[CI] Enable Windows ARM64 wheels - #691
Open
vortex-captain wants to merge 13 commits into
Open
Conversation
…sion - Fix Python library path detection in _build_optional_torch_c_dlpack.py to also check for versioned Python library (e.g. python312.lib) in addition to the major-only name (python3.lib). Some Python distributions like nuget-cpython (used by cibuildwheel on Windows) only provide the versioned library. When only the versioned library is found, explicitly add it to linker flags so pybind11's #pragma comment directive can resolve correctly. - Fix test_optional_torch_c_dlpack.py to use platform-appropriate library extension: .dll on Windows, .so elsewhere. Previously hardcoded .so caused test failures on Windows even when build succeeded.
…ilure Fix Windows wheel build: versioned Python lib detection and .dll extension
Contributor
|
Caution The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased. |
Member
|
good to understand usecases, likely we will align with pytorch on availability of win arm64 wheel |
vortex-captain
marked this pull request as draft
August 13, 2026 07:54
WoA CUDA support is to be added with Torch 2.14 release. There are nightlies in the nvidia index for now https://pypi.nvidia.com/nvtorch_oot_nightly/ |
Member
|
get it, likely we don't need torch c ext for windows arm64 since torch314 already comes with the functions we need, we can try enable tvm-ffi cibw wheel |
Keep the PR focused on core tvm-ffi wheels and leave the optional torch C extension workflow unchanged, following maintainer feedback. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Install PyTorch from NVIDIA's Windows ARM64 nightly index before cibuildwheel tests. Allow prereleases and bypass the upload-date cutoff because the nightly index does not expose upload timestamps. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
vortex-captain
marked this pull request as ready for review
September 9, 2026 01:56
Author
|
@tqchen @contentis comments addressed, please help review, thanks! |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
tvm-ffiwheel publishing matrixFollowing maintainer feedback, this PR enables and tests the core
tvm-ffiWindows ARM64 wheel without changing the separatetorch-c-dlpack-extwheel workflow.Validation
maintorch 2.15.0.dev20260904+cu134from NVIDIA’s nightly index on native Windows ARM64Follow-up
Replace the NVIDIA nightly override with the official PyTorch distribution source once stable Windows ARM64 wheels are available.