[doc, worker] feat: Enable Megatron-Bridge for MTP

[HollowMan6]

What does this PR do?

There's nothing specific in Megatron-Bridge that stops MTP support.

NVIDIA-NeMo/Megatron-Bridge#2387 adds MiMo dense MTP models bridge support so that examples/mtp_trainer/test_dapo_mimo_7b_with_mtp_math_megatron.sh can also be used together with Megatron-Bridge (setting vanilla_mbridge to be False).

Checklist Before Starting

• Search for similar PRs. Paste at least one query link here: ...
• Format the PR title as [{modules}] {type}: {description} (This will be checked by the CI)
  • {modules} include fsdp, megatron, veomni, sglang, vllm, rollout, trainer, ci, training_utils, recipe, hardware, deployment, ray, worker, single_controller, misc, perf, model, algo, env, tool, ckpt, doc, data, cfg, reward
  • If this PR involves multiple modules, separate them with , like [megatron, fsdp, doc]
  • {type} is in feat, fix, refactor, chore, test
  • If this PR breaks any API (CLI arguments, config, function signature, etc.), add [BREAKING] to the beginning of the title.
  • Example: [BREAKING][fsdp, megatron] feat: dynamic batching

Test

For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc.

API and Usage Example

Demonstrate how the API changes if any, and provide usage example(s) if possible.

# Add code snippet or script demonstrating how to use this

Design & Code Changes

Demonstrate the high-level design if this PR is complex, and list the specific changes.

Checklist Before Submitting

Important

Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review.

• Read the Contribute Guide.
• Apply pre-commit checks: pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always
• Add / Update the documentation.
• Add unit or end-to-end test(s) to the CI workflow to cover all the code. If not feasible, explain why: ...
• Once your PR is ready for CI, send a message in the ci-request channel in the verl Slack workspace. (If not accessible, please try the Feishu group (飞书群).)
• If your PR is related to the recipe submodule, please also update the reference to the submodule commit via git submodule update --remote or cd recipe && git pull origin main.

[gemini-code-assist]

Code Review

This pull request enables Multi-Token Prediction (MTP) support for Megatron-Bridge by removing an assertion that previously restricted MTP to vanilla_mbridge. The documentation is also updated to reflect this new capability and provide guidance on dependencies.

My main feedback is to consider adding a more specific check to ensure that MTP with Megatron-Bridge is only used with supported models, as the current implementation might be too permissive and could lead to runtime errors for users. This would improve the robustness of the new feature.

[Copilot]

Pull request overview

This PR enables Megatron-Bridge support for Multi-Token Prediction (MTP) training by removing an overly restrictive assertion that previously required vanilla_mbridge=True. According to the linked PR description, Megatron-Bridge now supports MTP models, so the restriction is no longer necessary.

Changes:

• Removed the vanilla_mbridge assertion for MTP in verl/workers/megatron_workers.py
• Updated documentation to reflect support for both mbridge and Megatron-Bridge with MTP
• Added Megatron-Bridge PR reference to the documentation

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
verl/workers/megatron_workers.py	Removed assertion requiring vanilla_mbridge=True for MTP, enabling Megatron-Bridge support
docs/advance/mtp.md	Updated training engine documentation to include Megatron-Bridge, added PR reference, and updated last modified date

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated no new comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[ArronHZG]

For the use of Megatron-Bridge where SFT supports MTP, can this feature be directly supported here by adapting to the model engine approach?

[HollowMan6]

For the use of Megatron-Bridge where SFT supports MTP, can this feature be directly supported here by adapting to the model engine approach?

Unfortunately I haven't tried using SFT with verl, but I don't think we call any API/methods in Megatron-Bridge's SFT for implementing Verl's SFT.

[ArronHZG]

Has used Mbridge for regression test codes? Currently, some users are reporting that when the model has MTP parameters but label = None (log_prob is recalculated), the original postprocess performs a label.clone operation, causing an error.

[HollowMan6]

I did test with mbridge as well (with the same mcore version as Megatron-Bridge though). This sounds like a mcore version mismatch, I was using https://github.com/NVIDIA/Megatron-LM/tree/8f1c2f8ae53b4e3f32c0ae7f397d8b38a675eaa2 because of the Megatron-Bridge requirements, though I didn't test on older mcore versions.