Skip to content

[PyT] [Common] Enable Context Parallelism for GDN attention - #3826

Draft
KshitijLakhani wants to merge 33 commits into
NVIDIA:mainfrom
KshitijLakhani:klakhani/feat/linear_attn_with_cp
Draft

KshitijLakhani wants to merge 33 commits into
NVIDIA:mainfrom
KshitijLakhani:klakhani/feat/linear_attn_with_cp

Conversation

@KshitijLakhani

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: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
@KshitijLakhani KshitijLakhani self-assigned this Nov 17, 2026
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
@KshitijLakhani
KshitijLakhani force-pushed the klakhani/feat/linear_attn_with_cp branch from f791ea9 to 341f835 Compare November 18, 2026 19:16
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Replace the CP design page with a user-facing linear attention guide and runnable dense and packed GDN examples.

Include the example source in the notebook and validate it in unit and distributed CI, with independent FP64 reference coverage.

Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Run end-to-end coverage at the largest available CP size among 2, 4 and 8, and remove standalone partition, validation and binding tests.

Exercise four-document, eight-head cases and the 256-dimensional summary fallback. Preserve CP/TP subgroup coverage when running on eight GPUs.

Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants