Skip to content

[CI] Publish GB200 aarch64 wheel artifacts - #3311

Open
bvolpato wants to merge 2 commits into
NVIDIA:mainfrom
bvolpato:bvolpato/publish-gb200-aarch64-wheels
Open

[CI] Publish GB200 aarch64 wheel artifacts#3311
bvolpato wants to merge 2 commits into
NVIDIA:mainfrom
bvolpato:bvolpato/publish-gb200-aarch64-wheels

Conversation

@bvolpato

@bvolpato bvolpato commented Aug 4, 2026 •
edited
Loading

Copy link
Copy Markdown

Description

Publish Linux aarch64 CUDA 13 core and PyTorch 2.11 extension wheel artifacts during releases. Preserve existing release defaults and limit scope to missing GB200 release tuples plus disk usage that blocked prior ARM jobs.

Addresses #3309
Addresses #3310

Type of change

  • Documentation change (change only to the documentation, either a fix or a new content)
  • Bug fix (non-breaking change which fixes an issue)
  • New feature (non-breaking change which adds functionality)
  • Breaking change (fix or feature that would cause existing functionality to not work as expected)
  • Infra/Build change
  • Code refactoring

Changes

  • Preserve existing PyTorch 2.8/CUDA 12.9 release matrix defaults and append exact PyTorch 2.11/CUDA 13 rows for x86_64 and aarch64.
  • Build and attach the CUDA 13 Linux aarch64 core wheel needed by the extension package.
  • Replace full CUDA toolkit meta-packages with compiler, development-library, and NVTX packages required by wheel builds.

Validation

  • Built the exact Python 3.12, PyTorch 2.11, CUDA 13 builder image.
  • Compiled the PyTorch extension and produced the expected cu13/torch2.11/cp312 wheel on x86_64.
  • Verified CUDA 13 ARM64 repositories provide the reduced package set, including NVTX.
  • Verified release matrix expands to existing x86_64/aarch64 2.8/12.9 rows plus new x86_64/aarch64 2.11/13 rows.
  • Ran pre-commit and Dockerfile static checks. Added workflow job has no actionlint findings; existing workflow warnings remain unchanged.

Checklist:

  • I have read and followed the contributing guidelines
  • The functionality is complete
  • I have commented my code, particularly in hard-to-understand areas
  • I have made corresponding changes to the documentation
  • My changes generate no new warnings
  • I have added tests that prove my fix is effective or that my feature works
  • New and existing unit tests pass locally with my changes

@bvolpato
[CI] Publish GB200 aarch64 wheel artifacts
17f4f0b 
Signed-off-by: bvolpato <brunocvcunha@gmail.com>
@github-actions github-actions Bot added the community-contribution PRs from external contributor outside the core maintainers, representing community-driven work. label Aug 4, 2026
@greptile-apps

greptile-apps Bot commented Aug 4, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Greptile Summary

The PR extends release wheel publishing to the missing Python 3.12, PyTorch 2.11, CUDA 13 tuples on x86_64 and ARM64.

  • Adds an ARM64 release job that builds and uploads the CUDA 13 core wheel.
  • Adds x86_64 and ARM64 PyTorch 2.11/CUDA 13 matrix entries.
  • Replaces full CUDA toolkit packages with the narrower compiler, development-library, and NVTX package set.
  • Pins the new checkout and release-upload actions to immutable commit SHAs.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains, and the previously reported mutable action references are pinned to full commit SHAs.

Important Files Changed

Filename Overview
.github/actions/build-pytorch-wheel/Dockerfile Uses a reduced CUDA package set and adds the PyTorch 2.11-to-CUDA 13 wheel mapping.
.github/workflows/attach-wheels-to-release.yml Adds CUDA 13/PyTorch 2.11 release tuples and an ARM64 core-wheel build-and-upload job; the previously reported action references are now pinned.
build_tools/wheel_utils/Dockerfile.aarch Adds the explicit NVTX package while removing the redundant full CUDA toolkit installation.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  Release["Release published"] --> Matrix["Generate wheel matrix"]
  Matrix --> X86["Build x86_64 PyTorch wheel"]
  Matrix --> ARMExt["Build ARM64 PyTorch wheel"]
  Release --> ARMCore["Build ARM64 CUDA 13 core wheel"]
  X86 --> Assets["Attach wheels to release"]
  ARMExt --> Assets
  ARMCore --> Assets
Loading

Reviews (2): Last reviewed commit: "[CI] Preserve existing release wheel def..." | Re-trigger Greptile

greptile-apps[bot]
greptile-apps Bot reviewed Aug 4, 2026
View reviewed changes
Comment thread .github/workflows/attach-wheels-to-release.yml Outdated
@bvolpato
[CI] Preserve existing release wheel defaults
cbfc834 
Signed-off-by: bvolpato <brunocvcunha@gmail.com>
@ptrendx
ptrendx requested a review from fheinecke August 4, 2026 16:16
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@greptile-apps greptile-apps[bot] greptile-apps[bot] left review comments
@fheinecke fheinecke Awaiting requested review from fheinecke

At least 1 approving review is required to merge this pull request.

Assignees

No one assigned

Labels

community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@bvolpato