- 06 Jun, 2024 1 commit
-
-
Kirthi Shankar Sivamani authored
Cleanup Signed-off-by:Kirthi Shankar Sivamani <ksivamani@nvidia.com>
-
- 03 Jun, 2024 1 commit
-
-
Tim Moon authored
* Modify CUDA graph tests to use grad accumulation steps Signed-off-by:
Tim Moon <tmoon@nvidia.com> * Initialize grad buffers before capturing CUDA graph in CUDA graph tests Signed-off-by:
Tim Moon <4406448+timmoon10@users.noreply.github.com> --------- Signed-off-by:
-
- 31 May, 2024 1 commit
-
-
Tim Moon authored
Replace int8_t in PyTorch extensions with int64_t Signed-off-by:
-
- 30 May, 2024 3 commits
-
-
Charlene Yang authored
* add THD support Signed-off-by:
Charlene Yang <8636796+cyanguwa@users.noreply.github.com> * add seq_offsets_o and use new offset calculation Signed-off-by:
-
Xin Yao authored
* add multi-tensor kernels Signed-off-by:
Xin Yao <xiny@nvidia.com> * add FusedAdam Signed-off-by:
-
Tim Moon authored
* Initial refactor of FP8 workspaces in Linear module Signed-off-by:
-
- 29 May, 2024 1 commit
-
-
Tim Moon authored
Make sure RoPE frequencies are in FP32 Signed-off-by:
-
- 25 May, 2024 1 commit
-
-
Paweł Gadziński authored
* Fixed Llama tutorial. Changed batch size and added fused=True. Signed-off-by:
Pawel Gadzinski <pgadzinski@nvidia.com> Signed-off-by:
root <root@ipp2-0037.nvidia.com> * Tutorial updated but not complete yet. Signed-off-by:
root <root@ipp2-1661.nvidia.com> * Comments Signed-off-by:
root <root@ipp2-2373.nvidia.com> Co-authored-by:
root <root@ipp2-1588.nvidia.com> Co-authored-by:
root <root@ipp2-2371.nvidia.com> Co-authored-by:
root <root@ipp2-1589.nvidia.com> Co-authored-by:
Sudhakar Singh <sudhakars@nvidia.com> Co-authored-by:
-
- 22 May, 2024 1 commit
-
-
Alp Dener authored
TE checkpoint now preserves the torch autocast context from the forward pass during the recompute phase Signed-off-by:Alp Dener <adener@nvidia.com>
-
- 21 May, 2024 2 commits
-
-
Alp Dener authored
replaced deprecated pkg_resources with packaging Signed-off-by: -
Pavel Shamis (Pasha) authored
-
- 20 May, 2024 1 commit
-
-
Paweł Gadziński authored
* Calibration fix Signed-off-by:
-
- 17 May, 2024 1 commit
-
-
Charlene Yang authored
* fix inconsistency for attn mask; now True means participating in attn Signed-off-by:
-
- 16 May, 2024 1 commit
-
-
Phuong Nguyen authored
* added squared relu in te-torch Signed-off-by:
Phuong Nguyen <phuonguyen@nvidia.com> --------- Signed-off-by:
-
- 13 May, 2024 1 commit
-
-
Kunlun Li authored
Signed-off-by:
kunlunl <kunlunl@nvidia.com> Co-authored-by:
-
- 09 May, 2024 1 commit
-
-
Kirthi Shankar Sivamani authored
Bump FA version to 2.5.8 Signed-off-by:
-
- 02 May, 2024 1 commit
-
-
cyanguwa authored
* initialize tp_group for FP8 DPA Signed-off-by:
-
- 01 May, 2024 1 commit
-
-
Jinze Xue authored
* Handle the scaling factor when amax is too tiny that leads to an infinite scale Signed-off-by:
Jinze Xue <jinzex@nvidia.com> * revert formatting changes Signed-off-by:
Jinze Xue <155670984+jinzex@users.noreply.github.com> * Apply review suggestion Co-authored-by:
-
- 30 Apr, 2024 2 commits
-
-
