All pull requests

Pull requests

• Build macOS NuGet binary without built-in WebGPU EP

  #34173·Edward Chen (edgchen1) opened Nov 13, 2026Contributor·

  ·
  1

• [CPU, CUDA] Add opset 29 Pack and Unpack kernels

  #34172·Justin Chu (justinchuby) opened Nov 13, 2026Contributor·

  ·
  2

• Fix CUDA training compatibility with CCCL 3

  #34171·Justin Chu (justinchuby) opened Nov 13, 2026Contributor·

  ·
  1

• [CUDA] Share MoE GEMM tactic profiling across identical-shape layers

  #34169·BBaditi opened Nov 12, 2026Contributor·

  ·
  7

• [CUDA] Size QMoE split-K SwiGLU GEMV partials by GEMV rows, not prefill rows

  #34168·BBaditi opened Nov 12, 2026Contributor·

  ·
  2

• [CUDA] Reserve MoE GEMM profiler scratch outside the arena

  #34167·BBaditi opened Nov 12, 2026Contributor·

  ·
  4

• [CoreML] Fall back for padded Conv with a widely dilated kernel

  #34166·Vijay Kapse (vijay-kapse) opened Nov 12, 2026Contributor·

  ·
  1

  3

• [CUDA] Mamba2Prefill op and fixed GQA query chunks for Nemotron Bolt bulk prefill (DONOT Merge)

  #34165·Rui Ren (rui-ren) opened Nov 12, 2026Contributor·

  ·
  5

• [CI] Repair portable rolling ccache snapshots for Linux builds

  #34163·eserscor opened Nov 12, 2026Contributor·

  ·
  2

• Enable sccache for Windows CI builds

  #34162·eserscor opened Nov 12, 2026Contributor·

  ·
  2

• [CUDA] Reuse session temporary allocator for plugin graph capture

  #34161·Sayan Shaw (sayanshaw24) opened Nov 12, 2026Contributor·

  ·
  18

• Apply OneCore CRT selection to Windows-hosted ARM64 cross builds

  #34160·adrastogi opened Nov 12, 2026Contributor·

  ·
  1

  

• Preserve small softmax losses and training gradients

  #34158·Justin Chu (justinchuby) opened Nov 12, 2026Contributor·

  ·
  6

• Prepare CPU/CUDA Gemm and GroupedMatMul kernels for opset 29

  #34157·Justin Chu (justinchuby) opened Nov 12, 2026Contributor·

  ·
  4

• Stage opset 29 indexing kernels for CPU and CUDA

  #34156·Justin Chu (justinchuby) opened Nov 12, 2026Contributor·

  ·
  3

• Prepare opset 29 normalization and causal convolution CPU/CUDA kernels

  #34155·Justin Chu (justinchuby) opened Nov 12, 2026Contributor·

  ·
  1

• [CPU/CUDA] Prepare DequantizePacked opset 29 kernels

  #34154·Justin Chu (justinchuby) opened Nov 12, 2026Contributor·

  ·
  1

• Add CPU and CUDA kernels for opset 29 stable math

  #34152·Justin Chu (justinchuby) opened Nov 12, 2026Contributor·

  ·
  4

• Replace vulnerable C# DocFX template

  #34151·eserscor opened Nov 12, 2026Contributor·

  ·
  1

• Adopt model editor custom ops

  #34149·Yateng Hong (toothache) opened Nov 12, 2026Member·

  ·
  2

• Fix symbolic Concat value propagation with negative axes

  #34144·itsrayeedscodespace opened Nov 12, 2026·

  ·
  1

  2

• Constant fold a sliced Shape when only the selected dims are known

  #34142·Qiming Bao (14H034160212) opened Nov 12, 2026·

  ·
  1

  2

• Treat an empty Conv bias as absent in Conv+BN/Add/Mul fusions

  #34141·Qiming Bao (14H034160212) opened Nov 12, 2026·

  ·
  1

  2

• Add OrtEpApi::GetTelemetryEventsEnabled for plugin EP telemetry consent

  #34135·adrastogi opened Nov 12, 2026Contributor·1/2·

  ·
  1

• Bump ruff from 0.12.12 to 0.17.4

  dependenciespython
  #34130·dependabot[bot] opened Nov 12, 2026Bot·

  ·
  2