Commits · 50352325811043208f5b2a5edbb941ca560df2d0 · OpenDAS / TransformerEngine

09 Dec, 2025 1 commit

Jax primitives for permutation on single GPU (#2473) · 46c6ef31

Teddy Do authored Dec 09, 2025



* branch off of initial permutation jax-triton PR
Signed-off-by: tdophung <tdophung@nvidia.com>

* Set 0 as the size of dummy tensors to reduce memory usage.
Signed-off-by: tdophung <tdophung@nvidia.com>

* Correct setting of permuted_probs_stride_token, unpermuted_probs_stride_token and unpermuted_probs_stride_expert in unpermutation
Signed-off-by: tdophung <tdophung@nvidia.com>

* Implement primitives, wrapper, test for wrapper, edit trit
on binding to accomodate scalars
Signed-off-by: tdophung <tdophung@nvidia.com>

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

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



* Change implemementation of VJP functions to match correct pattern. Deduce some static scalar args from shapes of inputs. Accept B, S instead of num_tokens. Change test to use value_and_grad to test vjp funcs properly
Signed-off-by: tdophung <tdophung@nvidia.com>

* formatting
Signed-off-by: tdophung <tdophung@nvidia.com>

* fix pylint
Signed-off-by: tdophung <tdophung@nvidia.com>

* fix test to compare to the correct reference impl. relax 1 tol for grad compare, fix lint the rightway
Signed-off-by: tdophung <tdophung@nvidia.com>

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

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



* fix test_permutation to use value_and_grad for reference impl, tighten tols, and add unpermute with probs for token combine bwd rule
Signed-off-by: tdophung <tdophung@nvidia.com>

* added forgotten file in prev commit
Signed-off-by: tdophung <tdophung@nvidia.com>

* format
Signed-off-by: tdophung <tdophung@nvidia.com>

* merge with_probs to without_probs
Signed-off-by: tdophung <tdophung@nvidia.com>

* add aserts and fix lint
Signed-off-by: tdophung <tdophung@nvidia.com>

---------
Signed-off-by: tdophung <tdophung@nvidia.com>
Co-authored-by: Ming Huang <mingh@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

46c6ef31

06 Dec, 2025 1 commit

[JAX] Add CP + THD + AG + Striped>1 + SWA support (#2379) · fd0cd12e

Kshitij Lakhani authored Dec 05, 2025



* Add generic stripe_height support for load balancing
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Fix imports in test for deprecated jax.experimental.pjit
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Add test case for stripe_height greater than 1. Add stripe_height arg to reordering methods
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-eos01.eos.clusters.nvidia.com>

* Add Striped 1 and 4 test cases. Refactor the Load Balancing test case. Fix the incorrect shape in striping inverser reordering
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-eos01.eos.clusters.nvidia.com>

* Modify test code for CP + AG + THD + stripe height greater than 1
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-eos01.eos.clusters.nvidia.com>

* Add stripe_height arg to fused attn and fused attn fwd API. Add appropriate mask checks for AG+THD+CP and pick BRCM to be executed per rank. Add Fused Attn Primitive for CP + THD +AG + Striping. Add a method to reorder and all gather segment ids and offsets for kv
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-eos01.eos.clusters.nvidia.com>

* TMP: Throwaway testing commit
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-eos01.eos.clusters.nvidia.com>

* Add comments in primitive registration process
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-eos01.eos.clusters.nvidia.com>

* TMP: Throwaway test commit
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Undoing incorrect rebase/merge leftovers
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-eos01.eos.clusters.nvidia.com>

* TMP: Throwaway test commits
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-eos01.eos.clusters.nvidia.com>

* Add support for calculating q and kv seqlens and offsets per rank for CP+THD+AG+SW+Striped>1 primitive
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-eos01.eos.clusters.nvidia.com>

* Augment jax primitive register code comments
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-eos01.eos.clusters.nvidia.com>

* Fix the array sizes and padding values returned for seqlens and offsets to fit what the fused attn primitive non cp computation
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Add support in new primitive for softmax_offset related changes. Put in missing primitive registering line in again. Increase the seqoffsets arrays lengths by 1
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-preos01.a51.clusters.nvidia.com>

