Skip to content

fix(packaging): AUT-1323 align dev package and TE revision - #6296

Closed
svcnemo-autobot wants to merge 4 commits into
NVIDIA:devfrom
svcnemo-autobot:fix/post-training-package-dev
Closed

fix(packaging): AUT-1323 align dev package and TE revision#6296
svcnemo-autobot wants to merge 4 commits into
NVIDIA:devfrom
svcnemo-autobot:fix/post-training-package-dev

Conversation

@svcnemo-autobot

@svcnemo-autobot svcnemo-autobot commented Aug 5, 2026

Copy link
Copy Markdown
Collaborator

Background

Bridge dev validation requires one consistent Transformer Engine revision across MCore and Bridge.

What changed

  • Retain the megatron.post_training packaging and lazy-import fixes.
  • Pin mutable recipe sources immutably.
  • Select TE cfda3834 and admit its 2.18.x version in the dev constraint.
  • Regenerate uv.lock.

Details

flowchart LR
 P[Dev constraint 2.18.x] --> T[TE cfda3834]
 T --> L[uv.lock 2.18.0]
 L --> B[Bridge dev validation]
Loading

Bridge PR #5326 will pin this exact head after its approval mirror exists.

Tested

  • uv lock; uv lock --check — 325 packages
  • Verified >=2.18.0,<2.19.0 admits 2.18.0+cfda3834
  • No f031cf87 remains in pyproject.toml or uv.lock
  • git diff --check
  • Signed commit with DCO

Add megatron.post_training and megatron.post_training.* to
pyproject.toml's [tool.setuptools.packages.find] include list.
Move modelopt imports to lazy (inside functions) so the module
is importable without modelopt installed.

Same fix as PR NVIDIA#6290 for the dev branch.

Signed-off-by: svcnemo-autobot <svcnemo-autobot@nvidia.com>
@svcnemo-autobot
svcnemo-autobot requested review from a team as code owners August 5, 2026 20:49
@copy-pr-bot

copy-pr-bot Bot commented Aug 5, 2026

Copy link
Copy Markdown

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

@svcnvidia-nemo-ci

Copy link
Copy Markdown
Contributor

/ok to test d1a503b

Pin fast-hadamard-transform to f134af63 and FlashMLA to
b7643bd in the DSV4 GB200 recipe YAML. Unpinned clones
of mutable branch HEADs can execute arbitrary future remote
code on CI runners.

Signed-off-by: svcnemo-autobot <svcnemo-autobot@nvidia.com>
Signed-off-by: svcnemo-autobot <svcnemo-autobot@nvidia.com>
Signed-off-by: svcnemo-autobot <svcnemo-autobot@nvidia.com>
@svcnemo-autobot svcnemo-autobot changed the title fix(packaging): AUT-1323 include megatron.post_training in package discovery (dev fix(packaging): AUT-1323 align dev package and TE revision Aug 6, 2026
@ko3n1g ko3n1g closed this Aug 6, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants