Build Windows CUDA Wheels

Workflow runs for Build Windows CUDA Wheels

• 2026-10-25 nightly release (d07c72ba19e9f6adc2e186c4c5f322db690a0251)

  completed successfully:
  Build Windows CUDA Wheels#728:Commit38b0117pushedbypytorchbot
  nightly

  Oct 25, 20261h 54m 59s
  More actions for workflow run 728

• [pytorch hash update] update the pinned pytorch hash

  completed successfully:
  Build Windows CUDA Wheels#727:Pull request#10955synchronizebypytorchupdatebot
  update-pytorch-commit-hash/15080763801-537-1

  Oct 25, 20261h 8m 8s
  More actions for workflow run 727

• Migrate runtime consumers to std::string_view

  completed successfully:
  Build Windows CUDA Wheels#726:Pull request#24139openedbyr-barnes
  r-barnes:export-D127278310

  Oct 24, 20261h 7m 2s
  More actions for workflow run 726

• Fix device placement of constants, shared inputs and delegate outputs

  completed successfully:
  Build Windows CUDA Wheels#725:Pull request#24088synchronizebyshoumikhin
  shoumikhin:device-pass-constants

  Oct 24, 20261h 9m 57s
  More actions for workflow run 725

• 2026-10-24 nightly release (141f543d4c2bcdacc1f8a1b2e7238193de8158fa)

  completed successfully:
  Build Windows CUDA Wheels#724:Commit57de19dpushedbypytorchbot
  nightly

  Oct 24, 20261h 44m 3s
  More actions for workflow run 724

• Fetch Muse Glimmer serving tokenizer files from the non-GGUF repo

  completed successfully:
  Build Windows CUDA Wheels#723:Pull request#24137openedbyGasoonjia
  fix-muse-glimmer-serving-tokenizer

  Oct 24, 20261h 9m 27s
  More actions for workflow run 723

• Pin OpenVINO CI to the 2026.4.1 release instead of a deleted nightly

  completed successfully:
  Build Windows CUDA Wheels#722:Pull request#23502synchronizebyshoumikhin
  shoumikhin/openvino-pin-release

  Oct 24, 20261h 36m 45s
  More actions for workflow run 722

• CUDA backend: optional cuBLAS for matrix multiplications

  completed successfully:
  Build Windows CUDA Wheels#721:Pull request#24129synchronizebyshoumikhin
  shoumikhin:cuda-cublas-gemm

  Oct 24, 202650m 51s
  More actions for workflow run 721

• [pytorch hash update] update the pinned pytorch hash

  completed successfully:
  Build Windows CUDA Wheels#720:Pull request#10955synchronizebypytorchupdatebot
  update-pytorch-commit-hash/15080763801-537-1

  Oct 24, 20261h 11m 5s
  More actions for workflow run 720

• Let a backend share one delegate across calls of the same lowered module

  completed successfully:
  Build Windows CUDA Wheels#719:Pull request#24127synchronizebyshoumikhin
  shoumikhin:share-delegate-across-calls

  Oct 23, 202648m 50s
  More actions for workflow run 719

• Generate ring masks directly as packed bits (#24089)

  completed successfully:
  Build Windows CUDA Wheels#718:Pull request#24089synchronizebyThomasJannaud
  ThomasJannaud:export-D126734014

  Oct 23, 202649m 49s
  More actions for workflow run 718

• Let a backend share one delegate across calls of the same lowered module

  completed successfully:
  Build Windows CUDA Wheels#717:Pull request#24127synchronizebyshoumikhin
  shoumikhin:share-delegate-across-calls

  Oct 23, 202649m 40s
  More actions for workflow run 717

• CUDA wheel: check every dependency, and give torch_cpu_library targets torch's compile settings back

  completed successfully:
  Build Windows CUDA Wheels#716:Pull request#24087synchronizebyshoumikhin
  shoumikhin:cuda-wheel-dep-followups

  Oct 23, 202653m 35s
  More actions for workflow run 716

• Let a backend share one delegate across calls of the same lowered module

  cancelled:
  Build Windows CUDA Wheels#715:Pull request#24127synchronizebyshoumikhin
  shoumikhin:share-delegate-across-calls

  Oct 23, 20268m 51s
  More actions for workflow run 715

• Generate ring masks directly as packed bits (#24089)

  completed successfully:
  Build Windows CUDA Wheels#714:Pull request#24089synchronizebyThomasJannaud
  ThomasJannaud:export-D126734014

  Oct 23, 20261h 0m 30s
  More actions for workflow run 714

• Let a backend share one delegate across calls of the same lowered module

  cancelled:
  Build Windows CUDA Wheels#713:Pull request#24127synchronizebyshoumikhin
  shoumikhin:share-delegate-across-calls

  Oct 23, 202643m 17s
  More actions for workflow run 713

• CUDA wheel: check every dependency, and give torch_cpu_library targets torch's compile settings back

  cancelled:
  Build Windows CUDA Wheels#712:Pull request#24087synchronizebyshoumikhin
  shoumikhin:cuda-wheel-dep-followups

  Oct 23, 202657m 52s
  More actions for workflow run 712

• Generate ring masks directly as packed bits (#24089)

  cancelled:
  Build Windows CUDA Wheels#711:Pull request#24089synchronizebyThomasJannaud
  ThomasJannaud:export-D126734014

  Oct 23, 202630m 35s
  More actions for workflow run 711

• Let a backend share one delegate across calls of the same lowered module

  cancelled:
  Build Windows CUDA Wheels#710:Pull request#24127synchronizebyshoumikhin
  shoumikhin:share-delegate-across-calls

  Oct 23, 202629m 33s
  More actions for workflow run 710

• CUDA backend: optional cuBLAS for matrix multiplications

  completed successfully:
  Build Windows CUDA Wheels#709:Pull request#24129synchronizebyshoumikhin
  shoumikhin:cuda-cublas-gemm

  Oct 23, 202656m 27s
  More actions for workflow run 709

• CUDA backend: optional cuBLAS for matrix multiplications

  cancelled:
  Build Windows CUDA Wheels#708:Pull request#24129openedbyshoumikhin
  shoumikhin:cuda-cublas-gemm

  Oct 23, 20265m 22s
  More actions for workflow run 708

• CUDA wheel: check every dependency, and give torch_cpu_library targets torch's compile settings back

  completed successfully:
  Build Windows CUDA Wheels#707:Pull request#24087synchronizebyshoumikhin
  shoumikhin:cuda-wheel-dep-followups

  Oct 23, 202656m 55s
  More actions for workflow run 707

• Generate ring masks directly as packed bits (#24089)

  completed successfully:
  Build Windows CUDA Wheels#706:Pull request#24089synchronizebyThomasJannaud
  ThomasJannaud:export-D126734014

  Oct 23, 20261h 3m 56s
  More actions for workflow run 706

• Implement the portable kernel for avg_pool3d

  completed successfully:
  Build Windows CUDA Wheels#705:Pull request#24126openedbyyenhao-huang
  yenhao-huang:feat/portable-avg-pool3d-20261022

  Oct 23, 202655m 3s
  More actions for workflow run 705

• Generate ring masks directly as packed bits (#24089)

  cancelled:
  Build Windows CUDA Wheels#704:Pull request#24089synchronizebyThomasJannaud
  ThomasJannaud:export-D126734014

  Oct 23, 202624m 45s
  More actions for workflow run 704