All pull requests

Pull requests

• Bundle SM90 TLX comparisons and fix GEMM tuning cache (#1322)

  cla signedmeta-exported
  #1322·dshi7 opened Oct 29, 2026Contributor·

  ·
  1

• Bump black from 25.11.0 to 26.10.1

  cla signeddependenciespython
  #1321·dependabot[bot] opened Oct 28, 2026Bot·

  ·

• Fix --benchmark-name crashing regular operators with KeyError

  cla signed
  #1314·u7k4rs6 opened Oct 24, 2026Contributor·

  ·

• Inline/thread constexpr globals without defaults

  cla signedmeta-exported
  #1312·njriasan opened Oct 21, 2026Contributor·

  ·
  3

• Compare production AutoWS and TLX gradients (#1305)

  cla signedmeta-exported
  #1305·manman-ren opened Oct 8, 2026Contributor·

  ·
  1

• Add production shapes (#1304)

  cla signedmeta-exported
  #1304·manman-ren opened Oct 8, 2026Contributor·

  ·
  1

• Pin FA backward CTA count (#1302)

  cla signedmeta-exported
  #1302·manman-ren opened Oct 7, 2026Contributor·

  ·
  2

• Share 32x32 MXFP8 payloads in Flash Attention (#1282)

  cla signedmeta-exported
  #1282·htyu opened Sep 26, 2026Contributor·

  ·
  1

• Add Blackwell MXFP8 Flash Attention to tlx.ops (#1270)

  cla signedmeta-exported
  #1270·htyu opened Sep 16, 2026Contributor·

  ·
  1

• Add Hopper Flash Attention to tlx.ops

  cla signedmeta-exported
  #1267·htyu opened Sep 15, 2026Contributor·

  ·
  2

• Add Hopper TLX Flash Attention kernel

  cla signedmeta-exported
  #1256·htyu opened Sep 4, 2026Contributor·

  ·
  1

• Streamline tritonbench commands torchTLX perf measurement

  cla signedmeta-exported
  #1250·dshi7 opened Aug 27, 2026Contributor·

  ·
  1

• Add variable-length attention shapes to TritonBench

  cla signedmeta-exported
  #1247·htyu opened Aug 25, 2026Contributor·

  ·
  1

• Disable ADS autotuning in TritonBench GPU tests

  cla signedmeta-exported
  #1236·njriasan opened Aug 20, 2026Contributor·

  ·
  1

• run HSTU self-attention compiler DP on b200

  cla signedmeta-exported
  #1202·manman-ren opened Jul 29, 2026Contributor·

  ·
  1

• Add sparsity, target_size, and max_attn_len arguments

  cla signedfb-exportedmeta-exported
  #1051·xuzhao9 opened May 1, 2026Contributor·

  ·
  1

• adding support for every-n-shape test sampling (#992)

  cla signedfb-exportedmeta-exported
  #992·jhou-jpg opened Apr 2, 2026Contributor·

  ·
  1

• [Flex Attention] Add a Triton version so we can test autoWS

  cla signed
  #983·manman-ren opened Mar 30, 2026Contributor·

  ·

• [TLX][FA] update tritonbench version to support rescale opt

  cla signed
  #884·manman-ren opened Feb 21, 2026Contributor·

  ·
  1

• Update gluon_attention_persistent_forward

  cla signed
  #518·neildhar opened Oct 6, 2025Contributor·

  ·

• Add new metrics for operators

  cla signed
  #458·bdbowyer opened Sep 19, 2025·

  ·
  3

• Add backward compatibility for TensorDescriptor

  cla signed
  #457·bdbowyer opened Sep 19, 2025·

  ·

• Add a Blackwell-specific scaled persistent + TMA template for GEMMs

  cla signedfb-exportedmeta-exported
  #432·jananisriram opened Sep 17, 2025Contributor·

  ·
  2

• [gdpa] Turn on on-host TMA for TLX fwd

  cla signed
  #365·htyu opened Aug 28, 2025Contributor·

  ·
  2

• Validate exhaustive autotuning for FP8 Inductor templates

  cla signedfb-exported
  #355·jananisriram opened Aug 25, 2025Contributor·

  ·
  5