Skip to content

[CICD] Add workflows to validate TE QA test cases#41

Merged
Darryl233 merged 11 commits into
flagos-ai:mainfrom
Darryl233:addworkflows260215
Mar 2, 2026
Merged

[CICD] Add workflows to validate TE QA test cases#41
Darryl233 merged 11 commits into
flagos-ai:mainfrom
Darryl233:addworkflows260215

Conversation

@Darryl233

@Darryl233 Darryl233 commented Feb 15, 2026

Copy link
Copy Markdown
Collaborator

Description

Validate TE QA test cases with new CI workflows

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:

  • Added code inspection and PyTorch/C++ unit tests to improve the TE testing system
  • Implemented end-to-end automation of TE wheel package building, installation, and verification, supporting multiple versions of Flash Attention and GPUs with different CUDA architectures
  • Verified TE's core functions (distributed communication, matrix multiplication, ONNX export) and compatibility with Megatron-LM/Lightning-Thunder
  • Completed the verification of the nvinspect debugging tool and re-verification of core numerical tests

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

@Darryl233 Darryl233 changed the title [wip] cicd workflow [CICD] Add workflows to validate TE QA test cases Feb 15, 2026
@lxd-cumt

Copy link
Copy Markdown
Collaborator

Does the current unit test in TE-FL CI cover the codes in the Plugin?

If the Plugin can be covered, it is recommended to test all three backends (FlagOS / Vendor / Reference) within the Plugin. Control this via TE-FL-Prefer.

Should we design it as three separate CI pipelines for three plugin backends, since we may extend to other vendors in the future?

@Darryl233

Darryl233 commented Feb 27, 2026

Copy link
Copy Markdown
Collaborator Author

@lxd-cumt
Yes, current tests have covered the Plugin.

Current workflow only tests the Vendor.CUDA backend.

We will improve the workflow to make sure tests can cover all three backends in next PR

@lxd-cumt lxd-cumt left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM for transformer_engine/plugin and transformer_engine/pytorch, mostly code style formatting.
Based on this PR, Qwen3 can be trained correctly on all three backends: FlagOS, CUDA, and Reference.

@Darryl233 Darryl233 merged commit f808816 into flagos-ai:main Mar 2, 2026
11 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants