[bnellnm]

Purpose

Integration for DeepGEMM MegaMoE kernel.

Weight creation/prep in the quant method is not right. I'm not sure if I should create a whole new quant method or hack up the existing create_weights or add some further processing step, .e.g. cast_grouped_weights_to_fp4

TODO: consider making "deep_gemm_mega_moe" backend?

Test Plan

Added unit test
TODO: test on mxfp4 model.

Test Result

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.

BEFORE SUBMITTING, PLEASE READ https://docs.vllm.ai/en/latest/contributing (anything written below this line will be removed by GitHub Actions)

[gemini-code-assist]

Code Review

This pull request implements the DeepGemmMegaExperts modular kernel, integrating DeepGEMM's mega_moe kernel for MXFP4 MoE. The changes include the new expert class, comprehensive unit tests, and FP4 quantization utilities. Review feedback focuses on improving code quality by removing redundant math utilities in favor of existing vLLM helpers and cleaning up dead code, placeholder comments, and unimplemented methods in the new implementation.

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @bnellnm.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork