perf: Update moe_token_dispatcher_type default to alltoall

[parthmannan]

What does this PR do ?

Add a one line overview of what this PR aims to accomplish.

Issues

List issues that this PR closes (syntax):

Usage

• You can potentially add a usage example below

# Add a code snippet demonstrating how to use this

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you run the unit tests and functional tests locally? Visit our Testing Guide for how to run tests
• Did you add or update any necessary documentation? Visit our Document Development Guide for how to write, build and test the docs.

Additional Information

• ...

Summary by CodeRabbit

• Chores
  • Updated MOE token dispatcher configuration from "allgather" to "alltoall" across multiple example configurations.
  • Updated corresponding test configurations and documentation to reflect the new dispatcher default.

[coderabbitai]

📝 Walkthrough

Walkthrough

The PR updates Megatron MoE token dispatcher configuration from "allgather" to "alltoall" across example configuration files, test files, and documentation, changing the distributed token routing strategy.

Changes

Cohort / File(s)	Summary
Example Configurations examples/configs/distillation_math.yaml, examples/configs/distillation_math_megatron.yaml, examples/configs/dpo.yaml, examples/configs/grpo_math_1B.yaml, examples/configs/grpo_math_1B_megatron.yaml, examples/configs/sft.yaml, examples/configs/sft_openmathinstruct2_megatron.yaml, examples/configs/vlm_grpo_3B.yaml, examples/configs/vlm_grpo_3B_megatron.yaml, examples/nemo_gym/grpo_workplace_assistant_nemotron_nano_v2_9b.yaml	Updated moe_token_dispatcher_type from "allgather" to "alltoall" in Megatron policy configurations across all example YAML files.
Documentation nemo_rl/models/policy/__init__.py	Updated documentation comment for MegatronConfig.moe_token_dispatcher_type to reflect new default value of "alltoall".
Test Configurations tests/unit/models/generation/test_vllm_generation.py, tests/unit/models/megatron/test_megatron_setup.py, tests/unit/models/policy/test_megatron_worker.py	Updated Megatron test configurations to expect moe_token_dispatcher_type value of "alltoall" instead of "allgather".

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~5 minutes

Possibly related PRs

• feat: refactor megatron init #1646: Adds and sets up Megatron MoE config handling infrastructure that this PR's dispatcher type changes depend on.
• perf: DeepEP interface in megatron backend #1794: Introduces the moe_token_dispatcher_type field to Megatron configuration; this PR changes its default value.

Suggested labels

Performance

Suggested reviewers

• terrykong

🚥 Pre-merge checks | ✅ 3 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Test Results For Major Changes	⚠️ Warning	PR changes moe_token_dispatcher_type default from 'allgather' to 'alltoall' across multiple configs, a significant distributed training change affecting performance and convergence, but PR description lacks test results or regression validation.	Add test results and performance metrics demonstrating no regressions from the dispatcher type change, including convergence validation and before/after performance numbers.

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and specifically describes the main change: updating the default value of moe_token_dispatcher_type from 'allgather' to 'alltoall' across multiple configuration files.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

• 📝 Generate docstrings (stacked PR)
• 📝 Generate docstrings (commit on current branch)

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

Tip

Issue Planner is now in beta. Read the docs and try it out! Share your feedback on Discord.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

nemo_rl/models/policy/__init__.py (1)

1-1: ⚠️ Potential issue | 🟡 Minor

Update the copyright year to 2026.

The file was modified but the header still shows 2025; update it to the current year.

🔧 Suggested fix

-# Copyright (c) 2025, NVIDIA CORPORATION.  All rights reserved.
+# Copyright (c) 2026, NVIDIA CORPORATION.  All rights reserved.

As per coding guidelines: Add the NVIDIA copyright header (with current year) to all Python files and shell scripts, excluding tests (files under tests/ or test-only scripts).

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@nemo_rl/models/policy/__init__.py` at line 1, Replace the outdated copyright
year in the top-of-file header comment from 2025 to 2026; locate the header
comment at the beginning of the module (the copyright comment line in
nemo_rl/models/policy/__init__.py) and update the year to "2026" so the NVIDIA
copyright header is current.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Outside diff comments:
In `@nemo_rl/models/policy/__init__.py`:
- Line 1: Replace the outdated copyright year in the top-of-file header comment
from 2025 to 2026; locate the header comment at the beginning of the module (the
copyright comment line in nemo_rl/models/policy/__init__.py) and update the year
to "2026" so the NVIDIA copyright header is current.

[guyueh1]

@terrykong can we merge this? It's a default config value change

[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.

[terrykong]

CI failed L1, but it's b/c it's on a fork.

https://github.com/NVIDIA-NeMo/RL/actions/runs/22741665979/job/65956222345?pr=2004

setting to docs CI since the other tests passed