* Add new set of helper functions for seqlens and seqoffsets fo AG+THD+CP+Stripe>1 which accounts for batching and seq offsets size b+1
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Add backward primitive for CP+THD+AG+Striped>1
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Modify tests for backward primitive for CP+THD+AG+Striped>1
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Move stripe_height along with other static args in fused_attn_bwd rule. Fix typo in CP+AG+TH+Striped>1 primitive
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-preos01.a51.clusters.nvidia.com>

* Code clean up: remove older version for calculating seqlens and offsets for CP+AG+THD+striped>1 primitive
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Add test for CP+THD+AG+Striped>1
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Fix missing var
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



* Add SWA tests for AG+Striped>1+CP+THD+SWA
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-preos01.a51.clusters.nvidia.com>

* Restoring test code
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-preos01.a51.clusters.nvidia.com>

* Remove assert preventing SWA code path in CP+AG+Striped primitive
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-preos01.a51.clusters.nvidia.com>

* Parametrize num_segments_per_seq in 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



* Clean up test code
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

Clean up test code in TE common
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

Clean up debug statements
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Rename stripe_height to stripe_size
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Code clean up and add additional comments
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

nit: Apply suggestions from code review
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Signed-off-by: Kshitij Lakhani <33047503+KshitijLakhani@users.noreply.github.com>

Fix type on fused attn tests
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-preos01.a51.clusters.nvidia.com>

* Fix seqoffsets length to be passed onto FusedAttn primitive as it is b and not b+1 needed by cuDNN
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-preos01.a51.clusters.nvidia.com>

* Remove commented code
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Signed-off-by: Kshitij Lakhani <33047503+KshitijLakhani@users.noreply.github.com>

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

Fix incorrect greptile change
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



* Skip THD test cases for CP + AG + Dual chunk. Skip BSHD cases for CP + AG + Striped>1. Correct the layout and shapr parameters passed to the tests
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-preos01.a51.clusters.nvidia.com>

* Pass stripe_size explicitly for ring attn tests for THD cases
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-preos01.a51.clusters.nvidia.com>

* Remove TODO
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-preos01.a51.clusters.nvidia.com>

* Explicitly fail if THD + AG is being used with a non padding causal mask
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-preos01.a51.clusters.nvidia.com>

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

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



* nit: Correct the ID for the test dist fused attn tests to account for cp*2 which is done under the hood
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Set num_segments_per_seq defaults to None instead of 0
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Augment comments. Add ValueError for stripe_size=0
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Test only 1 num_segments_per_seq combination for CP+AG+THD+Striped>1+SWA instead of 2. Modify the num segments and window size to easily to debug values
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Default stripe_size to None instead of 0. Modify stripe_size check for <=0 instead of ==0
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



* Remove incorrectly added file
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Explicitly pass zero sized arrays for seg ids and pos in the CP + AG + Striped primitive rather than using the seqlens or the offsets as placeholders
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Fix linting errors
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



* Add a deep dive doc for CP+THD+AG+Stripe>1+SWA regarding design considerations and decisions
Signed-off-by: Kshitij Janardan Lakhani <klakhani@nvidia.com>

* Put docs and pngs into it's separate dir
Signed-off-by: Kshitij Janardan Lakhani <klakhani@nvidia.com>

* Replace png screenshots with markdown coe blocks for the attention patterns. Remove unecessary pngs
Signed-off-by: Kshitij Janardan Lakhani <klakhani@nvidia.com>

* Add doc file to index.rst. Fix grammatical errors
Signed-off-by: Kshitij Janardan Lakhani <klakhani@nvidia.com>

---------
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-eos01.eos.clusters.nvidia.com>
Signed-off-by: Kshitij  Janardan Lakhani <klakhani@login-preos01.a51.clusters.nvidia.com>
Signed-off-by: Kshitij Janardan Lakhani <klakhani@nvidia.com>
Co-authored-by: Kshitij  Janardan Lakhani <klakhani@login-eos01.eos.clusters.nvidia.com>
Co-authored-by: Kshitij  Janardan Lakhani <klakhani@login-preos01.a51.clusters.nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

fd0cd12e

26 Nov, 2025 1 commit

Docs fix (#2301) · df39a7c2

Paweł Gadziński authored Nov 26, 2025



* init
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* lines lenght
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* subtitle --- fix in many files:
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* cross entropy _input -> input rename
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* cross entropy _input -> input rename
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* a lot of small fixes
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* torch_version() change
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* add missing module and fix warnings
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* removed training whitespace:
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* Update docs/api/pytorch.rst
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Signed-off-by: Paweł Gadziński <62263673+pggPL@users.noreply.github.com>

