Commits · acf98b5c79d7e1db5c3efab2fa09fae4c642ab2e · OpenDAS / TransformerEngine

03 Nov, 2022 1 commit

Conditional wgrad support (#21) · 78210127

schetlur-nv authored Nov 03, 2022



* Conditional dgrad/wgrad support
Signed-off-by: Sharan Chetlur <schetlur@dlcluster.nvidia.com>

* Fixing the change to depend only on requires_grad. Also updating LayerNorm MLP
Signed-off-by: Sharan Chetlur <schetlur@dlcluster.nvidia.com>

* Minor fixes.
Signed-off-by: Sharan Chetlur <schetlur@dlcluster.nvidia.com>

* Adding conditional wgrad for LayerNormLinear
Signed-off-by: Sharan Chetlur <schetlur@dlcluster.nvidia.com>

* bug fix and remove conditional dgrad

Co-authored-by: schetlur-nv schetlur@nvidia.com
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Adding unit test for wgrad disabled path
Signed-off-by: Sharan Chetlur <schetlur@dlcluster.nvidia.com>

* Adding more unit tests for wgrad disabled path
Signed-off-by: Sharan Chetlur <schetlur@dlcluster.nvidia.com>

* Adding unit tests for fp8 wgrad disabling, and cleaning up the code.
Signed-off-by: Sharan Chetlur <schetlur@dlcluster.nvidia.com>

* fix lint errors
Co-Authored-By: Sharan Chetlur <schetlur@nvidia.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Signed-off-by: Sharan Chetlur <schetlur@dlcluster.nvidia.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: Sharan Chetlur <schetlur@dlcluster.nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

78210127

12 Oct, 2022 1 commit

Remove fp8_out from the LN API (#8) · 6a2161bf

Przemyslaw Tredak authored Oct 12, 2022



* Remove fp8_out from LN API
Signed-off-by: Przemyslaw Tredak <ptredak@nvidia.com>

* fix LN test
Signed-off-by: Przemyslaw Tredak <ptredak@nvidia.com>

* Fixes
Signed-off-by: Przemyslaw Tredak <ptredak@nvidia.com>
Signed-off-by: Przemyslaw Tredak <ptredak@nvidia.com>
Co-authored-by: ksivamani <ksivamani@nvidia.com>

6a2161bf

04 Oct, 2022 3 commits

Improve check in world size; None is valid group (#4) · a6b5f96d

Kirthi Shankar Sivamani authored Oct 04, 2022


Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

a6b5f96d

fix TP case to rm redundant AR (#3) · 5612ba78

Kirthi Shankar Sivamani authored Oct 04, 2022


Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

5612ba78

fp8_autocast bug fix when switching from non-fp8 execution (#2) · a2caec1e

Kirthi Shankar Sivamani authored Oct 04, 2022


Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

a2caec1e

28 Sep, 2022 1 commit

Inital code drop · 996ea169

Przemek Tredak authored Sep 27, 2022


Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

996ea169