GitLab

* Add control of attention dropout and hidden dropout RNG state
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fix CI error
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

---------
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* [Paddle] Add TP, DP, PP, FSDP
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Minor fix
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fix CI failure
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Remove set_nccl_overlap_warning_if_tp
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Improve variable naming
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Refactor FP8 Buffer
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Stylic changes
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fix FP32 parallel training
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fix numel performance issue
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Squashed commit of the following:

commit 79e2e5fd774e67dcdda9aae01a9f31a6479c5d70
Author: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>
Date:   Sun Aug 20 14:39:16 2023 +0000

    Add TP test
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

commit 1d40ad60540490f97ed82ba877cc6eda8902cbf6
Author: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>
Date:   Sun Aug 20 14:22:25 2023 +0000

    Fix tp_size when disabled
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

commit 6632f735a0c8251862355fc74622af59fae3a509
Author: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>
Date:   Sun Aug 20 05:52:18 2023 +0000

    Add TP for attention and transformer layer
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Add shape check
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Add FSDP check for stage 1,2,3
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Review changes
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fix group_sharding test
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Support NVTE_FUSE_ATTN
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fix CI errors
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

---------
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* API change and some test fixes
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

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

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

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

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

* Fused attention kernel only supports sm80 and sm90
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Update transformer_engine/jax/csrc/modules.cpp
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* arbitary fused kernel supports sm86/sm89 after 8.9.3
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Skip sm70
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Forward is_fused_attn_kernel_available to cpp backend
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Remove cpp is_fused_attn_available API
Signed-off-by: Reese Wang <rewang@nvidia.com>

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

fix rng_state issue and minor compiler warning
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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

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

* Expand mask type and add no mask numerical test
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

* Add nn.layer: softmax, attention, transformer
Signed-off-by: Shijie Wang <jaywan@nvidia.com>

* code refactor
Signed-off-by: Shijie Wang <jaywan@nvidia.com>

* code refactor
Signed-off-by: Shijie Wang <jaywan@nvidia.com>

* update docs and set dropout=0.1
Signed-off-by: Shijie Wang <jaywan@nvidia.com>

* Update transformer_engine/paddle/layer/attention.py
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

---------
Signed-off-by: Shijie Wang <jaywan@nvidia.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

* Save full global state
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

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

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

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

* add back pytorch csrc extensions.h
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* add unit tests for dpa checkpointing
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove seqlen%32/64 checks for now
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix tests for core attn bias
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* add tests for changes regarding rng_state in aux_ctx_tensor
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* reuse rng tracker from numerics in fused attn; skip checkpointing if FAv2 in numerics
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* uncomment comments used for testing
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix pre/post scale bias
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* Update transformer_engine/pytorch/attention.py
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Signed-off-by: cyanguwa <8636796+cyanguwa@users.noreply.github.com>

* remove skipifs for FAv2 check after PR366
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove checkpointing tests for transformer layer; dpa tests still provide coverage
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* adjust random number range for tests
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* Add upper bound to FA version
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Check backend only when using FusedAttention
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* remove imports/variables related to FAv2 checks
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* further fix random number ranges for tests
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix variable referenced before assignment error
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: cyanguwa <8636796+cyanguwa@users.noreply.github.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Add dynamically shaped input mask in test_export_softmax
* Fix test_softmax_mask_fn - use env. var `NVTE_ONNX_KVCACHE_MAX_SEQ_LEN` to control whether the test uses the default mask generation function or dynamic TRILU mask slicing.
* Change core_attention ONNX export test: use "no_mask" as attn mask type when testing `te.attention.DotProductAttention` w/o masking.
* Use ORT CUDA backend by default.
Signed-off-by: Neta Zmora <nzmora@nvidia.com>

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

Fix for the RMSNorm tests/doc/ONNX export to match the actual
implementation
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Fix flash attention dropout probability with inference
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Add output as the fused attention ctx tensor
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Add rng_state as the fused attention ctx tensors
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Add flash attention supported lengths to the fused attention
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Refactor attention primitive to reuse abstract shaped array
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Detect backend type to allocate appropriate ctx size
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Skip dropout correctness instead of return success
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Use cudaMemsetAsync and enhance the error handling
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Add flash attention kernel elts_per_thread update
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Remove redundant max 512 suffix
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Keep only DType and remove NVTEDType from python
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Fix a float32_attention_logits bugs
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Re-calculate workspace size for self attention
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Enhance bias/dbias shape guard
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Enhance the seed/rng_state checker
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Use jax.core.ShapedArray as jax.abstract_arrays is deprecated
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Enhance the unittest docs
Signed-off-by: Reese Wang <rewang@nvidia.com>

---------
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Cast Flax collections to FrozenDict as WAR to adapt Flax 0.7.1.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Adding min version of flax to requirements.txt in examples
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Fix praxis tests and rename compare_frozen_dict to compare_dict
Signed-off-by: Ming Huang <mingh@nvidia.com>

* [Paddle] Refactor FP8 state (#350)

Refactor fp8 state
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Store FP8 checkpointing data in CPU (#351)
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Make test_layer be able to run on both Flax >=0.7.1 and <=0.7.0
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Update Flax version
Signed-off-by: Tim Moon <tmoon@nvidia.com>

---------
Signed-off-by: Ming Huang <mingh@nvidia.com>
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Signed-off-by: Tim Moon <tmoon@nvidia.com>
Co-authored-by: Tian Zheng <tizheng@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: Tim Moon <tmoon@nvidia.com>

* Add FP8 support

- Add FP8 recipe
- Add FP8 path for nn layers
- Add MNIST FP8 example
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Update README
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fix LayerNormMLP FP8 backward
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fix FP8 training in float32 accumulation
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fix FP8 checkpointing for non forward execution cases (same as #323)
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Refactors and improvements for better code stype, readability and organization
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Remove unnecassary pylint override
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

---------
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* add support for multi-query/grouped-query attention
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* revert to flash-attn 1.0.6 and build 2.0.0.post1 manually in CI
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* add keyword name for DPA input
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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

* Update transformer_engine/pytorch/attention.py
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Update tests/pytorch/test_fused_attn.py
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Fix TP and SP case
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* add skipifs for pytest
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove higher limit for flash-attn version
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: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Exposing RMSNorm in pyTorch extensions
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* First pass at the Python API
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

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

* Added numerics tests and fixed issues
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

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

* Added RMSNorm to LayerNormMLP
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Added ONNX export and tests for RMSNorm
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

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

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

* Added normalization option to the TransformerLayer
Added tests
Fixed test failures
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Fix documentation
Co-authored-by: Przemyslaw Tredak <ptrendx@gmail.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

* Fix IMA and invalid type error
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Increase RMSNorm threshold for bf16 case
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

* add flash attn tests
Signed-off-by: Shijie Wang <jaywan@nvidia.com>

* update flash attn
Signed-off-by: Shijie Wang <jaywan@nvidia.com>

* fix random seed
Signed-off-by: Shijie Wang <jaywan@nvidia.com>

---------
Signed-off-by: Shijie Wang <jaywan@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

Replace deprecated sharding API
Signed-off-by: Tim Moon <tmoon@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Add Linear layer (FP16)
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

- Add BF16 training example
- Add fp8_autocast (only supports non-fp8 for now)
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Remove FP8 stuff
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Simplify Linear layer forward
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Add LayerNorm layer (BF16)
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Add LayerNormLinear layer
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Store weights in BF16
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Add LayerNormMLP layer
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Add BF16 MNIST example
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Remove in-place cast for compatibility with Paddle AMP mechanism
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* README correction
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Add Paddle op as a backend option
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fix code format
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fix dtype change between iterations
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Minor fixes
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Move forward function out of base layer
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Use Paddle nvtx bindings
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

---------
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fully remove attn_type and set self_attn_mask_type default to 'causal'
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Fix tests with new arguments
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Explicit self_attn_mask_type for examples
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Update transformer_engine/jax/flax/transformer.py
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Signed-off-by: zlsh80826 <rewang@nvidia.com>

* Update transformer_engine/jax/flax/transformer.py
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Signed-off-by: zlsh80826 <rewang@nvidia.com>

---------
Signed-off-by: Reese Wang <rewang@nvidia.com>
Signed-off-by: zlsh80826 <rewang@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Disable TorchDynamo optimizations in PyTorch modules
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Add test for Torch Dynamo
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Add torch.dynamo test to qa
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Skip torch.compile test for <v2.0
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

