- 23 Nov, 2022 1 commit
-
-
Kirthi Shankar Sivamani authored
fix checkpoint loading bug for FAR Signed-off-by:
Kirthi Shankar Sivamani <ksivamani@nvidia.com> Signed-off-by:
Kirthi Shankar Sivamani <ksivamani@nvidia.com>
-
- 17 Nov, 2022 1 commit
-
-
Kirthi Shankar Sivamani authored
* 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>
-
- 16 Nov, 2022 1 commit
-
-
Kirthi Shankar Sivamani authored
* 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>
-
- 28 Sep, 2022 1 commit
-
-
Przemek Tredak authored
Co-authored-by:
Kirthi Shankar Sivamani <ksivamani@nvidia.com> Signed-off-by:
Przemek Tredak <ptredak@nvidia.com>
-