Add MimoOptimizer for heterogeneous parallelism with Mimo

[yashaswikarnati]

What does this PR do ?

Adds optimizer support for MIMO models with heterogeneous parallelism, where different modules (encoder, LLM) can have different DP/TP/PP configurations

Introduces MimoOptimizer class that:

• Manages per-module MegatronOptimizer instances
• Computes true global gradient norm via all_reduce MAX across all modules
• Clips gradients using the global norm
• Provides module-keyed state dicts for checkpointing

⚠️ For major changes (either in lines of code or in its impact), please make sure to first share a design doc with the team. If you're unsure what's the best way to do so, contact the @mcore-oncall.

Contribution process

flowchart LR
    A[Pre-checks] --> B[PR Tests]
    subgraph Code Review/Approval
        C1[Expert Review] --> C2[Final Review]
    end
    B --> C1
    C2 --> D[Merge]

Loading

Pre-checks

• I want this PR in a versioned release and have added the appropriate Milestone (e.g., Core 0.8)
• 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

The following process is enforced via the CODEOWNERS file for changes into megatron/core. For changes outside of megatron/core, it is up to the PR author whether or not to tag the Final Reviewer team.

For MRs into `main` branch

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!

(Step 1): Add PR label Expert Review

(Step 2): Collect the expert reviewers reviews

1. Attach the Expert Review label when your PR is ready for review.
2. GitHub auto-assigns expert reviewers based on your changes. They will get notified and pick up your PR soon.

⚠️ Only proceed to the next step once all reviewers have approved, merge-conflict are resolved and the CI is passing.
Final Review might get declined if these requirements are not fulfilled.

(Step 3): Final Review

1. Add Final Review label
2. GitHub auto-assigns final reviewers based on your changes. They will get notified and pick up your PR soon.

(Optional Step 4): Cherry-pick into release branch

If this PR also needs to be merged into core_r* release branches, after this PR has been merged, select Cherry-pick to open a new PR into the release branch.

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.

Merging your PR

Any member of core-adlr and core-nemo will be able to merge your PR.

[copy-pr-bot]

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.

[aroshanghias-nvd]

Quick note from trying to hook this into Megatron‑Bridge MIMO: there are a few small API mismatches that make the new MimoOptimizer hard to use without runtime shims in Bridge. To make the implementation work, core could expose these consistently:

• HyperCommGrid should have is_rank_in_grid() (Bridge grids already track rank_offset/size, and optimizer assumes this method exists).
• _get_pg_collection_from_grid() should set expt_dp=None so optimizer setup doesn’t fail when MoE isn’t used.
• get_mimo_optimizer() shouldn’t call .get() on modality_submodules if it’s a ModuleDict — use membership check + indexing.

If these land in core, Bridge can use MimoOptimizer directly without compatibility patches.

[yashaswikarnati]

Quick note from trying to hook this into Megatron‑Bridge MIMO: there are a few small API mismatches that make the new MimoOptimizer hard to use without runtime shims in Bridge. To make the implementation work, core could expose these consistently:

• HyperCommGrid should have is_rank_in_grid() (Bridge grids already track rank_offset/size, and optimizer assumes this method exists).
• _get_pg_collection_from_grid() should set expt_dp=None so optimizer setup doesn’t fail when MoE isn’t used.
• get_mimo_optimizer() shouldn’t call .get() on modality_submodules if it’s a ModuleDict — use membership check + indexing.

If these land in core, Bridge can use MimoOptimizer directly without compatibility patches.

thanks for raising these, i can try to address.

[dimapihtar]

/ok to test 214e4fd