[Bug] use_single_group_runtime_offsets arg is not supported for ScaledSrelu

[vthumbe1503]

Description

Please include a brief summary of the changes, relevant motivation and context.

Fixes # (issue)

Type of change

• Documentation change (change only to the documentation, either a fix or a new content)
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Infra/Build change
• Code refactoring

Changes

Please list the changes introduced in this PR:

• Change A
• Change B

Checklist:

• I have read and followed the contributing guidelines
• The functionality is complete
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

[vthumbe1503]

/te-ci pytorch

[greptile-apps]

Greptile Summary

This PR prevents the unsupported single-group runtime-offset specialization from being selected for ScaledSReLU while preserving it for compatible activations.

• Makes the cuDNN frontend capability check activation-aware.
• Applies the capability consistently to forward metadata selection and backward kernel arguments.
• Updates the ScaledSwiGLU single-group MXFP8 test monkeypatch for the new helper signature.

Confidence Score: 5/5

The PR appears safe to merge, with no concrete correctness, security, or compatibility failures identified.

ScaledSReLU now consistently avoids the unsupported runtime-offset specialization and retains complete generic offset metadata for forward and backward, while compatible activation behavior remains unchanged.

Important Files Changed

Filename	Overview
transformer_engine/pytorch/ops/fused/grouped_mlp.py	Adds activation-aware runtime-offset capability gating and consistently routes ScaledSReLU through the existing generic-offset path.
tests/pytorch/test_grouped_mlp.py	Updates the single-group MXFP8 test to call and monkeypatch the capability helper with an activation type.

_{Reviews (1): Last reviewed commit: "fix cudnn version gurad" | Re-trigger Greptile}

[zhongbozhu]

LGTM

[zhongbozhu]

/te-ci pytorch

[vthumbe1503]

Blackwell tests related to grouped mlp are passing now in CI.