep 26, 2026Collaborator·

  ·
  51

• chore: add security scanning workflows

  #743·haribabug opened Sep 24, 2026·

  ·
  1

• [JAX] grouped GEMM interface fix

  ci-level 3
  #742·shurale-nkn opened Sep 24, 2026Contributor·

  ·

• Enable gfx950 FP8 HD128/256 fused-attn ASM forward from aiter

  community-contribution
  #741·yaomingamd opened Sep 23, 2026·

  ·

• Fix rocprofv3 Hang in JAX Training Docker 26.6/26.7 by Skipping TheRock Core SDK Preload

  community-contribution
  #739·yaomingamd opened Sep 18, 2026·

  ·
  6

• Upgrade ci to therock 10.0

  ci-level 3
  #736·VeeraRajasekhar opened Sep 10, 2026Contributor·

  ·
  15

  

• Persistent AG + GEMM MXFP8 Enablement

  ci-level 1
  #734·aris134 opened Sep 9, 2026Contributor·

  ·
  77

  

• microbenchmarks: use pytest as execution backend

  #726·matthiasdiener opened Aug 31, 2026Contributor·

  ·
  12

  

• [proof-of-concept] comm overlap microbenchmarks

  #724·matthiasdiener opened Aug 31, 2026Contributor·

  ·
  2

  

• Ifu dev 20260828 v2.19

  ci-level 3
  #721·matthiasdiener opened Aug 28, 2026Contributor·

  ·
  82

  

• Add gfx950 MXFP8 CK grouped GEMM

  #718·aris134 opened Aug 26, 2026Contributor·

  ·
  

• Integrate Grouped Gemm v2 on ROCm

  #717·VeeraRajasekhar opened Aug 26, 2026Contributor·

  ·
  

• CI: upload Python coverage.json from pytest (non-blocking)

  community-contribution
  #712·jiagaoxiang opened Aug 21, 2026·

  ·
  1

• [ROCm] Route dense CP softmax-LSE correction through a native kernel

  community-contribution
  #708·zjin-lcf opened Aug 19, 2026·

  ·
  1

• [proof-of-concept] Kernel autotuning in TE

  #705·matthiasdiener opened Aug 18, 2026Contributor·

  ·
  

• [gfx1250] detect when building on gfx1250, add to build archs

  ci-level 1
  #700·matthiasdiener opened Aug 11, 2026Contributor·

  ·
  3

  

• sGPU Test Scheduling: Global Work Queue

  ci-level 3
  #696·VeeraRajasekhar opened Aug 7, 2026Contributor·

  ·
  24

  

• Compile CP softmax LSE corrections with dynamic shapes

  ci-level 3community-contribution
  #695·JessicaJiang-123 opened Aug 6, 2026·

  ·
  3

  

• Permute-Free Grouped GEMM for MoE (bf16, gfx950)

  ci-level 1
  #694·sudhu2k opened Aug 6, 2026Contributor·

  ·
  51

  

• microbenchmarks: usv implementation

  #679·matthiasdiener opened Jul 24, 2026Contributor·

  ·

• Performance dashboard

  #670·matthiasdiener opened Jul 14, 2026Contributor·

  ·
  

• Updated CK/AITER Cmake Build

  ci-level 1
  #666·Micky774 opened Jul 9, 2026Contributor·

  ·
  10