vasunvidia authored
Signed-off-by:
Vasudevan Rengasamy <vrengasamy@nvidia.com> Co-authored-by:
-
Tim Moon authored
* Fix linter warnings from unused args Signed-off-by:
-
- 29 Apr, 2024 2 commits
-
-
cyanguwa authored
remove tp_size/tp_group as amax reduction is handled by fp8_group() Signed-off-by: -
Zhenhuan Liu authored
* Add support for MoE with FP8. Signed-off-by:
Dennis Liu <denliu@nvidia.com> * Fix unittest. Signed-off-by:
Przemyslaw Tredak <ptredak@nvidia.com>
-
- 26 Apr, 2024 1 commit
-
-
Xiaowei Ren authored
* make FusedAttn with CP support bias Signed-off-by:
Xiaowei Ren <xren@nvidia.com> * assert Alibi cannot work with CP Signed-off-by:
-
- 24 Apr, 2024 1 commit
-
-
Kirthi Shankar Sivamani authored
* Try using global buffer for cu_seqlens Signed-off-by:
-
- 22 Apr, 2024 1 commit
-
-
Tim Moon authored
* Remove unnecessary Pylint overrides Signed-off-by:
-
- 19 Apr, 2024 1 commit
-
-
Tim Moon authored
* Support noop concat without providing full tensor Stop storing fused buffers in linear modules. Signed-off-by:
-
- 18 Apr, 2024 1 commit
-
-
Alp Dener authored
fix type checking in checkpointing to assume that there must be TE modules in custom callables Signed-off-by:
-
- 17 Apr, 2024 2 commits
-
-
Pavel Shamis (Pasha) authored
[UB] Adding configurable timeout for userbuffer and improving error reporting for potential hangs (#757) * Improving error reporting and hang detection logic * Adding verbose error reporting in case of UB hang * Adding CE hang detector * Replacing hard-coded timeout with configurable one Signed-off-by:
Pasha (Pavel) Shamis <pasharesearch@gmail.com> * Cleaning up warnings in the code Signed-off-by:
-
Kirthi Shankar Sivamani authored
* fixes; docs Signed-off-by:
-
- 16 Apr, 2024 3 commits
-
-
Alp Dener authored
* changed TE checkpoint passthrough logic to also recursively look for TE submodules Signed-off-by:
-
Kirthi Shankar Sivamani authored
Use torch function as a class method Signed-off-by: -
cyanguwa authored
* WIP: fp8 v1 fprop integration Signed-off-by:
-
- 15 Apr, 2024 1 commit
-
-
Kirthi Shankar Sivamani authored
Don't use autograd hook for bwd reduction Signed-off-by:
-
- 12 Apr, 2024 2 commits
-
-
Sangkug Lym authored
* Add LN margin to inference Signed-off-by:
Sangkug Lym <slym@nvidia.com> * cleanup Signed-off-by:
-
Kirthi Shankar Sivamani authored
* FP8 cuda graphs Signed-off-by:
Charlene Yang <charleney@nvidia.com> * Fix numerics Signed-off-by:
-
- 06 Apr, 2024 2 commits
-
-
Sangkug Lym authored
fix the default userbuffer communicator init settings Signed-off-by: -
Jaemin Choi authored
* Enable DGRAD RS overlap Signed-off-by:
Jaemin Choi <jaeminc@nvidia.com> * fix lint; apply suggestions Signed-off-by:
-
- 04 Apr, 2024 3 commits
-
-
Sangkug Lym authored
* userbuffer fp8 reduction support for individual overlap Signed-off-by:
-
Kirthi Shankar Sivamani authored
* Args can be None Signed-off-by:
-
Pavel Shamis (Pasha) authored
* Fixing potential integer overflow on sequence counter Current implementation may potential cause hangs or data corruption Signed-off-by:
-
