Skip to content

test(optimizer): MCORE-560 cover MoE gradient zero counts - #6050

Merged
ko3n1g merged 4 commits into
NVIDIA:mainfrom
svcnemo-autobot:fix/resume-metric-mismatch
Jul 29, 2026
Merged

test(optimizer): MCORE-560 cover MoE gradient zero counts#6050
ko3n1g merged 4 commits into
NVIDIA:mainfrom
svcnemo-autobot:fix/resume-metric-mismatch

Conversation

@svcnemo-autobot

@svcnemo-autobot svcnemo-autobot commented Jul 26, 2026

Copy link
Copy Markdown
Collaborator

Background

MCORE-560 followed the gradient-accounting correction in #5916. Its initial merge-queue head shifted deterministic loss and zero-count metrics before every affected golden was updated; the final merged change included those updates.

What changed

  • Extend the MoE gradient-statistics test to validate zero counts.
  • Exercise regular and distributed optimizers across TP/EP/ETP layouts.

Details

tests/unit_tests/training/test_param_norm.py asserts that an entirely zero logical model reports the reference model's full parameter count across topologies, then checks all-one gradients report zero alongside norm and clipping invariance.

Tested

  • git diff --check
  • Main run 30180965425: exact H100 and GB200 resume recipes passed five repeats
  • Fresh 4-H100 validation: ci-events-agent-write-4607a401 (pending)

Signed-off-by: svcnemo-autobot <svcnemo-autobot@nvidia.com>
@svcnvidia-nemo-ci
svcnvidia-nemo-ci marked this pull request as draft July 26, 2026 08:39
@github-actions

Copy link
Copy Markdown
Contributor

This PR has been automatically converted to draft because all PRs must start as drafts.

When you are ready for review, click Ready for Review to begin the review process. This will:

  1. Add the oncall reviewer (optional reviewer)
  2. Add required review teams based on your changes

See the contribution guide for more details.

@copy-pr-bot

copy-pr-bot Bot commented Jul 26, 2026

Copy link
Copy Markdown

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.

@svcnemo-autobot

Copy link
Copy Markdown
Collaborator Author

/ok to test b132d32

Signed-off-by: svcnemo-autobot <svcnemo-autobot@nvidia.com>
Signed-off-by: svcnemo-autobot <svcnemo-autobot@nvidia.com>
@svcnemo-autobot

Copy link
Copy Markdown
Collaborator Author

/ok to test a6445bd

Signed-off-by: svcnemo-autobot <svcnemo-autobot@nvidia.com>
@svcnemo-autobot

Copy link
Copy Markdown
Collaborator Author

/ok to test 3ba0d46

@svcnemo-autobot

Copy link
Copy Markdown
Collaborator Author

/ok to test 0333e00

@svcnemo-autobot

Copy link
Copy Markdown
Collaborator Author

/ok to test 0333e00

@ko3n1g ko3n1g added the core_r0.19.0 Auto-cherrypick to release branch. Apply before merge; cherrypick happens after merge. label Jul 29, 2026
@ko3n1g

ko3n1g commented Jul 29, 2026

Copy link
Copy Markdown
Contributor

@svcnemo-autobot fetch and merge upstream/main and set to auto-merge

@svcnvidia-nemo-ci

Copy link
Copy Markdown
Contributor

🛑 This session reached its turn budget — stopping to avoid runaway cost. Open a new request if there is more to do.

@ko3n1g
ko3n1g merged commit f43ff6b into NVIDIA:main Jul 29, 2026
183 of 185 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

complexity: low core_r0.19.0 Auto-cherrypick to release branch. Apply before merge; cherrypick happens after merge. Run functional tests

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants