GitLab

Signed-off-by: Frederic Bastien <fbastien@nvidia.com>

* fix mask conversion and rng_state
Signed-off-by: Shijie Wang <jaywan@nvidia.com>

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

* use CUB to do prefix sum
Signed-off-by: Shijie Wang <jaywan@nvidia.com>

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

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

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

* Debug merge errors
Signed-off-by: Tim Moon <tmoon@nvidia.com>

---------
Signed-off-by: Shijie Wang <jaywan@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>

* wgrad-accumulation patch for legacy megatron-LM
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

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

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

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

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

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

* add support for flexible qkv layout
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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

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

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

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

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

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

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

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

* sb3hd/bs3hd working on top of 3xsbhd/bshd/thd
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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

* remove qkvso_strides on torch side; cover it in generateQKVStrides
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* all 15 layouts pass
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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

* removed most debug info/clean up
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* add note to deprecate some qkv layouts
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix code for unit tests in test_fused_attn.py
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* remove a couple more comments
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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

* fix fp8 tests
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix onnx for core attn; not fixed
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove nvtx and add env var for workspace opt
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove testing for env var
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* replace zeros/zeros_like with empty/empty_like
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix nvtx marker name for _q_k_v API
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove sm80 when compiling for h100
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* add mapping from qkv layout to layout group and qkv format
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* clean up enums mapping and remove trailing spaces
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* simplify workspace opt control logic; only need env var
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix fp8 test, and minor modifications for other tests
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* avoid overwriting model configs in unit test
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* random fixes/improvements: get_qkv_format/etc, default values, docstrings, comments
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* change workspace opt logic back to FORCE_WORKSPACE_OPT
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix FP8 tests and generateStrides function
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix get_backend logic for max512/arbitrary
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix unit tests; need cleanup
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* clean up unit tests for layouts, and fix minor lint issue
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* minor tweaks for CI testing: onnx string issue and test fused attn first
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove one unsupported layout from max512 and add a check to qkvpacked API
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix te layer test; reduce test time
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* revert compiler option changes; add back sm80 for even h100
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove some unit tests or make them optional to reduce CI time
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove more unit tests temporarily
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove _q_k_v in naming and add NVTE_ERROR for FP8 Aux_CTX_Tensors size checks
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* add more deprecation notes
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove temp tests from last commit
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* remove prints from last commit
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove redundant contiguous()
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove thd->bs3hd user warning to avoid GPU sync
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* temporary fix for onnx issue; more fixes in PR 437
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove unused variables
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: Charlene Yang
Signed-off-by: cyanguwa <8636796+cyanguwa@users.noreply.github.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* [PyTorch] Implement GQA based on fused q, k, v projection. Additionally fixes #392
Signed-off-by: Markus Schnoes <markus.schnoes@gmx.de>

* [PyTorch] Extend parameters_split option in Linear and LayerNormLinear to support splitting with different sizes as required by unfused GQA.
Signed-off-by: Markus Schnoes <markus.schnoes@gmx.de>

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

* fix noop cat to bypass torch.cat and support uneven split
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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

* fix cuda graph due to noop_cat
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* remove the use of enumerate when possible
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

---------
Signed-off-by: Markus Schnoes <markus.schnoes@gmx.de>
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Co-authored-by: Markus Schnoes <markus.schnoes@gmx.de>

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

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

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

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

* Change scaling factor from E8M0 to E8M23
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

* add flash implementation with context parallelism
Signed-off-by: xren <xren@nvidia.com>

* next more comments
Signed-off-by: xren <xren@nvidia.com>

* code comment fix
Signed-off-by: xren <xren@nvidia.com>

* comment fix
Signed-off-by: xren <xren@nvidia.com>

* add missing space
Signed-off-by: xren <xren@nvidia.com>

* fix docstrings
Signed-off-by: xren <xren@nvidia.com>

* try to add fa v2 api
Signed-off-by: xren <xren@nvidia.com>

* fix a comment
Signed-off-by: xren <xren@nvidia.com>

* fix padded kv return
Signed-off-by: xren <xren@nvidia.com>

* add docstrings of context parallelism
Signed-off-by: xren <xren@nvidia.com>

* minor fix
Signed-off-by: xren <xren@nvidia.com>

* minor docstring fix
Signed-off-by: xren <xren@nvidia.com>

* fix positional arguments
Signed-off-by: xren <xren@nvidia.com>

* make docstring line shorter
Signed-off-by: xren <xren@nvidia.com>

* add fa v2 backward api for flash_attn_with_cp
Signed-off-by: xren <xren@nvidia.com>

* remove redundant code
Signed-off-by: xren <xren@nvidia.com>

* make sure hidden size per attn head is multiple of 8 for FA2
Signed-off-by: xren <xren@nvidia.com>

* remove an unnecessary assert check for FA2
Signed-off-by: xren <xren@nvidia.com>

* indention fix
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

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

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

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

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

* Eliminate amax_and_scale_update bubbles
Signed-off-by: rewang <rewang@nvidia.com>

* Add CUDA check
Signed-off-by: rewang <rewang@nvidia.com>

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

* Enable the model to be change precision between iterations
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

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

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

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

Signed-off-by: Sangkug Lym <slym@nvidia.com>

* add workspace optimization for arbitrary_seqlen fused attn
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* add use_workspace_opt to cudnn plan cache and fix workspace estimate
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* modify workspace opt logic; move zero fill to FP8 API only; other minor fixes
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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

* fix std string error when input is nullptr
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* Add = for required vs allowed workspace comparison
Signed-off-by: cyanguwa <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>

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

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

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

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

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

* Improve performance of stack backtrace
Signed-off-by: Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com>

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

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

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

Use the new API when it is available.
Signed-off-by: Frederic Bastien <fbastien@nvidia.com>

* 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>

* [JAX] Fix incorrect sharding when only enable FSDP.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* [JAX] Add WAR to memory misaligned issues of LN BWD.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* [JAX] Reuse sm_arch for avoiding duplicate code.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* [JAX] Support multiple sizes allocation in WorkspaceManager.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* [JAX] Use template and ariadic arguments to improve multple sizes allocator.
Signed-off-by: Ming Huang <mingh@nvidia.com>

---------
Signed-off-by: Ming Huang <mingh@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>

Catch misaligned address errors
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.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>

Signed-off-by: Sudhakar Singh <sudhakars@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>

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

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

* Change API and add to RMSNorm
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

---------
Signed-off-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 support for cuDNN fprop and FAv2 bprop
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* skip activation recompute tests if FAv2
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* restrict the use of FAv2 bprop to H100 only
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* move use_FAv2_bwd check to init
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove skipifs for FAv2 in test numerics
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix typos and wording for deterministic checks
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* Remove variables related to FAv2 skipifs
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Signed-off-by: cyanguwa <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>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

* fix linear sequence parallel when weight is frozen
Signed-off-by: jasonwan <jasonwan@nvidia.com>

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

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

* Initially commit for FSDP
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Adding support to fsdp xmap sharding
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Specify WeightHParamsCollection of fp8 meta.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Support partial FP8 custom calls with FSDP.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Adding amax reduction on the fsdp mesh dim.
Signed-off-by: Ming Huang <mingh@nvidia.com>

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

* Fix the wrong batch axis in logic_axis_rules and add sharding_constraint to BMM1
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Support FSDP in fMHA.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Fix missing all-reduce of wgrads along FSDP axis.
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Change default value of fsdp_axis_name to  for aligning with others
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Fix RuntimeError: with_sharding_constraint requires a non-empty
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Slightly changes (review feedback)
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Removed unnecessary comments
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Mergeing input_dp_dim into weight_fsdp_dim_map
Signed-off-by: Ming Huang <mingh@nvidia.com>

* Update transformer_engine/jax/sharding.py
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

---------
Signed-off-by: Ming Huang <mingh@nvidia.com>
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

* Disable FAv2 for deterministic use
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Also disable FusedAttention backend with deterministic
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

* Optimize calls to .cpu() during checkpointing
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Fixes for ONNX
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@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>

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