GitLab

* Fix long compile time in padding.cu
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

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

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



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

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

Update copyright to include 2026
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

Signed-off-by: fuyue.lj <fuyue.lj@antgroup.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* [PyTorch] Fuse permute+pad and unpermute+unpad ops for FP8 optimization

1.Fused `moe_permute_with_probs` + `Fp8Padding` and fused `moe_unpermute` + `Fp8Unpadding`,
  that can remove the explicit padding/unpadding of moe expert, improved performance and reduced peak gpu memory usage.
2.Add tests of fused permute/pad and unpermute/unpad.
Signed-off-by: xiaoxi-wangfj <690912414@qq.com>

* [PyTorch/Common] Fuse permute+pad and unpermute+unpad support with_merging_probs
Signed-off-by: xiaoxi-wangfj <690912414@qq.com>

* [PyTorch]format code
Signed-off-by: xiaoxi-wangfj <690912414@qq.com>

* [Common]perf expert_idx loaded once
Signed-off-by: xiaoxi-wangfj <690912414@qq.com>

* fix: pad_offsets can be None
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Signed-off-by: xiaoxi-wangfj <690912414@qq.com>

* add padding + merging probs bwd support. Not tested
Signed-off-by: tdophung <tdophung@nvidia.com>

* Fix garbage initialized act grad
Signed-off-by: tdophung <tdophung@nvidia.com>

* all test passing for jax permutation + pad
Signed-off-by: tdophung <tdophung@nvidia.com>

* change tokens_per_experts APIs to num_out_tokens with conservative allocation of worst case padding for output buffer
Signed-off-by: tdophung <tdophung@nvidia.com>

* change test permutation to reduce test time
Signed-off-by: tdophung <tdophung@nvidia.com>

* triggering PR refresh
Signed-off-by: tdophung <tdophung@nvidia.com>

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

* Remove some tests cases from pytorch side. Add a separate toekn_dispatch test for sanity in case combine accidentally undo an error on dispatch in the roundtrip test. Add distinction between L0 and L2 in test cases in jax
Signed-off-by: tdophung <tdophung@nvidia.com>

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

* remove chance for inefficiency in moving between CPU and GPU, remove redundant primitive using a new static bool for padding, add assert for align size
Signed-off-by: tdophung <tdophung@nvidia.com>

* fix lint in jax
Signed-off-by: tdophung <tdophung@nvidia.com>

* account for both jax newer and older than version 0.8.2. Adjusted gpu triton binding accordingly
Signed-off-by: tdophung <tdophung@nvidia.com>

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

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

---------
Signed-off-by: xiaoxi-wangfj <690912414@qq.com>
Signed-off-by: tdophung <tdophung@nvidia.com>
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Co-authored-by: tdophung <tdophung@nvidia.com>

* rowwise colwise RHT group quant v1
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* remove local array RW
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* change wait_barrier
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* fast math options
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* use mult to replace div
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* format
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* bulk move random states
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* greptile
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* lint
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* revert to use divides
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* avoid fp32 bf16 round-trip in RHT cast fusion
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* trigger fastmath by toggle NVTE_RHT_CAST_FUSION_USE_FAST_MATH
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* integrate row col rht fusion, functional
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* numerics aligned
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* style
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* remove device sync
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* 128 padding
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* revert colwise rng state creation because of row-col fused kernel
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* fix CI, linter
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* refactor RS for generating two random values
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* Avoid invalid configs with templated kernel
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* fix acc pipeline init with 0 arrival count
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* restore rowwise-only mode
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* switch to dynamic atomic scheduler
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* Avoid instantiating group RHT+cast kernel without row-wise or col-wise output
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Include fast math option in quantization config
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Fix linter warnings and review nits
Signed-off-by: Tim Moon <tmoon@nvidia.com>

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

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



* Use TE license
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Fix bug where kernel is always launched on stream
Signed-off-by: Tim Moon <tmoon@nvidia.com>

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

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



* Restore BF16 intermediate downcast in fused RHT-cast kernels
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* fix numerical test of grouped kernel
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* Make sure row-wise and col-wise quantization use different RNG seeds
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

* Restore autoformatter
Signed-off-by: Tim Moon <tmoon@nvidia.com>

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

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



---------
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>
Signed-off-by: Tim Moon <tmoon@nvidia.com>
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Co-authored-by: Tim Moon <tmoon@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

* Check calling convention for amax switch.

