Skip to content

[PyTorch] Fix EP zero-copy MXFP8 scale offset to be window-relative - #3447

Open
phu0ngng wants to merge 1 commit into
NVIDIA:mainfrom
phu0ngng:fix/ep_pyt
Open

[PyTorch] Fix EP zero-copy MXFP8 scale offset to be window-relative#3447
phu0ngng wants to merge 1 commit into
NVIDIA:mainfrom
phu0ngng:fix/ep_pyt

Conversation

@phu0ngng

Copy link
Copy Markdown
Collaborator

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

Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>
@phu0ngng
phu0ngng requested a review from ksivaman as a code owner August 31, 2026 18:19
@phu0ngng
phu0ngng requested a review from vthumbe1503 August 31, 2026 18:19
@phu0ngng

Copy link
Copy Markdown
Collaborator Author

/te-ci L1 pytorch.

@greptile-apps

greptile-apps Bot commented Aug 31, 2026

Copy link
Copy Markdown
Contributor

Greptile Summary

Updates zero-copy expert-parallel communication to calculate symmetric-memory tensor positions relative to the NCCL window.

  • Uses NCCLSymmetricMemory::get_window_offset() while retaining the tensor view’s byte storage offset.
  • Advances the pinned nccl-extensions submodule revision supporting the updated window-offset behavior.

Confidence Score: 5/5

The PR appears safe to merge because no concrete changed-code failure was established.

The offset calculation now accounts for the NCCL window origin while preserving the separate storage offset needed for tensor views, and no incompatible reachable contract was identified.

Important Files Changed

Filename Overview
transformer_engine/pytorch/csrc/extensions/ep.cpp Changes zero-copy NCCL communication-window offsets from allocation-relative to window-relative coordinates for base tensors and carved MXFP8 scale views.
3rdparty/nccl-extensions Advances the pinned NCCL extensions revision associated with the corrected communication-window offset contract.

Reviews (1): Last reviewed commit: "[PyTorch] Fix EP zero-copy MXFP8 scale o..." | Re-trigger Greptile

@vthumbe1503

Copy link
Copy Markdown
Collaborator

LGTM. Pending CI

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants