Add the YARN support for hybrid_model by guihong-nv · Pull Request #4244 · NVIDIA/Megatron-LM

guihong-nv · 2026-04-09T22:37:27Z

Add YaRN positional embedding support to MambaModel

Summary

megatron/core/models/mamba/mamba_model.py: Extended position_embedding_type to accept 'yarn'. Adds a YarnRotaryEmbedding in __init__ and calls it in forward (discarding the returned mscale), mirroring the existing RoPE path.
mamba_builders.py: Injects YaRN-specific config attributes (yarn_rotary_scaling_factor, yarn_original_max_position_embeddings, yarn_beta_fast, yarn_beta_slow, yarn_mscale, yarn_mscale_all_dim, yarn_correction_range_round_to_int) onto TransformerConfig before MambaModel is constructed, so pretrain_mamba.py can accept --position-embedding-type yarn.
megatron/training/arguments.py: Added 'yarn' to the allowed values for --position-embedding-type.
tests/unit_tests/models/test_mamba_model.py: Added TestMambaModelWithYarn with three tests — constructor (checks YarnRotaryEmbedding is created), forward pass, and inference (static context, prefill + decode steps).

Contribution process

Pre-checks

I have added relevant unit tests
I have added relevant functional tests
I have added proper typing to my code Typing guidelines
I have added relevant documentation
I have run the autoformatter.sh on my PR

Code review

Feel free to message or comment the @mcore-oncall to help accelerate your merge into main. The less complex your PR is, the faster it will be approved and merged!

All PRs start as draft. If you open a non-draft PR, it will be automatically converted to draft.

Step 1: Mark PR as "Ready for Review"

When your PR is ready, click Ready for Review.
An oncall reviewer is auto-assigned and expert reviewers are notified based on your changes.
- Some PRs may jump straight to step 2. This is determined by .github/CODEOWNERS.

⚠️ Only mark as ready once merge-conflicts are resolved and the CI is passing.
Final Review might get declined if these requirements are not fulfilled.

Step 2: Final Review

For PRs that change megatron/core, once all expert reviewers have approved, the Final Review label is applied automatically and final reviewers are assigned.

For PRs outside megatron/core, this step is skipped.

Step 3: Approved

Once all required reviewers have approved, the Approved label is applied automatically.

Merge

Any member of mcore-engineers will be able to merge your PR.

For MRs into `dev` branch

The proposed review process for `dev` branch is under active discussion.

MRs are mergable after one approval by either eharper@nvidia.com or zijiey@nvidia.com.

copy-pr-bot · 2026-04-09T22:37:31Z

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.

Phlip79 · 2026-04-09T22:40:23Z

/ok to test c7c9d79

guihong-nv · 2026-04-10T20:05:34Z

/ok to test 3ace27b

Phlip79 · 2026-04-10T21:23:52Z

/claude review

claude

LGTM

svcnvidia-nemo-ci · 2026-04-23T13:34:16Z

🔄 Merge queue validation started!

You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/24838266239

svcnvidia-nemo-ci · 2026-04-24T01:19:17Z

🔄 Merge queue validation started!

You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/24867274470

Phlip79 · 2026-04-24T02:58:21Z

/ok to test 5f895dd

svcnvidia-nemo-ci · 2026-04-24T04:51:13Z

🔄 Merge queue validation started!

You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/24872833192

svcnvidia-nemo-ci · 2026-04-24T05:03:50Z

🔄 Merge queue validation started!

You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/24873177343

Co-authored-by: Philip Petrakian <ppetrakian@nvidia.com>

Co-authored-by: Philip Petrakian <ppetrakian@nvidia.com> Signed-off-by: yhgalaxy <yhgalaxy@outlook.com>

HybridModel already builds YarnRotaryEmbedding from getattr(self.config, "yarn_*") (NVIDIA#4244), but those attributes were never declared on TransformerConfig and had no CLI plumbing, so --position-embedding-type yarn raised AttributeError and YaRN could not be configured from the command line. - transformer_config.py: declare the 7 yarn_* fields with safe defaults (yarn_rotary_scaling_factor default 1.0 is a no-op). - arguments.py: add matching --yarn-* args; dest names match the field names so core_transformer_config_from_args plumbs them through automatically. - test_hybrid_model.py: drive the yarn test config via the constructor and add tests for the dataclass fields and config->embedding plumbing. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Signed-off-by: Guihong Li <guihongl@nvidia.com>

Co-authored-by: Philip Petrakian <ppetrakian@nvidia.com> Signed-off-by: Jon Barker <jbarker@aws-cmh-slurm-1-vscode-02.cm.cluster>

Co-authored-by: Philip Petrakian <ppetrakian@nvidia.com>

svcnvidia-nemo-ci added this to the Core 0.16 milestone Apr 9, 2026

copy-pr-bot Bot temporarily deployed to test April 10, 2026 20:06 Inactive

guihong-nv marked this pull request as ready for review April 10, 2026 21:14

guihong-nv requested review from a team as code owners April 10, 2026 21:14

svcnvidia-nemo-ci requested a review from a team April 10, 2026 21:14

svcnvidia-nemo-ci added the complexity: low label Apr 10, 2026

claude Bot approved these changes Apr 10, 2026

View reviewed changes

jaredcasper approved these changes Apr 10, 2026

View reviewed changes

guihong-nv mentioned this pull request Apr 12, 2026

[model] feat: Add YARN support for mamba_model from MCORE NVIDIA-NeMo/Megatron-Bridge#3289

Closed

5 tasks

Phlip79 approved these changes Apr 13, 2026

View reviewed changes

janEbert reviewed Apr 13, 2026

View reviewed changes

Comment thread megatron/core/models/mamba/mamba_model.py Outdated

guihong-nv added 2 commits April 20, 2026 18:45

add the YARN support for mamba_model

6b1b1d1

Fix linting issues

213fe44

guihong-nv force-pushed the main branch from 3ace27b to 213fe44 Compare April 21, 2026 01:49

guihong-nv requested a review from a team as a code owner April 21, 2026 01:49

svcnvidia-nemo-ci added the Final Review PR is in the "final review" stage label Apr 21, 2026

guihong-nv added 2 commits April 20, 2026 18:55

fix linting issue

2397c1e

Fix formatting issues

19b23ac

copy-pr-bot Bot temporarily deployed to test April 21, 2026 02:03 Inactive

Phlip79 removed request for a team April 21, 2026 02:37

guihong-nv enabled auto-merge April 21, 2026 12:18

guihong-nv disabled auto-merge April 21, 2026 12:18

Merge branch 'main' into main

303b749

guihong-nv changed the title ~~Add the YARN support for mamba_model~~ Add the YARN support for hybrid_model Apr 22, 2026

Phlip79 enabled auto-merge April 23, 2026 04:55

Phlip79 added this pull request to the merge queue Apr 23, 2026

github-merge-queue Bot removed this pull request from the merge queue due to a conflict with the base branch Apr 23, 2026

Merge branch 'main' into yarn_in_hybrid

05f4718

guihong-nv enabled auto-merge April 23, 2026 12:33

copy-pr-bot Bot temporarily deployed to test April 23, 2026 12:34 Inactive

guihong-nv added this pull request to the merge queue Apr 23, 2026

github-merge-queue Bot removed this pull request from the merge queue due to no response for status checks Apr 23, 2026

Phlip79 added this pull request to the merge queue Apr 24, 2026

github-merge-queue Bot removed this pull request from the merge queue due to failed status checks Apr 24, 2026

Merge branch 'main' into yarn_in_hybrid

5f895dd

Phlip79 enabled auto-merge April 24, 2026 02:58

copy-pr-bot Bot temporarily deployed to test April 24, 2026 02:59 Inactive

Phlip79 added this pull request to the merge queue Apr 24, 2026

Merged via the queue into NVIDIA:main with commit 4d6cdd5 Apr 24, 2026
68 checks passed

yangbofun pushed a commit to xlm-research/Megatron-LM that referenced this pull request May 22, 2026

Add the YARN support for hybrid_model (NVIDIA#4244)

f9a2770

Co-authored-by: Philip Petrakian <ppetrakian@nvidia.com>

yhgalaxy pushed a commit to yhgalaxy/Megatron-LM that referenced this pull request Jun 17, 2026

Add the YARN support for hybrid_model (NVIDIA#4244)

65d1103

Co-authored-by: Philip Petrakian <ppetrakian@nvidia.com> Signed-off-by: yhgalaxy <yhgalaxy@outlook.com>

guihong-nv mentioned this pull request Jun 23, 2026

Add formal YaRN RoPE config fields and CLI plumbing for hybrid models #5460

Closed

6 tasks

This was referenced Jun 23, 2026

Add formal YaRN RoPE config fields and CLI plumbing for hybrid models #5461

Closed

Add formal YaRN RoPE config fields for hybrid models #5465

Draft

jon-barker pushed a commit to jon-barker/Megatron-LM that referenced this pull request Jul 10, 2026

Add the YARN support for hybrid_model (NVIDIA#4244)

2d08e11

Co-authored-by: Philip Petrakian <ppetrakian@nvidia.com> Signed-off-by: Jon Barker <jbarker@aws-cmh-slurm-1-vscode-02.cm.cluster>

terminator123 pushed a commit to 021ai/Megatron-LM that referenced this pull request Aug 3, 2026

Add the YARN support for hybrid_model (NVIDIA#4244)

d985bb7

Co-authored-by: Philip Petrakian <ppetrakian@nvidia.com>

Navigation Menu

Uh oh!

Add the YARN support for hybrid_model - #4244

Add the YARN support for hybrid_model#4244
Phlip79 merged 7 commits into
NVIDIA:mainfrom
guihong-nv:yarn_in_hybrid

guihong-nv commented Apr 9, 2026 •

edited

Loading

Uh oh!

copy-pr-bot Bot commented Apr 9, 2026

Uh oh!

Phlip79 commented Apr 9, 2026

Uh oh!

guihong-nv commented Apr 10, 2026

Uh oh!

Phlip79 commented Apr 10, 2026

Uh oh!

claude Bot left a comment

Uh oh!

Uh oh!

svcnvidia-nemo-ci commented Apr 23, 2026

Uh oh!

Uh oh!

svcnvidia-nemo-ci commented Apr 24, 2026

Uh oh!

Uh oh!

Phlip79 commented Apr 24, 2026

Uh oh!

svcnvidia-nemo-ci commented Apr 24, 2026

Uh oh!

svcnvidia-nemo-ci commented Apr 24, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

Uh oh!

Conversation

guihong-nv commented Apr 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Add YaRN positional embedding support to MambaModel

Summary

Contribution process

Pre-checks

Code review

Step 1: Mark PR as "Ready for Review"

Step 2: Final Review

Step 3: Approved

Merge

Uh oh!

copy-pr-bot Bot commented Apr 9, 2026

Uh oh!

Phlip79 commented Apr 9, 2026

Uh oh!

guihong-nv commented Apr 10, 2026

Uh oh!

Phlip79 commented Apr 10, 2026

Uh oh!

claude Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

svcnvidia-nemo-ci commented Apr 23, 2026

Uh oh!

Uh oh!

svcnvidia-nemo-ci commented Apr 24, 2026

Uh oh!

Uh oh!

Phlip79 commented Apr 24, 2026

Uh oh!

svcnvidia-nemo-ci commented Apr 24, 2026

Uh oh!

svcnvidia-nemo-ci commented Apr 24, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

guihong-nv commented Apr 9, 2026 •

edited

Loading