GitLab

* Make amax reduction optional
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* remove setup for global amax redux for optional case
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

* Address documentation review
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

* better FP8 checkpointing
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Making checkpointing backwards compatible
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Add deprecation warning for old checkpoint loading
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* fix checkpointing for fp8 recompute case
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* improvements to deprecation warning
Co-authored-by: Przemyslaw Tredak <ptrendx@gmail.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: Przemyslaw Tredak <ptredak@nvidia.com>
Co-authored-by: Przemyslaw Tredak <ptrendx@gmail.com>

* Fix bugs for full activation recompute in FP8
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Ensure identical numerics in recomputation for pipeline parallelism
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* expose checkpoint API and add docs
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* complete checkpointing docs
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

addressed LayerNormMLP bias issue #26
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* 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>

* 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>

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

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

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

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