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

* Fix more imports
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Fix NumPy docstring parameter spacing and indentation

- Standardize parameter documentation to use 'param : type' format (space before and after colon) per NumPy style guide
- Fix inconsistent indentation in cpu_offload.py docstring
- Modified 51 Python files across transformer_engine/pytorch
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

---------
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>
Signed-off-by: Paweł Gadziński <62263673+pggPL@users.noreply.github.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

df39a7c2

25 Nov, 2025 1 commit

[JAX] Allow DP + FSDP and fixed sr_rng_state partitioning (#2418) · 66ae3030

Phuong Nguyen authored Nov 25, 2025



* allow dp + fsdp and fixed sr_rng_state partitioning
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

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

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



* cleanup for lint test
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

66ae3030

18 Nov, 2025 1 commit

[JAX] Add support for sink attention in JAX (#2225) · 15cefbc5

Paweł Gadziński authored Nov 18, 2025



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* removed packed versions
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* jax
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fixes
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix:
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fixes
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* fixes
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fixes
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* sofmtax_fusion -> softmax_fusion_type
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

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

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



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

---------
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

15cefbc5

14 Nov, 2025 2 commits

[JAX] Use TE quant if TE fused act is disabled (#2374) · 389a6ba4

jberchtold-nvidia authored Nov 14, 2025



* Use TE quant if TE fused act is disabled
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Keep existing precision
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

---------
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

389a6ba4

[JAX] Improve support and testing for direct recipe usage without autocast contexts (#2366) · a0754757

jberchtold-nvidia authored Nov 14, 2025



* Refactor to avoid storing a global quantization config so direct recipe passing works as intended
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* fix use_split_accumulator for current scaling recipe
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* fix tests that pass direct recipe and were missing quantize meta set
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Revert "fix use_split_accumulator for current scaling recipe"

This reverts commit a74ab7df812ec0a069b1bdd208debb93ec25a900.
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* fix ci failures
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Fix amax_history post_init
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Signed-off-by: jberchtold-nvidia <158520091+jberchtold-nvidia@users.noreply.github.com>

* Update transformer_engine/jax/quantize/quantizer.py
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Signed-off-by: jberchtold-nvidia <158520091+jberchtold-nvidia@users.noreply.github.com>

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

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



* fix ci failures
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* fix ci issue
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* address comments
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* make recipe assertion classes in test_recipe_characteristics not inherit from unittest.TestCase
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

---------
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>
Signed-off-by: jberchtold-nvidia <158520091+jberchtold-nvidia@users.noreply.github.com>
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

a0754757

13 Nov, 2025 2 commits

[JAX] Shardy rule + QuantizeLayout Rework (#2364) · 9440b76a

Phuong Nguyen authored Nov 13, 2025



* shardy + quantize_layout rework
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* add assertion for NVFP4 in fused act and fused norm primitive
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* add assertions
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

9440b76a

[JAX] NVFP4 scale swizzling via nvte kernel (#2350) · ef28c865

Phuong Nguyen authored Nov 13, 2025



* swizzle via nvte
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

ef28c865

07 Nov, 2025 1 commit
- Fix sharding of segment position to match id in ring attention. (#2349) · f62cad90
  Michael Goldfarb authored Nov 06, 2025
  
  f62cad90
30 Oct, 2025 1 commit

[JAX] Fix: Skip determinism tests for bprop for all sm >=100 (#2315) · 5e8a9a96

Kshitij Lakhani authored Oct 30, 2025



* Fix: Skip determinism tests for bprop for all sm >=100
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Add username to TODO
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Assert in fused attn bwd pass for sm100+
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



