[Feature]: CompressedTensors: NVFP4A16 not supported for MoE models

[zhangyimi]

🚀 The feature, motivation and pitch

NVFP4A16 (W4A16 FP4) quantization via compressed_tensors works for dense models but fails on MoE models like Qwen3-30B-A3B.

Looking at compressed_tensors_moe.py, _is_fp4a16_nvfp4 is checked for Linear layers but not in get_moe_method() for FusedMoE. Only W4A4 has a MoE method (CompressedTensorsW4A4Nvfp4MoEMethod).

Since the Marlin kernel already supports FP4 weights + FP16 activations, is there a plan to add W4A16 MoE support for compressed_tensors?

Alternatives

No response

Additional context

No response

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[robertgshaw2-redhat]

Would you be able to open a PR?

[mratsim]

Duplicate #21977

The fix seems simple but it might be just the tree that hides the forest: #21977 (comment)

[zhangyimi]

Would you be able to open a PR?

Hi @robertgshaw2-redhat ,
I've implemented W4A16 NVFP4 MoE support for CompressedTensors and submitted a PR: #32318
The implementation adds CompressedTensorsW4A16Nvfp4MoEMethod class using the Marlin backend. I've tested it with pe-nlp/Qwen3-30B-A3B-NVFP4A16 and it works as expected.
Would you mind helping review this PR? Thanks!

[loqs]

@dsikka fixed this in 5d3b609?

[dsikka]

Yup, this should be resolved!

[github-actions]

This issue has been automatically marked as stale because it has not had any activity within 90 days. It will be automatically closed if no further activity occurs within 30 days. Leave a comment if you feel this issue should remain open. Thank you!

[github-actions]

This issue has been automatically closed due to inactivity. Please feel free to reopen if you feel it is still relevant. Thank you!