ci: Build and attach bdist wheels to release page

[ko3n1g]

Description

This PR adds CI workflows to build and attach PyTorch wheels to a release page of TE. Thus, the workflows are triggered on release creation. Manual runs are also possible via dispatch event.

By default a hardcoded torch and latest 5 NGC PyT versions are chosen.

it also fixed an issue where the existing build workflows ran out of storage.

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

Please list the changes introduced in this PR:

• Change A
• Change B

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

[ko3n1g]

Thanks for your review @timmoon10 . The suggestions all make sense to me. I'm still syncing with @ksivaman on some details and hope we can hash it out end of the week

[greptile-apps]

Greptile Overview

Greptile Summary

This PR introduces automated CI infrastructure to build and distribute PyTorch wheels for TransformerEngine releases. The changes implement a new GitHub Actions workflow that can be triggered on release creation or manually via dispatch events. The solution builds wheels against multiple PyTorch/CUDA/Python version combinations using both standard Docker containers and NGC (NVIDIA GPU Cloud) images to cover the last 5 months of available versions.

The implementation includes a new composite GitHub Action (build-pytorch-wheel) that handles containerized wheel building with disk space optimization, a shell script to dynamically discover available NGC images, and updates to the existing build workflow to address storage limitations. The PyTorch setup script was modified to support NGC container environments and drop CUDA 11 support in favor of CUDA 12/13 only. Additionally, the changes include standard Python development practices like adding .venv to .gitignore.

Important Files Changed

Filename	Score	Overview
.github/workflows/attach-wheels-to-release.yml	1/5	New workflow for automated wheel building with critical issues including undefined job references, deprecated actions, and logical errors
.github/actions/build-pytorch-wheel/action.yml	1/5	New composite action for wheel building with exit code masking, missing outputs, and unreachable cleanup code
.github/actions/build-pytorch-wheel/build.sh	1/5	Build script that captures but ignores exit codes, causing failed builds to appear successful
.github/scripts/check_for_ngc_images.sh	2/5	NGC image discovery script with macOS compatibility issues and error handling problems
.github/workflows/build.yml	2/5	Refactored build workflow addressing disk space issues but introducing manual container management complexity
.github/actions/build-pytorch-wheel/Dockerfile	3/5	Docker container setup with complex version mapping logic that may be fragile
transformer_engine/pytorch/setup.py	4/5	Updated wheel URL construction with improved NGC support and cleaner code structure
.gitignore	5/5	Standard addition of .venv to exclude Python virtual environments

Confidence score: 1/5

• This PR requires extensive review and testing before merging due to multiple critical implementation issues that will prevent proper functionality
• Score reflects serious problems with error handling, deprecated action usage, undefined references, and logical inconsistencies that would cause CI failures
• Pay close attention to the workflow files, composite action, and build scripts which contain fundamental flaws that need correction

[greptile-apps]

_{4 files reviewed, 6 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{4 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{4 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{4 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{4 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{4 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{4 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{6 files reviewed, 5 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{6 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{6 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{6 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{6 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{5 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{6 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{6 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{6 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

Additional Comments (1)

1. .github/workflows/attach-wheels-to-release.yml, line 152-170 (link)

   logic: Incomplete job - builds wheels for NGC images but never uploads them to the release. This job will build wheels and then discard them. Should add upload step similar to lines 141-150.

   Should this job also upload the NGC-built wheels to the release, or are they intended for a different purpose?

_{7 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{7 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{7 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{7 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{7 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{7 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{7 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{7 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{8 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile
React with 👍 or 👎 to share your feedback on this new summary format}

[greptile-apps]

_{8 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

logic: Container name 'builder' could conflict if multiple jobs run simultaneously - consider using unique names per job

Suggested change

	- name: Start named container
	run: |
	docker run -v $(pwd):$(pwd) -w $(pwd) --name builder -d nvcr.io/nvidia/cuda:12.8.0-devel-ubuntu22.04 sleep infinity
	- name: Start named container
	run: |
	docker run -v $(pwd):$(pwd) -w $(pwd) --name builder-pytorch -d nvcr.io/nvidia/cuda:12.8.0-devel-ubuntu22.04 sleep infinity

Are these jobs guaranteed to run sequentially, or could they run in parallel causing container name conflicts?

[greptile-apps]

logic: Same container name 'builder' used in multiple jobs - will cause conflicts if jobs run in parallel

Suggested change

	- name: Start named container
	run: |
	docker run -v $(pwd):$(pwd) -w $(pwd) --name builder -d ghcr.io/nvidia/jax:jax sleep infinity
	- name: Start named container
	run: |
	docker run -v $(pwd):$(pwd) -w $(pwd) --name builder-all -d ghcr.io/nvidia/jax:jax sleep infinity

[greptile-apps]

_{8 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[ksivaman]

/te-ci pytorch

[ksivaman]

Most review was conducted offline. For release v2.10 we will release as is. For release 2.11, we plan some generalizations of the pytorch wheel.

[greptile-apps]

_{8 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

_{8 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}