All pull requests

Pull requests

• Avoid quadratic allocation-planner free-buffer scans

  #33407·Rui Ren (rui-ren) opened Oct 16, 2026Contributor·

  ·
  1

• [CUDA plugin EP] Improve NuGet package metadata

  #33405·Tianlei Wu (tianleiwu) opened Oct 16, 2026Contributor·

  ·
  1

• Design post-partition runtime fusion for existing CUDA kernels

  #33404·titaiwangms with CopilotAI opened Oct 15, 2026·

  ·
  1

  3

  

• [float16] Keep float outputs of LayerNormalization, MelWeightMatrix and window ops

  #33402·Kayvan Zahiri (Kayvan-Zahiri) opened Oct 15, 2026·

  ·
  4

• Fix QLinearSoftmax output for empty inputs

  #33401·David Luong (DavidLuong98) opened Oct 15, 2026Member·

  ·
  1

  2

• [CUDA] Add source-ordered CUDA IQ2_XXS, Q8_0 and NVFP4 A16 matmul

  #33400·Rui Ren (rui-ren) opened Oct 15, 2026Contributor·

  ·
  2

  

• Fix CPU Range integer boundary counts and final increment overflow

  #33399·Christopher Boseak (cboseak) opened Oct 15, 2026·

  ·
  1

  3

• Validate quantization parameters before QDQ Add, Mul and Concat fusion

  #33397·rookieLiu2018 opened Oct 15, 2026Contributor·

  ·
  1

  2

• [WebGPU] Automate Dawn upgrades and component inventories

  #33396·Yang Gu (gyagp) opened Oct 15, 2026Contributor·

  ·
  1

• [WebGPU] Unroll bounded NVIDIA FlashAttention reductions in WGSL

  #33395·Yang Gu (gyagp) opened Oct 15, 2026Contributor·

  ·
  1

• P0/P1: packed-code foundation and versioned INT3 QMoE contract

  #33394·David Fan (jiafatom) opened Oct 15, 2026Contributor·

  ·
  15

• Optimize Qwen CUDA sparse attention, indexer, and quantized MoE kernels

  #33393·Akshay Sonawane (apsonawane) opened Oct 15, 2026Contributor·

  ·
  9

• Skip MatMul BatchNormalization fusion when the channel axis moves

  #33392·Samridh (samridh10exe) opened Oct 15, 2026·

  ·
  1

  3

• Revert "webgpu: Fix Python factory selection for plugin EP hosts (#33241)"

  #33391·Jiajia Qin (qjia7) opened Oct 15, 2026Contributor·

  ·
  4

• [WebGPU] Upgrade Dawn to the latest version

  #33390·Jiawei Shao (Jiawei-Shao) opened Oct 15, 2026Contributor·

  ·
  15

• Bump github/issue-labeler from 3.4 to 3.5

  dependenciesgithub_actions
  #33388·dependabot[bot] opened Oct 15, 2026Bot·

  ·
  3

• Bump actions/download-artifact from 8.0.1 to 8.0.2

  dependenciesgithub_actions
  #33387·dependabot[bot] opened Oct 15, 2026Bot·

  ·
  3

• Bump actions/checkout from 6 to 7

  dependenciesgithub_actions
  #33386·dependabot[bot] opened Oct 15, 2026Bot·

  ·
  3

• Bump github/codeql-action/autobuild from 4.38.2 to 4.38.3

  dependenciesgithub_actions
  #33385·dependabot[bot] opened Oct 15, 2026Bot·

  ·
  2

• [CUDA] Add HyperConnectionLowRankMix fused low-rank hyper-connection op

  #33384·Tianlei Wu (tianleiwu) opened Oct 15, 2026Contributor·3 tasks done·

  ·

• [CUDA] Add state_status output to PackedSparseAttentionIndexer

  #33383·Tianlei Wu (tianleiwu) opened Oct 15, 2026Contributor·2/2·3 tasks done·

  ·

• Fix ConvMulFusion with non-finite scales

  #33382·King Star (jstar0) opened Oct 14, 2026·

  ·
  1

  3

• Fix arena configuration handling and optimize CUDA GEMV and prefill

  #33381·Akshay Sonawane (apsonawane) opened Oct 14, 2026Contributor·

  ·
  3

• Skip QLinearWhere fusion for per-axis quantization parameters

  #33380·Samridh (samridh10exe) opened Oct 14, 2026·

  ·
  1

  3

• Fix GatherToSliceFusion with negative Range limit

  #33377·Sahil Choudhary (Choudhary-Sahil099) opened Oct 14, 2026·

  ·
  1

  5