Fuse permute+pad and unpermute+unpad ops for FP8/FP4 training

[xiaoxi-wangfj]

What does this PR do ?

Integrate TransformerEngine fused MoE permute+pad and unpermute+unpad FP8 fast-path (NVIDIA/TransformerEngine#1921) into Megatron-LM MoE token dispatching, eliminating explicit padding/unpadding steps in the experts path to improve performance and reduce peak GPU memory usage.

Background / Motivation

TransformerEngine PR NVIDIA/TransformerEngine#1921 introduces:

• moe_permute_and_pad_with_probs (fused moe_permute_with_probs + Fp8Padding)
• extended moe_unpermute(..., pad_offsets=...) (fused moe_unpermute + Fp8Unpadding)

These TE-side fused kernels require Megatron-LM changes to:

• pass tokens_per_expert and FP8 alignment size into permute
• carry pad_offsets through dispatch and feed it back into unpermute
• gate the new path behind a configuration switch / availability checks

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

(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.

[xiaoxi-wangfj]

@yaox12 This PR integrates support for the fused permute+pad and unpermute+unpad APIs in Megatron-LM. Appreciate your review.

[xiaoxi-wangfj]

Hi @yaox12
The moe-permute-padding-for-quantization flag has been removed.
The system now defaults to fused_permute_and_pad_with_probs.
When this kernel is not available (e.g., due to TE version constraints),
it transparently falls back to fused_permute_with_probs + FP8 padding.

[yaox12]

Hi @yaox12 The moe-permute-padding-for-quantization flag has been removed. The system now defaults to fused_permute_and_pad_with_probs. When this kernel is not available (e.g., due to TE version constraints), it transparently falls back to fused_permute_with_probs + FP8 padding.

Thanks. I'll review it later.

[yaox12]

/ok to test 911548e

[yaox12]

/ok to test ea197bc