[Main] Partial CUDA Graph support for EP Overlap

[Wohox]

Based on #1920

What does this PR do ?

EP Overlap brings extra cpu overhead and may cause GPU bubble during execution, partial CUDA graph helps release cpu pressure within the selected scope. This PR supports partial CUDA graph for EP Overlap, the supported scopes are attn, moe_router, moe_preprocess (moe and mlp are not supported).

Usage
To enable this feature, refer to the following example:

--overlap-moe-expert-parallel-comm \
--delay-wgrad-compute \
--cuda-graph-scope attn moe_router moe_preprocess \

(--delay-wgrad-compute is optional)

Correctness
The loss value can be bitwise aligned when enable & disable partial CUDA graph.

This PR is for main branch and the PR for dev branch is in PR2168

Dependencies
enable CUDA graph with delay-wgrad-compute relies on TE PR:

• [Merged] [PyTorch] Support delay_wgrad_compute cudagraph TransformerEngine#1948
• [InReview] [Pytorch] Fix backward_dw cuda graph order TransformerEngine#2376

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.

[kvareddy]

@fanshiqing @jiemingz can you please take a look at this MR?

[jiemingz]

it looks like this include changes from #1920 , is that supposed to be merged first and this be rebased?

[Wohox]

it looks like this include changes from #1920 , is that supposed to be merged first and this be rebased?

@jiemingz Yes, but rebase can happen later, since this MR requires #1920.

[Wohox]

/ok to test 125fa43

[lhb8125]

LGTM, thanks!

[jiemingz]

LGTM!

[Wohox]

/ok to test 9eababb

[Wohox]

@ericharper Can you help review this PR, still needs approval from NeMo and GPT group, thanks~

[Phlip79]

/ok to test ecbaff4

[Phlip79]

/ok to test c3219e6

[Phlip79]

/ok to test e01fcab

[Wohox]

/ok to test e180d4d