All pull requests

Pull requests

• Skip bf16_grouped_fallback_test in OSS ROCm CI

  cla signedmeta-exportedmodule: rocm
  #548·bottler opened Oct 11, 2026Contributor·

  ·
  1

• Support sliding-window local attention in FlyDSL backward

  cla signed
  #546·amd-wsung102 opened Oct 9, 2026Collaborator·

  ·
  1

• Remove dead int32-KV branch from both split-K decode ops

  cla signed
  #545·aryaman-gupta opened Oct 9, 2026Contributor·3 of 4 tasks·

  ·
  1

• Add tensor bias in FlyDSL backward on gfx950

  cla signed
  #542·amd-wsung102 opened Oct 6, 2026Collaborator·

  ·
  1

• Route short paged query blocks to the head-packed decode kernel

  cla signed
  #538·RichardChamberlain1 opened Sep 30, 2026·

  ·
  1

• Support sliding-window attention on paged KV and size split-K from chain length

  cla signed
  #537·RichardChamberlain1 opened Sep 30, 2026·

  ·
  1

• Honour per-request seqlen_k and fold GQA query heads into the decode M axis

  cla signed
  #536·RichardChamberlain1 opened Sep 30, 2026·

  ·
  1

• Memoise paged launch metadata on the attention bias

  cla signed
  #535·RichardChamberlain1 opened Sep 30, 2026·

  ·
  1

• Split-K combine: correctness and wave parallelism

  cla signed
  #534·RichardChamberlain1 opened Sep 30, 2026·

  ·
  1

• Suppress type errors for Pyre upgrade - mslk

  cla signedmeta-exported
  #532·maggiemoss opened Sep 29, 2026Contributor·

  ·
  1

• FlyDSL paged decode: maintain MFMA occupancy on short-query workload

  cla signed
  #531·RichardChamberlain1 opened Sep 29, 2026·

  ·
  1

• Skip dispatch reporting during torch.compile (#530)

  cla signedmeta-exported
  #530·amolagrawalmeta opened Sep 28, 2026·

  ·
  1

• Stop the CUDA conda-build CI job from OOMing the worker

  cla signedmeta-exported
  #511·sryap opened Sep 4, 2026Contributor·

  ·
  1

• Autotune async copy for rowwise FP8 GEMM (#494)

  cla signedmeta-exported
  #494·warrendeng opened Aug 21, 2026·

  ·

• Add Triton FP8×INT4 rowwise GEMM for ROCm gfx942/gfx950

  cla signedmodule: rocm
  #479·apicciau opened Aug 11, 2026Contributor·

  ·

• Fix flash_attn_bench crash on the flash_attn.cute (FA4) backend: pass window_size=(-1, -1) instead of None

  #397·Maurits-de-Groot opened Jun 19, 2026·

  ·
  3

• [CUDA] [PERFORMANCE] Increase speed of bf16bf16bf16_grouped_wgrad via indicating that ElementC is void / nullptr

  cla signed
  #329·benediktjohannes opened Apr 19, 2026·

  ·
  2

• [WIP] Migrate MOE kernels to libtorch ABI stable

  cla signed
  #325·janeyx99 opened Apr 14, 2026Contributor·

  ·