---------
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

5e8a9a96

28 Oct, 2025 1 commit

[JAX] Use TE quantization when TE fused norm is disable (#2303) · a8e4346e

Phuong Nguyen authored Oct 28, 2025



* jax norm + te quant
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>


---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

a8e4346e

22 Oct, 2025 2 commits

[JAX] Defer TE/JAX cublas shape check on fp8 gemms until lowering (#2292) · 2ac3c168
jberchtold-nvidia authored Oct 22, 2025
```
Defer cublas check on fp8 gemms until lowering
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>
```
2ac3c168

[JAX] NVFP4 recipe with option to enable/disable SR, RHT, and 2D quantization (#2270) · 818b30cc

jberchtold-nvidia authored Oct 22, 2025



* [JAX] Support recipe flags for disabling SR, RHT, and 2D quantization
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* lint
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Fix issue with SR state being erased due to pytree handling of NVFP4Quantizer
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Add test for SR state preservation across VJP boundaries
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Fix sharding of SR rng state
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* lint
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* update tolerances slightly now that SR is enabled
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* lint
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Use hashlib for deterministic hashes across runs for SR
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* rename uses_rht on scaled tensors to has_applied_rht
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* add assert
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Move decision of whether to use RHT into helper.py and add dedicated RHT tests
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* lint
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* fix use_rht attr usage
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* fix pure-jax rht usage criteria
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Adjust tolerances after rebase
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

---------
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

818b30cc

13 Oct, 2025 1 commit

[Common][JAX] Improve error message for cublas fp8 gemm with incorrect shape (#2261) · 8c364b4d

jberchtold-nvidia authored Oct 13, 2025



* Improve error message for cublas fp8 gemm with incorrect shape
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* lint
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Removed unnecessary non-contracting size check
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* rename inner dim -> leading dim
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

---------
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

8c364b4d

09 Oct, 2025 1 commit

[JAX] NVFP4 support in TE/JAX (#2254) · 8a7ab3dd

jberchtold-nvidia authored Oct 09, 2025


Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>
Co-authored-by: Phuong Nguyen <phuonguyen@nvidia.com>

8a7ab3dd

08 Oct, 2025 1 commit

[JAX] Async issuing D2H memcpy for grouped_gemm group_sizes array (#2213) · af2a0c16

Hua Huang authored Oct 08, 2025



* Try async copy of grouped GEMM group_sizes data
Signed-off-by: Hua Huang <huah@nvidia.com>

---------
Signed-off-by: Hua Huang <huah@nvidia.com>
Co-authored-by: Phuong Nguyen <phuonguyen@nvidia.com>

af2a0c16

07 Oct, 2025 1 commit

[JAX] Activation/Normalization to output amax for later quantization in CurrentScaling (#2238) · 127b6d3a

Phuong Nguyen authored Oct 07, 2025



* reuse amax for current scaling
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

127b6d3a

06 Oct, 2025 1 commit

[JAX] Fix for GEMM + fuse bias + AllReduce (#2230) · 0db0f4d2

Phuong Nguyen authored Oct 06, 2025



* not fuse bias for output all reduction case + unit tests
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* norm to reduce dgamma along tpsp as well
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* clean up tests
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* fix test_distributed_layernorm byte counts
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* increase tols for jax_gemm
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

0db0f4d2

03 Oct, 2025 1 commit

[JAX] Clamped Swiglu Integration (#2194) · b840898b

vthumbe1503 authored Oct 03, 2025


Signed-off-by: Varun Thumbe <vthumbe@nvidia.com>
*Jax integration for clamped swiglu. This is the continuation of PR which added Clamped Swiglu(used in GPT OSS) support in TE along with Pytorch integration. This PR hooks up the clamped swiglu and dswiglu's nvte APIs to TE Jax.

