All pull requests

Pull requests

• [CUDA] Include SM103 in Linux aarch64 CUDA 13 packages

  #33649·haoxiz-nvidia opened Oct 27, 2026·

  ·
  1

• Stage Dawn Agility SDK through configured NuGet feed

  #33648·Edward Chen (edgchen1) opened Oct 27, 2026Contributor·

  ·
  1

• Add CUDA BitShift integer kernels

  #33647·Justin Chu (justinchuby) opened Oct 27, 2026Contributor·

  ·
  1

• [CUDA] Add BitwiseAnd, BitwiseOr, and BitwiseNot integer kernels

  #33646·Justin Chu (justinchuby) opened Oct 27, 2026Contributor·

  ·
  1

• Add CUDA BitwiseXor integer kernels

  #33645·Justin Chu (justinchuby) opened Oct 27, 2026Contributor·

  ·
  1

• [CUDA] Add native dynamic W8A8 execution to block-FP8 QMoE

  #33644·Justin Chu (justinchuby) opened Oct 27, 2026Contributor·

  ·
  3

• Fix external tensor copy validation, dense-to-sparse bounds, and ORT format initializer state

  #33643·m00dy (professor-moody) opened Oct 27, 2026·

  ·
  5

• Add pybindings for Model Packaging

  #33640·Preetha Veeramalai (preetha-intel) opened Oct 27, 2026Contributor·

  ·
  1

• [WebGPU] Use shared memory tiles in subgroup matrix MatMul on NVIDIA

  #33639·Rithik Sharma (SharmaRithik) opened Oct 27, 2026Contributor·

  ·
  5

• [WebGPU] Share MatMulNBits input preparation and tune narrow DP4A

  #33635·Jiajia Qin (qjia7) opened Oct 27, 2026Contributor·

  ·
  7

• [WebGPU] Fix plugin tensor conversion for CPU memory descriptors

  #33634·Yang Gu (gyagp) opened Oct 27, 2026Contributor·

  ·
  3

• [CUDA] Enable tanh-GEGLU for MoE and integer QMoE

  #33633·Ti-Tai Wang (titaiwangms) opened Oct 27, 2026Contributor·

  ·
  4

• Validate tensor slicing, sparse CSR shapes, and opset versions

  #33632·m00dy (professor-moody) opened Oct 26, 2026·

  ·
  2

• Add composable C++ graph pattern rewriter

  #33631·G. Ramalingam (gramalingam) opened Oct 26, 2026Contributor·

  ·

• [WebGPU EP] 0.5.0 cherry-picks round 3

  #33630·Edward Chen (edgchen1) opened Oct 26, 2026Contributor·

  ·

• [CUDA/CPU] Add AVX2 INT3 cold-expert kernel, 8-route grouped INT3 offload prefill, and S=8 QMoE offload optimizations

  #33629·Tianlei Wu (tianleiwu) opened Oct 26, 2026Contributor·

  ·
  2

• WebGPU: Enable small-M SIMT MatMulNBits for large INT4 projections

  #33628·aciddelgado opened Oct 26, 2026Contributor·

  ·
  4

• [CUDA] Add opt-in tail expert substitution for hybrid QMoE decode

  #33625·BBaditi opened Oct 26, 2026·

  ·
  9

• [CUDA/CPU] Add AVX2 INT4 cold-expert kernel and zero-cold doorbell short-circuit for QMoE offload

  #33624·BBaditi opened Oct 26, 2026·

  ·
  10

• Fix MulInteger inference for omitted zero points

  #33623·Justin Chu (justinchuby) opened Oct 26, 2026Contributor·

  ·
  1

• [CUDA/CPU] Optimize INT4 QMoE offload prefill/decode and S=8 doorbell execution

  #33622·Tianlei Wu (tianleiwu) opened Oct 26, 2026Contributor·5/5·3 tasks done·

  ·
  3

• Fix: [Python] A requested execution provider that can't be created is silently dropped, e.g. a plugin MIGraphXExecutionProvider

  #33621·SIDDARTHA REDDY (SIDDARTHAREDDY8) opened Oct 26, 2026Contributor·

  ·
  1

  4

• Fix more dead links in docs and training API README

  #33620·Pratik Gandhi (pratikgx) opened Oct 26, 2026Contributor·

  ·
  2

• Migrate managed targets and CI tooling to .NET 10

  #33618·eserscor opened Oct 26, 2026Contributor·2/2·

  ·

• Fix QDQ MatMul fusion by validating quantization parameters

  #33616·Theodore Cooper (the0cp) opened Oct 26, 2026Contributor·

  ·
  1

  4