All pull requests

Pull requests

• [CPU] Speed up sub-byte DequantizeLinear (int4/uint4/int2/uint2)

  #33784·Bharat Kaurav (TheNetherWatcher) opened Oct 31, 2026·

  ·
  3

• Update Kusto dashboard SDK dependencies to 6.0.0

  #33783·Mustapha Jaber (mustjab) opened Oct 31, 2026Contributor·

  ·
  1

• fix(web): handle opaque origins for bundled file URLs

  #33782·Lenin Mookiah (leninworld) opened Oct 31, 2026Contributor·

  ·
  1

  2

• Align Windows WebGPU profiling and visualize CPU/GPU timelines

  #33781·Yang Gu (gyagp) opened Oct 31, 2026Contributor·

  ·
  1

• Fix spelling typos flagged by misspell linter

  #33780·adrastogi opened Oct 31, 2026Contributor·

  ·
  1

• Make misspell lint actionable: PR-only, blocking, no false positives

  #33779·adrastogi opened Oct 31, 2026Contributor·

  ·
  1

• Migrate GitHub Actions away from deprecated macos-14 image

  #33778·Edward Chen (edgchen1) opened Oct 31, 2026Contributor·

  ·
  1

• Refresh Docker base pulls and coordinated Node tooling

  #33777·eserscor opened Oct 30, 2026Contributor·

  ·

• Avoid repeated CUDA device-property queries during convolution plan rebuilds

  #33776·tianleiwu with CopilotAI opened Oct 30, 2026·

  ·
  1

  1

  

• [CPU] Fuse 16-bit QuantizeLinear -> DequantizeLinear into QuantizeDeq…

  #33775·mirounga opened Oct 30, 2026Contributor·

  ·
  2

• Delete Implementation Documentation File

  #33774·aciddelgado opened Oct 30, 2026Contributor·

  ·
  1

• Fix macOS vcpkg binary-cache environment identity

  #33773·eserscor opened Oct 30, 2026Contributor·

  ·
  1

• Improve GetPrimitveElementType

  #33772·Christian Bourjau (cbourjau) opened Oct 30, 2026Contributor·

  ·
  3

• [CPU] Relax GroupQueryAttention fused RoPE validation to 8-aligned head sizes

  #33768·Shoryamishra61 opened Oct 30, 2026·

  ·
  7

• [WebGPU] Add per-channel (block_size = K) DP4A MatMulNBits path

  #33767·Zhenwei Jin (kylo5aby) opened Oct 30, 2026Contributor·

  ·
  3

• Skip MatMulAdd reshaping for zero-sized dimensions

  #33766·rookieLiu2018 opened Oct 30, 2026Contributor·

  ·
  1

  2

• [WebGPU] Isolate graph buffer pools and release by ownership

  #33764·Xiaofei Han (xiaofeihan1) opened Oct 30, 2026Contributor·2/2·

  ·
  1

• [WebGPU] Synchronize Session recording across Run, replay, and CPU IoBinding

  #33763·Xiaofei Han (xiaofeihan1) opened Oct 30, 2026Contributor·1/2·

  ·
  2

• [CUDA] GroupQueryAttention: opt-in tensor-core split-KV decode kernel

  #33759·Yuto Ashida (y-chan) opened Oct 30, 2026·

  ·
  6

• [CUDA] Add independent FLOAT/BF16 unit-Swish runtime fusion

  #33757·Ti-Tai Wang (titaiwangms) opened Oct 30, 2026Contributor·

  ·
  2

• [CUDA] Skip the tiled Transpose kernel below half a tile of rows

  #33756·Athrva Pandhare (athrva98) opened Oct 30, 2026·

  ·
  5

• Fix JSEP Reshape and Unsqueeze schema registrations

  #33749·Justin Chu (justinchuby) opened Oct 29, 2026Contributor·

  ·
  4

• [CUDA] Route training-only cuDNN calls through the runtime loader

  #33744·Dante (KafKafrnZ) opened Oct 29, 2026·

  ·
  1

  3

• fix: transpose-replacement-zero-sized-dims

  #33743·Vinay Pepakayala (vinay-852) opened Oct 29, 2026Contributor·

  ·
  1

  2

• fix(web): use the session's WebGPU device id for gpu-buffer inputs and outputs

  #33742·Salman Muin Kayser Chishti (salmanmkc) opened Oct 29, 2026·

  ·
  1

  3