Skip to content

[CI] Publish PyTorch 2.11 CUDA 13 release wheels - #3311

Open
bvolpato wants to merge 3 commits into
NVIDIA:mainfrom
bvolpato:bvolpato/publish-gb200-aarch64-wheels
Open

[CI] Publish PyTorch 2.11 CUDA 13 release wheels#3311
bvolpato wants to merge 3 commits into
NVIDIA:mainfrom
bvolpato:bvolpato/publish-gb200-aarch64-wheels

Conversation

@bvolpato

@bvolpato bvolpato commented Aug 4, 2026

Copy link
Copy Markdown

Description

Publish Python 3.12, PyTorch 2.11, CUDA 13 framework-extension release assets for Linux x86_64 and aarch64. Source installs can fetch matching artifacts instead of compiling locally, while existing PyTorch 2.8/CUDA 12.9 release coverage remains unchanged.

Fixes #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

  • Add Python 3.12, PyTorch 2.11, CUDA 13, C++11 ABI release rows for x86_64 and aarch64.
  • Include SM100 in the build architecture list and replace the full CUDA toolkit with required compiler, development-library, and NVTX packages.
  • Stream complete compiler output and propagate build or timeout failures without hiding diagnostics.
  • Validate CPython, platform, and C++11 ABI tags before upload.
  • Install and import both new CUDA 13 wheel tuples in their native build containers using CUDA driver stubs.

Validation

  • Built the exact Python 3.12, PyTorch 2.11, CUDA 13, C++11 ABI wheel on x86_64 from the v2.18 release source.
  • Validated cp312-cp312-linux_x86_64, cu13, torch2.11.0+cu130, and cxx11abiTRUE artifact tags.
  • Installed the wheel with matching 2.18.0 metapackage/core dependencies and completed a driverless import transformer_engine.pytorch smoke check.
  • Confirmed CUDA 13 ARM64 repositories provide the reduced package set and PyTorch publishes the matching CPython 3.12 aarch64 wheel.
  • Verified release matrix expands to existing x86_64/aarch64 PyTorch 2.8 rows plus new x86_64/aarch64 PyTorch 2.11 rows.
  • Ran pre-commit across the repository, shell syntax checks, ShellCheck, YAML parsing, and actionlint. 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

@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

Copy link
Copy Markdown
Contributor

Greptile Summary

The PR expands release wheel publication to PyTorch 2.11/CUDA 13 on x86_64 and aarch64 while preserving the existing release matrix.

  • Replaces the full CUDA toolkit package with the compiler, development libraries, and NVTX packages needed for wheel builds.
  • Adds wheel tag, filename, architecture, and optional import-smoke-test validation.
  • Adds PyTorch 2.11/CUDA 13 release-matrix entries for x86_64 and ARM runners.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains; the previously reported mutable references belonged to the ARM core-wheel job that has been removed from the current workflow.

Important Files Changed

Filename Overview
.github/actions/build-pytorch-wheel/Dockerfile Adds compute capability 10.0 and PyTorch 2.11/CUDA 13 support while reducing installed CUDA packages.
.github/actions/build-pytorch-wheel/action.yml Passes architecture and release metadata into the container and preserves the Docker process exit status.
.github/actions/build-pytorch-wheel/build.sh Validates the generated wheel's count, platform tags, filename, and optional installation/import behavior.
.github/workflows/attach-wheels-to-release.yml Adds x86_64 and aarch64 PyTorch 2.11/CUDA 13 tuples and enables smoke testing for those release builds.

Sequence Diagram

sequenceDiagram
    participant Release as Release workflow
    participant Matrix as Release matrix
    participant Action as build-pytorch-wheel action
    participant Docker as Wheel builder container
    participant Assets as Release assets
    Release->>Matrix: Expand existing and CUDA 13/PyTorch 2.11 tuples
    Matrix->>Action: Pass architecture, versions, and smoke-test flag
    Action->>Docker: Build and validate extension wheel
    Docker-->>Action: Return validated wheel name
    Action->>Assets: Upload wheel artifact
Loading

Reviews (3): Last reviewed commit: "[CI] Validate CUDA 13 PyTorch release wh..." | Re-trigger Greptile

Comment thread .github/workflows/attach-wheels-to-release.yml Outdated
Signed-off-by: bvolpato <brunocvcunha@gmail.com>
Signed-off-by: bvolpato <brunocvcunha@gmail.com>
Signed-off-by: bvolpato <brunocvcunha@gmail.com>
@bvolpato
bvolpato force-pushed the bvolpato/publish-gb200-aarch64-wheels branch from cbfc834 to 2da90bf Compare August 16, 2026 02:47
@bvolpato bvolpato changed the title [CI] Publish GB200 aarch64 wheel artifacts [CI] Publish PyTorch 2.11 CUDA 13 release wheels Aug 16, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

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

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[Build] Publish PyTorch 2.11 CUDA 13 aarch64 wheels for GB200

2 participants