* Fix bprop for cuDNN 8.9.3
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* Update cuDNN version requirement to 8.9.3
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* debug paddle CI; force LD_LIBRARY
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* debug paddle CI; force LD_LIBRARY to /opt
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove debug info for paddle
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* change cudnn requirement to 8.9.1 for v1 and 8.9.0 for v2; add batch size 32 for unit test; add LD library path for paddle tests temporarily
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove printf line in fused_attn.cpp
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* add batch size 32 for unit test
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* update cudnn-frontend to 0.9.2
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove temporary LD library path used for testing pre-released cudnn 8.9.3
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* Fix FP32 LayerNorm ONNX export

When running inference use a fwd method that is registered with torchscript.
Signed-off-by: Neta Zmora <nzmora@nvidia.com>

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

---------
Signed-off-by: Neta Zmora <nzmora@nvidia.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

* add more ops
Signed-off-by: Shijie Wang <jaywan@nvidia.com>

* add skipif
Signed-off-by: Shijie Wang <jaywan@nvidia.com>

* fix bug
Signed-off-by: Shijie Wang <jaywan@nvidia.com>

* minor change
Signed-off-by: Shijie Wang <jaywan@nvidia.com>

* minor change on coding style
Signed-off-by: Shijie Wang <jaywan@nvidia.com>

---------
Signed-off-by: Shijie Wang <jaywan@nvidia.com>

Signed-off-by: Gal Hubara Agam <ghubaraagam@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Add cast_transpose

Add gelu, gelu_fp8

Add cast_transpose_bgrad_dgelu

Add layernorm_fwd and layernorm_fwd_fp8

Add layernorm_bwd
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fix missing header
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

---------
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* add long sequence support and unify three backends for fused attention
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* update cudnn-frontend to v0.9.1
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* replace cpu_float2half_rn with __float2half_rn
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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

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

* make cudnn plan caches thread_local
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* replace cuDNN throw with NVTE_CHECK
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix replacement of cuDNN throw with NVTE_CHECK
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* force dropout probablity to 0 in inference mode
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* change negInfinity to be consistent with m512 fused attn
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove float2half conversion for scale_dropout
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* add back runtime api for sm detection
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* add gemm3 to enums FP8Fwd/BwdTensors
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* change dropout from no to yes for fmha_v1
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove output_rng_state in m512 kernels
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix elts_per_thread calculation in kvpacked fwd
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove dropout=0.0 restriction for m512 fused attn
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove output_rng_state completely from m512 kernels
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: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

Same filename is used for all activations, file is getting overridden to last activation type
Signed-off-by: Gal Hubara Agam <ghubaraagam@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Enable fused attention dropout
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Cast the uint32 key/counter to int64
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Update dropout support in fused attention docs
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Revise devPtrCuSeqlen* to align the naming
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Support different Jax PRNG impls
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Revert CastAsync since it is not used
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Implement is_training for 16-bit fused attn
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Add fused attn with dropout sanity unit tests
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Enhance the comments readability and rng_state checker
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Change the attention dropout shape to align other frameworks
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Make encoder tests deterministic
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Change the default seed for the jax encoder tests
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Maintain offset in TE
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Enhance the resource safety
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Revert rng_state type to allow only i64
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Handle the corner case for elts_per_threads calculation
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Populate rng state by kernels
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Rename rng_state as seed in cpp_extensions
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Update the attention dropout comment
Signed-off-by: Reese Wang <rewang@nvidia.com>

---------
Signed-off-by: Reese Wang <rewang@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Add self_attn_mask_type and replace attn_type
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Refine the keyword style for the better readability
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Replace attn_type with attn_mask_type in praxis transformer
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Fix typos
Signed-off-by: Reese Wang <rewang@nvidia.com>

---------
Signed-off-by: Reese Wang <rewang@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Fix softmax ONNX export

* BF16 is validated using "fake i/o": ie. instead of using BF16 as input/output, use FP32 input/output and convert to/from BF16 in the forward method.

