Skip to content

[CICD] Add KunlunXin unit and MCore integration test support - #94

Open
AlexMa616 wants to merge 4 commits into
flagos-ai:mainfrom
BrianPei:Kunlun-dev
Open

[CICD] Add KunlunXin unit and MCore integration test support#94
AlexMa616 wants to merge 4 commits into
flagos-ai:mainfrom
BrianPei:Kunlun-dev

Conversation

@AlexMa616

Copy link
Copy Markdown
Collaborator

Description

Summary

This PR adds dedicated KunlunXin CI support for TransformerEngine-FL using the existing vendor.kunlunxin backend.

Changes

  • Added the KunlunXin platform configuration, environment setup script, and test workflow.
  • Added KunlunXin launchers for:
    • PyTorch debug tests
    • PyTorch unit tests
    • Distributed tests
    • ONNX export tests
    • MCore integration tests
  • Added an explicit setup check to ensure:
    • transformer_engine_klx_torch can be imported.
    • The KunlunXin backend is available.
    • generic_gemm selects vendor.kunlunxin instead of falling back to another backend.
  • Kept KunlunXin-specific environment handling and compatibility logic outside the common workflows.
  • Added generic MCore entrypoint and CUDA-default controls for platform-specific launchers.
  • Pinned the Megatron-LM-FL revision used by the MCore integration test.
  • Enabled coverage collection and aggregation.

No TransformerEngine core source, upstream test cases, other platform configurations, or third-party submodules are changed.

Testing

The previous self-hosted KunlunXin workflow run passed:

  • PyTorch debug tests
  • PyTorch unit tests
  • Distributed tests
  • ONNX export tests
  • Coverage aggregation
  • MCore integration tests

The final cleaned branch additionally passed shell syntax, YAML parsing, embedded Python syntax, configuration path, and diff consistency checks.

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.

2 participants