Inference: Add the embedding and output layer in the full_iteration_inference cuda graph scope for hybrid models

[sidsingh-nvidia]

What does this PR do ?

⚠️ 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.

Issue tracking

For PRs from open-source community contributors:

• New features: a linked issue is required. Please open a feature request and reference it here before submitting the PR.
• Small updates (bug fixes, minor improvements): a linked issue is recommended and will accelerate the PR review process.

Linked issue:

Contribution process

Pre-checks

• 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

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!

All PRs start as draft. If you open a non-draft PR, it will be automatically converted to draft.

Step 1: Mark PR as "Ready for Review"

1. When your PR is ready, click Ready for Review.
2. An oncall reviewer is auto-assigned and expert reviewers are notified based on your changes.
   • Some PRs may jump straight to step 2. This is determined by .github/CODEOWNERS.

⚠️ Only mark as ready once merge-conflicts are resolved and the CI is passing.
Final Review might get declined if these requirements are not fulfilled.

Step 2: Final Review

For PRs that change megatron/core, once all expert reviewers have approved, the Final Review label is applied automatically and final reviewers are assigned.

For PRs outside megatron/core, this step is skipped.

Step 3: Approved

Once all required reviewers have approved, the Approved label is applied automatically.

Merge

Any member of mcore-engineers will be able to merge your PR.

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.

[github-actions]

This PR has been automatically converted to draft because all PRs must start as drafts.

When you are ready for review, click Ready for Review to begin the review process. This will:

1. Add the oncall reviewer (optional reviewer)
2. Add required review teams based on your changes

See the contribution guide for more details.

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[tdene]

Does this work when the CUDA graph scope is something other than full_iteration_inference?

If not, could you please leave a comment in the code mentioning this, so that it doesn't trip up future users.

[tdene]

This does not capture the .float() inside Float16Module.forward, nor the inference_context.last_token_logits and the transpose(0, 1).contiguous() inside the model's postprocess calls. I did not check the embedding layer.

Overall, these uncaptured calls add latency and unnecessary memcopies.

References:

• https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/core/transformer/module.py#L499
• https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/core/models/hybrid/hybrid_model.py#L480
• https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/core/models/hybrid/hybrid_model.py#L498

EDIT:
After discussion, the above is wrong. The hybrid_model.py calls are captured. What I see in my trace as uncaptured is just the Float16Module.forward and a different memcopy (like from abstract_inference_wrapper).

Approved; the other sites will be handled by a different PR.

[mathemakitten]

LGTM. Can you document the speedup?

[mathemakitten]

Are we sure that and kwargs['attention_mask'] is None is no longer load-bearing? I think it is now superfluous because of the check against inference_context but just double checking.

[sidsingh-nvidia]

attention_mask seems like a superfluous check. On this note, we need a standardized way to check if we are in inference_mode. Every module seems to be doing it's own thing atm :). Some check if torch grad is enabled, other's check for the inference context etc.

[sidsingh-nvidia]

@mathemakitten we gain around 200-300us with this.

[svcnvidia-nemo-ci]

🔄 Merge queue validation started!

You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/24908769581