Wgrad gemms with colwise x colwise require
rowwise data via general_gemm. Since dy
has both for dgrad and wgrad, the brittleness
has likely not affected results.
Signed-off-by: Keith Wyss <kwyss@nvidia.com>

* Clear rowwise data when applicable.
Signed-off-by: Keith Wyss <kwyss@nvidia.com>

* Update test with columnwise cases.
Signed-off-by: Keith Wyss <kwyss@nvidia.com>

* Check enum value rather than implicit cast.
Signed-off-by: Keith Wyss <kwyss@nvidia.com>

---------
Signed-off-by: Keith Wyss <kwyss@nvidia.com>

* fix ce loss with ignore idx
Signed-off-by: ykarnati <ykarnati@nvidia.com>

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

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

Signed-off-by: ykarnati <ykarnati@nvidia.com>

* remove fix comments
Signed-off-by: ykarnati <ykarnati@nvidia.com>

* fallback divisor to 1
Signed-off-by: ykarnati <ykarnati@nvidia.com>

* have arg for n_rows and n_non_ignore
Signed-off-by: ykarnati <ykarnati@nvidia.com>

* fuse n_non_ignore to softmax kernel
Signed-off-by: ykarnati <ykarnati@nvidia.com>

* fix incorrect arg
Signed-off-by: ykarnati <ykarnati@nvidia.com>

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

* update FE; initial pass at thd
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* produce Stats+Max instead of Max+Sum_Exp
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* Revert "produce Stats+Max instead of Max+Sum_Exp"

This reverts commit c7d2b77b2da9ff3f68344097284187ac427eeb6a.
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

---------
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

change order
Signed-off-by: tdophung <tdophung@nvidia.com>

Fixes to runtime loading logic and add missing deps
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

Remve assumption in quantize/activation kernels that data buffer is initialized
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Initialize empty tensors with shape=[0] instead of shape=[].
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Fix runtime crash in LayerNorm

Still seeing correctness issues.
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Make sure norm workspace sizes are not zero
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Remove assumption in swizzle kernel that data is available.
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Remove assumption in multi-swizzle kernel that data is available.
Signed-off-by: Tim Moon <tmoon@nvidia.com>

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

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



* Remove unnecessary explicit call to default constructor
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Avoid accessing tensor data pointer if tensor has no entries
Signed-off-by: Tim Moon <tmoon@nvidia.com>

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

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



* Apply suggestions from code review
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

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

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



* Update transformer_engine/common/swizzle/swizzle.cu
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

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

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



* Review suggestions from @ptrendx and @greptile-apps
Signed-off-by: Tim Moon <tmoon@nvidia.com>

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

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



* Prefer using row-wise/col-wise shape based on which has data
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Fix merge conflict, expand docs, fix inconsistency in dim function
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Change Tensor::has_data to check whether tensor is initialized, not whether pointer is valid.
Signed-off-by: Tim Moon <tmoon@nvidia.com>

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

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



* Review suggestion from @greptile-apps
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

* Debug incorrect tensor initialization in tests
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Clarify comments that has_data does not guarantee safe pointer accesses
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Debug test failure when computing amaxes
Signed-off-by: Tim Moon <tmoon@nvidia.com>

---------
Signed-off-by: Tim Moon <tmoon@nvidia.com>
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.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>

* Add primary weighs fp8 support for mxfp8
Signed-off-by: kunlunl <kunlunl@nvidia.com>

* Fix unit test and add better error log to unit test
Signed-off-by: kunlunl <kunlunl@nvidia.com>

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

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



* Move post all-gather processing out of for loop
Signed-off-by: kunlunl <kunlunl@nvidia.com>

* Add descriptions and ASCII diagrams for partial cast and partial amax functions
Signed-off-by: kunlunl <kunlunl@nvidia.com>

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

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



* Minor fix based on greptile bot
Signed-off-by: kunlunl <kunlunl@nvidia.com>

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

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



* Fix compilation errors due to arch-specific PTX instructions
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Remove unused noop flag from C API
Signed-off-by: Tim Moon <tmoon@nvidia.com>

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

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



* Expose test_partial_cast
Signed-off-by: kunlunl <kunlunl@nvidia.com>

* Skip mxfp8 partial cast test if mxfp8 is not available
Signed-off-by: kunlunl <kunlunl@nvidia.com>

* Fix pytest error
Signed-off-by: kunlunl <kunlunl@nvidia.com>

* pylint ignore unused manual_post_all_gather_processing
Signed-off-by: kunlunl <kunlunl@nvidia.com>

* Fix error when using is_mxfp8_available
Signed-off-by: kunlunl <kunlunl@nvidia.com>

---------
Signed-off-by: kunlunl <kunlunl@nvidia.com>
Signed-off-by: Tim Moon <tmoon@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Co-authored-by: Tim Moon <tmoon@nvidia.com>

* add grouped_tensor classes and helpers
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* rm non-contiguous option and dptrs
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



* address comments + rework CheckIn/OutputGroupedTensor
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 for compilation
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* make first_dims/last_dims optional + data.shape 2d
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* added assertion
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



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

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

* added logical shape field
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



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

* fixed issues raised by greptile
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



* return default dtype when grouped_tensor is empty
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



* use has_data() for dim queries
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



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

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

* Update transformer_engine/common/transformer_engine.cpp
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* Update transformer_engine/common/transformer_engine.cpp
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* restore Tensor.has_data() + add experimental marks
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



* restore Tensor::has_columnwise_data
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* cleanup
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>
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>

* 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>

* minor fix of torch view dtype
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* multi-tensor RHT amax, compiles
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* setup multi_tensor_quantize_nvfp4_impl
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* wire things up and run without crash
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* numerical test
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* unit test passing
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* finish unit test of split quantize api
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* bump up padding to 64 for nvfp4 grouped quantize
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* fix stochastic rounding
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* lint
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* change error message
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* clean up
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* enable multi-amax without RHT
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* fix col-only quantize mode
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* improve benchmark script
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* add NCU example script
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* add larger test case
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* add contiguous_data_and_scale check to bulk allocator
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* unified naming and differentiate between group_ and multi_
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* move regular amax into multi_tensor.h
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* Disentangle logic for split-quantize and general multi-tensor quantize
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Use size_t for split sections
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Suggestions from @greptile-apps
Signed-off-by: Tim Moon <tmoon@nvidia.com>

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

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



---------
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>
Signed-off-by: Tim Moon <tmoon@nvidia.com>
Co-authored-by: Tim Moon <tmoon@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

* Change order of arguments to make jax works
Signed-off-by: tdophung <tdophung@nvidia.com>

* make num_experts a tl.constepxr again
Signed-off-by: tdophung <tdophung@nvidia.com>

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

* Minor CPU overhead changes
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Cache per device
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

* refactor mxfp8_cast_only kernel
Signed-off-by: Jianbing Dong <jianbingd@nvidia.com>

* fix ptx.cuh after format
Signed-off-by: Jianbing Dong <jianbingd@nvidia.com>

---------
Signed-off-by: Jianbing Dong <jianbingd@nvidia.com>
Co-authored-by: Oleg Goncharov <64355998+Oleg-Goncharov@users.noreply.github.com>

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

* add notes
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

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

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



* small fixes
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>

Fix amax computation using output_t data in normalization
Signed-off-by: Evgeny <etsykunov@nvidia.com>

* enable applying rope offsets in backwared
Signed-off-by: Sudhakar Singh <sudhakars@nvidia.com>

* add tests for rope offsets for thd/bshd/sbhd formats
Signed-off-by: Sudhakar Singh <sudhakars@nvidia.com>

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

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



* minor fixes
Signed-off-by: Sudhakar Singh <sudhakars@nvidia.com>

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

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



---------
Signed-off-by: Sudhakar Singh <sudhakars@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* move triton to common and change paths
Signed-off-by: tdophung <tdophung@nvidia.com>

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

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

* code drop
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>

* [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>

* depracted compile time warning + \warning -> \deprecated
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

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

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



---------
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* Fix cuDNN backend selection for more case. Add CG as a option as well
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

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

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

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

* Add check for window size
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

Deleted unused header
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

* Separated gated and dequantize kernels
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

* Separated quantize, dequantize and gated functions
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

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

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



* Fixed lint issues
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

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

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



* Fixed persistent lint issues
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

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

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



* Added missing compute capability 10.0 check for Quantize FP8 TMA kernels
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

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

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



* Fixed the issue which was added again by autofix
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

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

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



* Changed files description. Completely removed non-identity activations from the NVFP4 transpose test suite
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

* Removed unsupported template arguments in NVFP4 quantize
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

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

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



* Fixed undefined symbol error
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

* Fixed condition
Signed-off-by: Oleg Goncharov <64355998+Oleg-Goncharov@users.noreply.github.com>

* Fixed CUDA version check
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

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

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



* Changed arch conditions order
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

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

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



* Fix
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

* Clean up
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

* Small fix
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

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

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



* Small fix
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

* Fixes per the PR review
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

* Fix
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

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

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



* Split quantize helper into two (FWD and BWD) functions
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

