[Common] Default EP comm kernels to 32 SMs; keep shuffle/preprocess on all SMs

[phu0ngng]

Description

Set a sane default SM budget for EP dispatch/combine (32 SMs) while leaving the local shuffle/preprocess kernels on all device SMs.

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

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

[phu0ngng]

/te-ci L1

[greptile-apps]

Greptile Summary

The PR changes the default NCCL EP dispatch/combine budget from automatic sizing to 32 SMs, clamped to the active device, while defaulting local shuffle and preprocessing kernels to all device SMs.

• Updates the native EP group configuration and process environment defaults.
• Updates the public C and JAX documentation for the new communication-kernel default.
• Leaves one native validation diagnostic using the old zero-means-auto terminology.

Confidence Score: 4/5

The PR appears safe to merge, with one non-blocking diagnostic inconsistency that should be corrected.

The implementation and public API documentation consistently apply the new 32-SM default, but invalid native configurations can still produce an error message claiming that zero enables automatic sizing.

Files Needing Attention: transformer_engine/common/ep/ep_backend.cpp

Important Files Changed

Filename	Overview
transformer_engine/common/ep/ep_backend.cpp	Implements the new SM defaults correctly at initialization, but leaves the validation message describing the former zero-value semantics.
transformer_engine/common/include/transformer_engine/ep.h	Updates the public native configuration documentation to describe the 32-SM clamped default.
transformer_engine/jax/ep.py	Updates the JAX API documentation consistently with the new dispatch/combine default.

_{Reviews (1): Last reviewed commit: "[Common] Default EP comm kernels to 32 S..." | Re-trigger Greptile}

[greptile-apps]

Stale zero-budget diagnostic

The implementation now maps num_comm_sms == 0 to min(32, device_sms), but the native validation message still describes zero as automatic sizing, misleading users who are diagnosing or selecting the communication-kernel SM budget.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!