b840898b

01 Oct, 2025 2 commits

[JAX] Fix `rng_state` shape in fused attention (#2217) · b0d562d8

Phuong Nguyen authored Oct 01, 2025



fix rng_state shape
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>
Co-authored-by: jberchtold-nvidia <158520091+jberchtold-nvidia@users.noreply.github.com>

b0d562d8

[JAX] Rework amax reduction over TPSP (#2218) · ac4e0fd6

Phuong Nguyen authored Oct 01, 2025



* rm using_global_amax_of_x
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

ac4e0fd6

27 Sep, 2025 1 commit

[JAX] CollectiveGemm (#2166) · d75bf43f

Phuong Nguyen authored Sep 27, 2025



* init cgemm + unit tests

* UB bootstrap with NCCL, no MPI dependency

* add NVLINK-P2P check + error message

* skip tests if no NVLINK available

* use std::vector to store ncclComm_t

* update misuse of TP warning
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

d75bf43f

23 Sep, 2025 2 commits

[JAX] Restore Shardy Rule with CompoundFactor (#2167) · 3f875fb5

Phuong Nguyen authored Sep 23, 2025



* Rework shardy rules

* WAR for compound factor=1
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

3f875fb5

[JAX] Local-Amax for Current-Scaling (#2183) · a92a0ad2

Ming-Xu Huang authored Sep 23, 2025



* Adding Amax Primitive and related args.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Enable local-amax for current-scaling and optionally run AR aross FSDP/TP/SP.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Adding doc for Amax Primitive.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Fix the function name conflict.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Modification as feedback suggested.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Fix errors from lint.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Fix the wrong amax-scope in the bwd.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Added more description for amax-scope
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Fix the wrong attribute name.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Keep dim for AmaxCalcuation.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Remove keepDim and add shardy_rule
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Fix shardy_rule
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Remove extra-collective bytes from ref_coll_count due to local amax.
Signed-off-by: Ming Huang <mingh@nvidia.com>

---------
Signed-off-by: Ming Huang <mingh@nvidia.com>
Signed-off-by: Ming-Xu Huang <mingh@nvidia.com>
Co-authored-by: Phuong Nguyen <phuonguyen@nvidia.com>

a92a0ad2

22 Sep, 2025 1 commit

[JAX] Remove import jax.extend.ffi (#2193) · 57b4d7bc

Phuong Nguyen authored Sep 22, 2025



* remove import jax.extend.ffi
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

57b4d7bc

15 Sep, 2025 1 commit

Lower precision gated-act to accelerate FP8 current-scaling. (#2153) · cd2034f3

Ming-Xu Huang authored Sep 15, 2025



* Applying the original precision as Norm outputs' and activation compuations.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Adding knob to control norm output precision.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Removing the knob and applying lower-precision norm with current-scaling only.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Fix the error when quantizer==None
Signed-off-by: Ming Huang <mingh@nvidia.com>

---------
Signed-off-by: Ming Huang <mingh@nvidia.com>

cd2034f3

09 Sep, 2025 1 commit

[JAX] Scale swizzling via JAX transpose op (#2163) · 5f2b8310

Phuong Nguyen authored Sep 09, 2025



* add swizzle in jax
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* added outer_impl
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* clean up FFI
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

5f2b8310

05 Sep, 2025 1 commit

[JAX] NoScaleTensor wrapper for non-quantized data (#2136) · c47f329b

jberchtold-nvidia authored Sep 05, 2025



* Custom call tests passing
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Fix test_layer.py
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Lint
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Fix comments
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Support using amax on HighPrecision tensor if it exists instead of recomputing for current scaling
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Fix shardy issue with amax being shape 1,1,1 instead of shape (1,)
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Add higher-precision VJP tests to test_distributed_layernorm_mlp
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Cast non-quantized kernels to input dtype in VJPs
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Rename HighPrecisionTensor to NoScaleTensor
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Use NoScaleTensor in pure JAX impls where it was missing
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Fix tests
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

---------
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

c47f329b

03 Sep, 2025 1 commit

[JAX] Fix failing fused attn tests for dropout=0.1 and bias for sm100 (#2135) · f378eaf2

Kshitij Lakhani authored Sep 03, 2025



* Fix failing tests for dropout=0.1 and bias for fused attn for blackwell
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 the skip message
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Assert in fused attn bwd pass for sm100
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

Add check for sm100
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



* Add support to get all devs in the process for jax
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



* Code clean up
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Make get_all_device_compute_capability more pythonic, thereby avoiding unnecessary type conversion
Signed-off-by: Kshitij Lakhani <klakhani@nvidia.com>