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

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



* Moved activation functions from cast.cu. Removed cast.cu from the fast-math compilation list
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

* Enabled fast math for activations by default
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

* Disabled fast math for activations by default
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>

---------
Signed-off-by: Oleg Goncharov <ogoncharov@nvidia.com>
Signed-off-by: Oleg Goncharov <64355998+Oleg-Goncharov@users.noreply.github.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

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

* Remove nvidia-mathdx dep
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

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

* add max_score for fused/unfused F16 non-CP
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* calculate max per head instead of max over all heads
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix fused attn max_score shape
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* revert FE to github
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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



* update FE to 1.15.0-rc
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix merge
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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



* reduce ew kernels; fix causal masks; add more tests
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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



* minor fix to tests
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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



* remove logic for flash-attn
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* WIP: add CP support for p2p/a2a/all_gather
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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



* minor improvements of implementation/tests
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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



* WIP: add thd support
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* add thd to UnfusedDPA
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix lint
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* more fixes for lint
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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



* update to FE 1.15
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove unneeded changes
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* disable unfused for thd + pad_between_seqs
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* minor fixes
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* disable thd for unfused until bug is fixed
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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



* fix all_gather
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix all gather
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* rename max_score to max_logit
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix all_gather
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix all_gather
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* disable fused attn + thd
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

---------
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

fix checks in unoptimized non-rht fp4 quantize kernel
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Added sm_120f to the build
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Change the arch specific handling
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Fix
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Support for CUDA<12.9
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Moved through the rest of the files
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Fix
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Common cases
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Remove pure 100 from the list
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Fix
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* CMake changes, (not yet working)
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Fix
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Do not pass the arch-specific thing from build_tools
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Fix
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Moved some of the files to arch-specific compilation
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Fix and also changing the order of compilation to hopefully get the
compilation time lower
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Fix for the files overwriting custom compile properties
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Actually make this whole thing work
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

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

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



* Add space to the error message
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Signed-off-by: Przemyslaw Tredak <ptrendx@gmail.com>

* Apply suggestions from code review
Co-authored-by: Oleg Goncharov <64355998+Oleg-Goncharov@users.noreply.github.com>
Signed-off-by: Przemyslaw Tredak <ptrendx@gmail.com>

* Fixes from review
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Changing the naming to be more intuitive
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

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

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



* Add missing cassert include for device-side asserts
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

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

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



---------
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>
Signed-off-by: Przemyslaw Tredak <ptrendx@gmail.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Oleg Goncharov <64355998+Oleg-Goncharov@users.noreply.github.com>

* pipeclean, fix nvfp4 padding of 32 alignment
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

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

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



* numerical test passed
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* fix CI failure with test_cast_master_weights_to_fp8 (in a hacky way)
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* found CUDA mis-aligned address error in training in multi-swizzle, hack the vec_load_size to 1 to unblock
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* leave comments about alignment issue
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* fused bulk alloc nvfp4
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* fix RHT sign mask CPU overhead
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* fix
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* resolve comments
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>

* Remove incorrect logic that treats 0-D tensor as uninitialized

Tensor shape logic still requires treating 0-D tensor as uninitialized.
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Fix invalid conversion from tensor to int
Signed-off-by: Tim Moon <tmoon@nvidia.com>

---------
Signed-off-by: Zhongbo Zhu <zhongboz@nvidia.com>
Signed-off-by: Tim Moon <tmoon@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Tim Moon <tmoon@nvidia.com>
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

* Support wheel build for cuda 13
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

* Fixes for cu13 runtime, format
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

* Better error handling
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

* Modify function names
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

Make `CanonicalizeGemmInput()` support non-TN layout FP8 GEMM on Blackwell with column-wise/transposed data (#2233)

Modified CanonicalizeGemmInput() logic to pull from column-wise data for FP8 GEMM on Blackwell when row-wise is not available.
Signed-off-by: Alp Dener <adener@nvidia.com>

Signed-off-by: Tim Geypens <tim.geypens@gmail.com>

* Initial API change
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Change all imports and api
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

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

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

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

* fix docs, tests, and make Jax change as well
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Change internal uses of fp8_autocast
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

* CG function, and small test fixes
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Change instances of make_graphed_callables internally
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

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

* Fix test and add more docs
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Cleanup test imports and minimize internal file imports
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Make is_bf16_available public
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

* Better docs and better api
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

* Apply suggestions from code review
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

* fix nvfp4 test
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

---------
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>