Skip to content

Comments

[pull] main from NVIDIA:main#492

Merged
pull[bot] merged 1 commit intophu0ngng:mainfrom
NVIDIA:main
Feb 19, 2026
Merged

[pull] main from NVIDIA:main#492
pull[bot] merged 1 commit intophu0ngng:mainfrom
NVIDIA:main

Conversation

@pull
Copy link

@pull pull bot commented Feb 19, 2026

See Commits and Changes for more details.


Created by pull[bot] (v2.0.0-alpha.4)

Can you help keep this open source service alive? 💖 Please sponsor : )

…rt for additional bias shapes (#2537)

* Plumbing correct bias dims from TE to cudnn

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

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Make changes for cp bias code

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

* Add dbias and dbias_ to run_dpa_with_cp test

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

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Fix: Use output_dBias instead of input_dBias to extract the shape

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

* Add guards for bias/bias_/dbias/dbias_ being None

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

* Add support for bias shape 111s in addition to the original 1hss, 11ss, b1ss and bhss

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

* Add support for dbias calculation and variant packing for the dbias shapes b1ss, bhss, 11ss in addition to the already supported 1hss

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

* Add support for 111s bias shape in DPA

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

* Allow fused attn for dbias calculation for 11ss, b1ss, bhss. Disable fused attn if dbias calculation for 111s is required, else enable

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

* Disable requires_grad for bias for shape 111s in tests

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

* Disable bias grad / training flag for 111s bias in the non-CP attn tests. Add bias shape 111s to test_dpa_bias_shapes

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

* Fix to correctly create the bias shape tensor instead of the hard coded shape. Fix the comparison logic shapes for bias/dbias

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

* Add fused attn cp test cases for all supported bias shapes

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

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* nit: switch to elif for bias grad conditional

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

* Add CP support for bias/dbias shape 111s

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

* Add support for is_training in CP attn tests

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

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* nit: Fix incorrect comment

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

* nit: Fix incorrect comment and assert string

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

* Create the dbias graph tensor only if it is a cuDNN supported bias shape

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

* Fix the dim that is being compared for the two cp chunks in the test

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

* nit: Reinstate the original test for right side swa

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

---------

Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
@pull pull bot locked and limited conversation to collaborators Feb 19, 2026
@pull pull bot added the ⤵️ pull label Feb 19, 2026
@pull pull bot merged commit 2d0d276 into phu0ngng:main Feb 19, 2026
8 of 10 checks passed
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant