chore: nightly sync main into dev (05_05_2026) - #4619
Closed
svcnvidia-nemo-ci wants to merge 86 commits into
Closed
Conversation
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Co-authored-by: Xin Yao <xiny@nvidia.com>
Co-authored-by: john2 <john2@jrlogin01.jureca>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Co-authored-by: root <root@nvl72098-T17.cm.cluster> Co-authored-by: William Dykas <wdykas@oci-hsg-cs-001-vscode-03.cm.cluster> Co-authored-by: root <root@nvl72160-T13.cm.cluster>
…classmethod (#3812) Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: Dmytro Pykhtar <37850217+dimapihtar@users.noreply.github.com>
Signed-off-by: oliver könig <okoenig@nvidia.com>
#4403) Co-authored-by: Antoni-Joan Solergibert <asolergibert@nvidia.com>
Co-authored-by: Philip Petrakian <ppetrakian@nvidia.com>
Signed-off-by: Maanu Grover <maanug@nvidia.com> Co-authored-by: claude[bot] <209825114+claude[bot]@users.noreply.github.com>
Signed-off-by: dimapihtar <dpykhtar@nvidia.com>
Signed-off-by: dimapihtar <dpykhtar@nvidia.com>
Co-authored-by: Siddharth Singh <sidsingh@nvidia.com> Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…nference cuda graph scope for hybrid models (#4440)
…ss curve gaps for latent MoE models (#4433) Signed-off-by: root <jiemingz@nvidia.com>
…4158) Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…4422) Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Signed-off-by: rprenger <rprenger@nvidia.com>
Signed-off-by: qiyuw <qiyuw@nvidia.com> Co-authored-by: Antoni-Joan Solergibert <asolergibert@nvidia.com>
Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Co-authored-by: root <root@nvl72114-T04.cm.cluster> Co-authored-by: root <root@nvl72116-T13.cm.cluster> Co-authored-by: root <root@nvl72102-T01.cm.cluster> Co-authored-by: root <root@nvl72150-T11.cm.cluster> Co-authored-by: root <root@nvl72045-T09.cm.cluster> Co-authored-by: root <root@nvl72102-T04.cm.cluster>
Signed-off-by: Keshav Santhanam <ksanthanam@nvidia.com> Co-authored-by: Siddharth Singh <sidsingh@nvidia.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: Asha Anoosheh <aanoosheh@nvidia.com> Co-authored-by: oliver könig <okoenig@nvidia.com> Co-authored-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: oliver könig <okoenig@nvidia.com>
Signed-off-by: Maanu Grover <maanug@nvidia.com> Co-authored-by: oliver könig <okoenig@nvidia.com>
Signed-off-by: guihong-nv <guihongl@nvidia.com>
Contributor
Author
|
/ok to test b187f43 |
Contributor
Author
|
/ok to test e879a55 |
svcnvidia-nemo-ci
force-pushed
the
main2dev/05_05_2026
branch
from
May 5, 2026 04:17
e879a55 to
4de6661
Compare
Contributor
Author
|
/ok to test 4de6661 |
svcnvidia-nemo-ci
force-pushed
the
main2dev/05_05_2026
branch
from
May 5, 2026 05:30
4de6661 to
3567666
Compare
Contributor
Author
|
/ok to test 3567666 |
- Restore moe_expert_rank_capacity_factor field on TransformerConfig (used at line 1534 and 1801 but field definition was lost in merge) - Restore moe_mlp_glu_interleave_size field used by megatron/core/transformer/ moe/experts.py - Restore moe_hybridep_num_sms_preprocessing field - Update InferenceTopKRouter.forward signature to accept input_ids and forward it to TopKRouter.forward, matching the new caller signature in moe_layer.py:446 (apply_module(self.router)(hidden_states, padding_mask, input_ids)). Without this, tests/unit_tests/inference/test_hybrid_moe.py:: TestDynamicInferenceNVLS fails with TypeError: forward() takes from 2 to 3 positional arguments but 4 were given. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phlip79
force-pushed
the
main2dev/05_05_2026
branch
from
May 5, 2026 16:59
3567666 to
9a0cbdd
Compare
Member
|
/ok to test 9a0cbdd |
Member
|
/ok to test 9a0cbdd |
Contributor
Author
|
Superseded by today's nightly sync. |
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
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.
Summary
Nightly automated sync of
mainintodev(2026-05-05).mainFiles taken from
main(semantic conflicts)These files have known semantic conflicts where dev's versions reference args
or APIs main has removed/renamed. Taken from main wholesale:
megatron/training/training.pymegatron/training/initialize.pymegatron/training/utils.pymegatron/training/datasets/data_samplers.pyFiles restored from
mainmegatron/core/pipeline_parallel/hybrid_cp_schedule.py— providesBalancedCPScheduler, required by main'sHybridCPDataLoaderWrapper.Dev had renamed it to
dynamic_cp_schedule.pyand then deleted thatfile in PR [Dev] feat: Dynamic CP (part 2) #2000 ("Dynamic CP part 2") in favor of a new in-place
scheduler. Restored so main's training code continues to import.
Special handling
megatron/core/datasets/data_schedule.py: kept dev's classes(
BasePackingScheduler,DpBalancedScheduler,DefaultDynamicCPScheduler,wrap_data_iterator,get_batch_on_this_rank_for_sequence_packing) andappended main's
HybridCPDataLoaderWrapper. Imports unioned (Any,List,BalancedCPScheduleradded).megatron/training/training.py: changedif args.hybrid_context_parallel:toif getattr(args, 'hybrid_context_parallel', False):since the argno longer exists in dev's
arguments.py(replaced by dev's new dynamicCP system). The
HybridCPDataLoaderWrapperimport path remains valid.Files restored to dev's version
.github/CODEOWNERS— sync bot must never modify CODEOWNERS.pyproject.toml— dev includesfast-hadamard-transformgit source andnewer
nvidia-resiliency-extrevision.uv.lock— paired with dev'spyproject.toml.docker/Dockerfile.ci.dev— uses--group no_pypi_wheelsflag matchingdev's
pyproject.toml.Conflict resolution summary
-X theirs(favor main) was used as the merge strategy. The fullgit show --remerge-diffoutput is attached below for review.Files that had genuine merge conflicts (25):
.github/workflows/cicd-main.ymldocker/Dockerfile.ci.dev(restored to dev)docs/conf.pyexamples/mimo/train.pymegatron/core/datasets/readme.mdmegatron/core/transformer/moe/fused_a2a.pymegatron/core/transformer/moe/moe_layer.pymegatron/core/transformer/moe/token_dispatcher.pymegatron/core/transformer/transformer_config.pymegatron/core/transformer/transformer_layer.pymegatron/training/activation_logging.pymegatron/training/arguments.pymegatron/training/checkpointing.pymegatron/training/training.py(taken from main, then patched)megatron/training/utils.py(taken from main)pretrain_bert.pypretrain_gpt.pypretrain_mamba.pypretrain_vlm.pypyproject.toml(restored to dev)tests/unit_tests/models/test_hybrid_moe_model.pytools/checkpoint/checkpoint_inspector.pytools/checkpoint/remap_gpt_dsa_to_mamba.pytrain_rl.pyuv.lock(restored to dev)Pre-merge formatting
Ran
black==24.4.2(with--skip-magic-trailing-comma --skip-string-normalization)and
isort==5.13.2on the full set of changed files.pylint==3.2.6andruff~=0.9produce zero warnings on the autoformat target set.🤖 Generated by automated nightly sync bot.