[sync] Update FLOPs calculation for hybrid MoE and latent MoE

[ananthsub]

What does this PR do ?

Sync with changes from:

• Update flops calculation (for throughput) for hybrid MoEs NVIDIA/Megatron-LM#2198
• Changes to support latent MoEs NVIDIA/Megatron-LM#2296

Changelog

• Update FLOPs calculation for hybrid MoE and latent MoE

GitHub Actions CI

See the CI sectionin the Contributing doc for how to trigger the CI. A Nvidia developer will need to approve and trigger the CI for external contributors.

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 add or update any necessary documentation?
• Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

If you haven't finished some of the above items you can still open "Draft" PR.

Additional Information

• Related to # (issue)

Summary by CodeRabbit

• New Features

  • Enabled floating-point operations (FLOPs) calculation support for Mixture-of-Experts models and hybrid model configurations.

• Tests

  • Added comprehensive test suite covering MoE FLOP calculations, including latent compression and activation function scaling scenarios.

_{✏️ Tip: You can customize this high-level summary in your review settings.}

[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.

[ananthsub]

/ok to test c09ba77

[liding-nv]

lgtm