All pull requests

Pull requests

• Update parameterized requirement from >=0.8.1 to >=0.9.0

  dependenciespython
  #33022·dependabot[bot] opened Oct 1, 2026Bot·

  ·
  3

• Bump numpy from 2.4.2 to 2.4.6

  dependenciespython
  #33021·dependabot[bot] opened Oct 1, 2026Bot·

  ·
  5

• Bump clang-format from 20.1.8 to 23.1.1

  dependenciespython
  #33020·dependabot[bot] opened Oct 1, 2026Bot·

  ·
  2

• [CUDA] Bound GQA non-windowed workspace via the KV-length envelope (route-aware L1 estimate)

  #33016·Chi Lo (chilo-ms) opened Sep 30, 2026Contributor·

  ·
  3

• [WebGPU EP] gated delta net subgroups optimization

  #33015·Prathik Rao (prathikr) opened Sep 30, 2026Contributor·

  ·
  2

• [WebGPU EP] matmul n-bits decode optimization

  #33009·Prathik Rao (prathikr) opened Sep 30, 2026Contributor·

  ·
  19

• Test transformer requirements workflow against main

  #33004·eserscor opened Sep 30, 2026Contributor·

  ·

• [MIGraphX] Fix empty model cache dir

  #33000·Christian Sonnabend (ChSonnabend) opened Sep 30, 2026·

  ·
  3

• Preserve graph interfaces during cleanup

  #32993·Yuhx141 opened Sep 30, 2026Contributor·

  ·
  2

  3

• Keep opset 11 DequantizeLinear schema valid

  #32992·Yuhx141 opened Sep 30, 2026Contributor·

  ·
  1

  7

• Bump transformers to 5.10.1

  #32989·eserscor opened Sep 30, 2026Contributor·

  ·

• Discover GPUs through D3DKMT on WSL2

  #32988·Flora Cui (fcui-amd) opened Sep 30, 2026·

  ·
  6

• Fix safety guards in training loss fusions

  #32987·Yuhx141 opened Sep 30, 2026Contributor·

  ·
  2

  7

• Require zero threshold in IsInf ReduceSum fusion

  #32986·Yuhx141 opened Sep 30, 2026Contributor·

  ·
  1

  9

• Fix LayerNorm fusion safety checks

  #32984·Yuhx141 opened Sep 30, 2026Contributor·

  ·
  5

  5

• Preserve double constants in graph fusions

  #32983·Yuhx141 opened Sep 30, 2026Contributor·

  ·
  2

  19

• Fix correctness guards in training graph optimizers

  #32982·Yuhx141 opened Sep 30, 2026Contributor·

  ·
  6

  10

• Fix CUDA training stream API callers

  #32981·Yuhx141 opened Sep 30, 2026Contributor·

  ·
  2

• Fix semantic guards in basic graph optimizers

  #32975·Yuhx141 opened Sep 30, 2026Contributor·

  ·
  4

  4

• Fix Split num_outputs with an empty optional input

  #32972·Tianlei Wu (tianleiwu) opened Sep 30, 2026Contributor·

  ·
  4

• Honor explicit rotary positions in CUDA GroupQueryAttention fast decode

  #32951·BBaditi opened Sep 30, 2026·

  ·
  1

• Validate Attention projection and cache dimensions

  #32946·Gopalakrishnan Nallasamy (GopalakrishnanN) opened Sep 29, 2026Contributor·

  ·
  4

• [docs] Policy-based, route-aware workspace estimation design

  #32944·Chi Lo (chilo-ms) opened Sep 29, 2026Contributor·

  ·
  14

• check new cuda image

  #32941·eserscor opened Sep 29, 2026Contributor·

  ·

• [js/webgpu] Submit pending commands before uploading to a GPU buffer

  #32939·Mrmaxmeier opened Sep 29, 2026·

  ·
  1

  2