* Represent attn bias using enum instead of string
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>

f378eaf2

27 Aug, 2025 3 commits

[JAX] Add amax input to DBiasQuantizePrimitive and FFI (#2118) · 1e2c68d6

Phuong Nguyen authored Aug 27, 2025



* add amax input to DBiasQuantizePrimitive and FFI
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

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

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



* make sure amax is init with zero
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* fix sharding rule
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

1e2c68d6

[JAX] Decouple Recipe and ScalingMode (#1728) · c9508000

jberchtold-nvidia authored Aug 27, 2025



* Decouple recipe and scaling mode
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Expose global QuantizeConfig instance as a getter
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Format and lint
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Merge branch 'main' into dev/jberchtold/jax-scaling-mode-and-recipe-decoupling
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Rename UsageType to TensorSource
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Update test_layer.py
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

---------
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>
Signed-off-by: jberchtold-nvidia <158520091+jberchtold-nvidia@users.noreply.github.com>

c9508000

FP8 AllGather in FP8 GroupedGEMM + Fix Stream Usage Issue. (#2086) · 62a57dd4

Ming-Xu Huang authored Aug 27, 2025



* FP8 AllGather in FP8 GroupedGEMM

1. Support current scaling FP8 quantation with a given amax.
2. Support FP8 AG in fwd and BF16 RS in bwd.
3. The workflow is AR-max -> FP8 Quant -> FP8 AG -> FP8 GroupedGEMM.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Slightly refactor
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Adding documents of new args.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Adding unit-tests.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Adding license.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Move unit-tests to L1.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Move quantizaer store/reset into FP8 only.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Adding all layout support for Blackwell+
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Adopt the feedback from code-review.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Fixed the wrong stream used by d2d in groupedGEMM FFI.
Signed-off-by: Ming Huang <mingh@nvidia.com>

---------
Signed-off-by: Ming Huang <mingh@nvidia.com>
Co-authored-by: Phuong Nguyen <phuonguyen@nvidia.com>

62a57dd4

26 Aug, 2025 2 commits

[JAX] Add `tpsp_resource` in the `MeshResource` map (#2113) · d770886f

Phuong Nguyen authored Aug 26, 2025



* clean up sharding
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* added tpsp_resource
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* update tests
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* rework test for MeshResource
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* add mesh_resource into fp8_autocast in test_helper.py
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

d770886f

[JAX] `ScaledTensor1x` to store `amax` (#2117) · 3d0ea80a

Phuong Nguyen authored Aug 26, 2025



* added amax as an optional arg
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

3d0ea80a

25 Aug, 2025 1 commit

[JAX] Add Shardy warning in GEMM custom call (#2101) · 2e23ad71

Phuong Nguyen authored Aug 25, 2025



* added shardy warning
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>


---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

2e23ad71

20 Aug, 2025 1 commit

[JAX] Error checking for mesh resource and update GemmPrimitive to use... · bc99a88d

jberchtold-nvidia authored Aug 20, 2025


[JAX] Error checking for mesh resource and update GemmPrimitive to use global_mesh_resource().fsdp_resource (#2088)

* Enforce global MeshResource is set
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Use global_mesh_resource().fsdp_resource in gemm primitive
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Update tests
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Update gemm.py
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Update test_layer.py
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

---------
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

bc99a88d

18 Aug, 2025 1 commit
- [JAX] Fix for TE GEMM - Always AllGather RHS non-contracting dims with FSDP axis (#2075) · 3fc1e4bf
  Phuong Nguyen authored Aug 18, 2025
```
* fix fsdp
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>
```
  3fc1e4bf
12 Aug, 2025 1 commit

[JAX] Support custom recipe and custom collection name when creating quantizer sets (#2059) · 6a4e871e

jberchtold-nvidia authored Aug 12, 2025



* Support setting collection name for quantizer set Flax variables in TransformerEngineBase flax module
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Support creating quantizer set from a recipe directly
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Fix debug error format string in gemm.py
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Lint
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

---------
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

6a4e871e

08 Aug, 2025 1 commit

[JAX] Enable TE GEMM custom call for all recipes (#2047) · 235c8d00

Phuong Nguyen authored Aug 08, 2025



* enabled TE GEMM for all recipes
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* add warnings
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

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

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



* fix lint
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

235c8d00