cp: Move TE cross entropy guard to training args (5162) into core_r0.18.0 - #5185
Merged
Conversation
Co-authored-by: Mike Chrzanowski <mchrzanowski@gcp-nrt-cs-001-login-001.cm.cluster> Signed-off-by: NeMo Bot <nemo-bot@nvidia.com>
…r0.18.0` Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com> (cherry picked from commit b574499) Signed-off-by: oliver könig <okoenig@nvidia.com>
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
Contributor
Author
|
/ok to test 3f9ed1e |
yaoyu-33
approved these changes
Jun 5, 2026
…162-core_r0.18.0 Resolve TE cross-entropy guard conflict: keep the soft warnings.warn in model_parallel_config.py (PR #5162); the hard reject now lives in validate_args. Tests updated to cover both. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Signed-off-by: oliver könig <okoenig@nvidia.com>
Contributor
Author
|
/ok to test 004603b |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Claude summary
Background / motivation
core_r0.18.0.ModelParallelConfigfrom a hardassertinto awarnings.warn, leavingvalidate_args(megatron/training/arguments.py) as the authoritative reject.assert— were introduced by Disable TE cross entropy loss fusion #5115, whichcore_r0.18.0does not yet have.Dependency — do not merge before #5144
Disable TE cross entropy loss fusion (5115)intocore_r0.18.0#5144 (cp: Disable TE cross entropy loss fusion (5115)), the backport of the prerequisite Disable TE cross entropy loss fusion #5115.cherry-pick-5115-core_r0.18.0; this PR auto-retargets tocore_r0.18.0once cp:Disable TE cross entropy loss fusion (5115)intocore_r0.18.0#5144 merges.test_te_cross_entropy_loss_fusion_is_disabled_by_training_argswould fail —core_r0.18.0'svalidate_argshas no"Transformer Engine cross entropy loss fusion"assert.What changed
megatron/core/model_parallel_config.py: TE cross-entropyassert→UserWarning.tests/unit_tests/test_model_parallel_config.py: warn-instead-of-raise test + DP-safevalidate_argsrejection test.Details
b57449928aafaaec7a36e8325d83aacf05e91bd4.Disable TE cross entropy loss fusion (5115)intocore_r0.18.0#5144 — final diff is exactly the upstream change (2 files, +39 / −7); no manual conflict markers.