* Wrap softmax symbolic functions with conversion to/from FP32 to produce the same semantics as TE's softmax (compute is performed at FP32 precision regardless of input/output data type).
Signed-off-by: Neta Zmora <nzmora@nvidia.com>

* ONNX export Code refactoring

Share function compute_in_fp32 between softmax.py (softmax symbolic functions) and te_onnx_extensions.py (the rest of the symbolic functions).
Signed-off-by: Neta Zmora <nzmora@nvidia.com>

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

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

---------
Signed-off-by: Neta Zmora <nzmora@nvidia.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

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

* Export reglu, geglu, swiglu
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

* Added ReLU and GLU variants to common
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

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

* PyTorch C++ lint
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

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

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

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

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

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

* Review comments
Co-authored-by: Przemyslaw Tredak <ptrendx@gmail.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

Move jax.experimental.maps.Mesh to jax.sharding.Mesh
Signed-off-by: Reese Wang <rewang@nvidia.com>

* Only use one gpu for tensorflow tests
Signed-off-by: kaixih <kaixih@nvidia.com>

* Simplify the change
Signed-off-by: kaixih <kaixih@nvidia.com>

* Final fix
Signed-off-by: kaixih <kaixih@nvidia.com>

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

* Use the same default in the function to what the class default.
Signed-off-by: Frederic Bastien <fbastien@nvidia.com>

* Assert instead of silently ignoring not supported variation. Small doc correction, amax_compute_algo is partially supported.
Signed-off-by: Frederic Bastien <fbastien@nvidia.com>

* Fix line lenght to fix the CI.
Signed-off-by: Frederic Bastien <fbastien@nvidia.com>

* Apply suggestions from code review
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Signed-off-by: Frédéric Bastien <frederic.bastien@gmail.com>

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

* Clarify that it is only TE/JAX that don't support that faeture.
Signed-off-by: Frederic Bastien <fbastien@nvidia.com>

* Update transformer_engine/jax/fp8.py
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Signed-off-by: Frédéric Bastien <frederic.bastien@gmail.com>

* Update the test following the change in default value
Signed-off-by: Frederic Bastien <fbastien@nvidia.com>

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

---------
Signed-off-by: Frederic Bastien <fbastien@nvidia.com>
Signed-off-by: Frédéric Bastien <frederic.bastien@gmail.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* First step of PaddlePaddle integration
- Add build option for paddle
- Add basic test framework
- Add 3 basic operators: cast_from_fp8, cast_to_fp8, gemm
Signed-off-by: Tian Zheng <tizheng@nvidia.com>
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fix review comments
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Support paddle build
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Add paddle build support for new building framework
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fix review comments
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Clean up build process for Paddle stub file
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Minor fixes
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fix pylint "wrong-import-order" warning
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Fix review comments
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

* Skip BF16 GEMM tests for unsupported arch
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

---------
Signed-off-by: Tian Zheng <tizheng@nvidia.com>
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>
Signed-off-by: Tim Moon <tmoon@nvidia.com>
Co-authored-by: Tim Moon <tmoon@nvidia.com>
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* add bf16 subgraph tests
Signed-off-by: Asfiya Baig <asfiyab@nvidia.com>

* changes:
1. Add normal mode BF16 tests for all subgraphs
2. Add fake BF16 tests for low-level subgraphs
3. Separate IO serialization from validation
Signed-off-by: Asfiya Baig <asfiyab@nvidia.com>

* ONNX export test - BF16 support part 1

TE infer returns torch.tensor, to support output of bf16 which is
currently not supported in numpy
Signed-off-by: Gal Hubara Agam <ghubaraagam@nvidia.com>

* ONNX export test - BF16 support part 2

- Separate TE infer from serialize
- Fix serialize function to use full path
- Set unique filenames for fake bf16 (avoid overriding standard bf16)
- Remove overwriting fake_bf16_io value
Signed-off-by: Gal Hubara Agam <ghubaraagam@nvidia.com>

* Export test: Slight tolerance increase in test_export_gpt_generation

Causes sporadic failures ~1% of all runs
Signed-off-by: Gal Hubara Agam <ghubaraagam@nvidia.com>

* Remove GEMM fake-bf16 export test and patch to enable it
Signed-off-by: Gal Hubara Agam <ghubaraagam@nvidia.com>

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

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

---------
Signed-off-by: Asfiya Baig <asfiyab@nvidia.com>
Signed-off-by: Gal Hubara Agam <ghubaraagam@nvidia.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: Asfiya Baig <asfiyab@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>