Skip to content

Conversation

@sudhu2k
Copy link
Contributor

@sudhu2k sudhu2k commented Feb 11, 2026

Description

This PR fixes some bugs in the release v2.2 branch that I found when working on the Megatron IFU r0.15.0.

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

  1. Fixes assertion of Float8Tensor to include Float8Currentscaling tensor.
  2. cherry picks a commit which checks if the function is of type module, before accessing fsdp_wrapped, fsdp_group. (It was trying to access them even when it isn't of type TEModule.

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

sugovind and others added 2 commits January 27, 2026 04:13
* fix

Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* added test

Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* test change

Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* changed the test

Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
@ipanfilo
Copy link
Collaborator

I think it is better to create separate branch rather than using release_v2.2_rocm.
Is there corresponding PR for dev branch?
Also, please update copyright on modified files

@sudhu2k
Copy link
Contributor Author

sudhu2k commented Feb 11, 2026

Dev branch doesn't need these fixes, since they already have them

if isinstance(tensor, QuantizedTensor):

if isinstance(function, TransformerEngineBaseModule):
# If this TE module is FSDP-wrapped, clear its FSDP group information because there's no need
# to scatter/gather activations that we will recompute anyway.
setattr(function, "fsdp_wrapped", False)
setattr(function, "fsdp_group", None)

This is only for release_v2.2 branch since Megatron's CI uses this specific commit.
Let me know if you still think we should have a separate branch. I've updated copyright on the files which I made changes to, not the cherry-picks.

@ipanfilo
Copy link
Collaborator

It is not a question of branch per se but rather versioning. TE 2.2 was released a while ago. So changes on top of that should be versioned like 2.2.post0 or something like this.

@sudhu2k
Copy link
Contributor Author

sudhu2k commented Feb 12, 2026

Yeah that makes sense! I'll bump the version to 2.2.post0.

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