GitLab

Fixed offloading for PyT version/ Added Attention activation offloading support/ Native FP8 support (#632)

* Fixed offloading for PyT version/ Added Attention activation offloading support/ Native FP8 support
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

* Removed activation offloading for fused attention
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

* Fixed the illegal memory access issue for activation offloading of attention
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

* Removed the version guard
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

* Pipeline failures fix
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

* Fixed lint erros
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

* Lint error fix
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

---------
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>
Co-authored-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

Fix MHA docstring
Signed-off-by: Isaac Ong <isaacong.jw@gmail.com>

* fused apply rope
Signed-off-by: Xin Yao <xiny@nvidia.com>

* Apply suggestions from code review
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Signed-off-by: Xin Yao <yaox12@outlook.com>

* resolve comments
Signed-off-by: Xin Yao <xiny@nvidia.com>

* make rotary_percent optional
Signed-off-by: Xin Yao <xiny@nvidia.com>

* fix ci
Signed-off-by: Xin Yao <xiny@nvidia.com>

* fix test
Signed-off-by: Xin Yao <xiny@nvidia.com>

* add rope test to qa
Signed-off-by: Xin Yao <xiny@nvidia.com>

* fix linting
Signed-off-by: Xin Yao <xiny@nvidia.com>

* sync apex: add transpose_output_memory
Signed-off-by: Xin Yao <xiny@nvidia.com>

* small fix
Signed-off-by: Xin Yao <xiny@nvidia.com>

* sync apex: fuse sin/cos
Signed-off-by: Xin Yao <xiny@nvidia.com>

* sync apex: fused rope for thd format
Signed-off-by: Xin Yao <xiny@nvidia.com>

* fix lint
Signed-off-by: Xin Yao <xiny@nvidia.com>

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

* add support for bshd format
Signed-off-by: Xin Yao <xiny@nvidia.com>

* support different seq length
Signed-off-by: Xin Yao <xiny@nvidia.com>

* update
Signed-off-by: Xin Yao <xiny@nvidia.com>

* update copyright
Signed-off-by: Xin Yao <xiny@nvidia.com>

* remove transpose_output_memory
Signed-off-by: Xin Yao <xiny@nvidia.com>

* Make outputs contiguous in SBHD case
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

---------
Signed-off-by: Xin Yao <xiny@nvidia.com>
Signed-off-by: Xin Yao <yaox12@outlook.com>
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Co-authored-by: Przemyslaw Tredak <ptredak@nvidia.com>

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

[PyTorch] fix forward attention_type in MultiheadAttention
Signed-off-by: Markus Schnoes <markus.schnoes@gmx.de>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

[PyTorch] Workaround for incorrect output from torch.cuda.is_bf16_compatible() on V100s and TU102s (#626)

* replaced torch.cuda.is_bf16_compatible() with explicit sm_80 check via torch.cuda.get_device_capability()
Signed-off-by: Alp Dener <adener@nvidia.com>

* implement te.utils.is_bf16_compatible() to replace torch.cuda counterpart
Signed-off-by: Alp Dener <adener@nvidia.com>

---------
Signed-off-by: Alp Dener <adener@nvidia.com>

* added missing parameter materialization on real device for LayerNorm and RMSNorm
Signed-off-by: Alp Dener <adener@nvidia.com>

* added new unittest for deferred initialization and modified parameter materialization to support standalone execution outside of FSDP
Signed-off-by: Alp Dener <adener@nvidia.com>

* restored tensor parallel attributes that were being wiped out by the parameter reset
Signed-off-by: Alp Dener <adener@nvidia.com>

* fixed incorrect order of fp8 metadata initialization
Signed-off-by: Alp Dener <adener@nvidia.com>

* added deferred init unittest to the QA script
Signed-off-by: Alp Dener <adener@nvidia.com>

---------
Signed-off-by: Alp Dener <adener@nvidia.com>

Signed-off-by: Markus Schnoes <markus.schnoes@gmx.de>

Activation offloading to CPU's for the Linear, Layernorm Linear and the Layernorm MLP modules (#571)

* Added support activation offloading to CPU's
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

* Moving CPU offloading library to TE
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

* Restructured code, added switch to choose between weight/activation offloading
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

* Removed arg during constructor
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

* Fix nit-pick errors
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

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

* Fix to the code block in docs
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

* Added offloading unit test
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

* Fixed formatting
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

* wgrad fusion fix, minor errors and lint
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

* Fixed stray PyT tensors in LayernormMLP getting offloaded
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

* Fixed typi
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>

* Fix offloading for rmsnorm, rm test
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

* Float8Tensor compatible offloading
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

---------
Signed-off-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: Selvaraj Anandaraj <selvaraja@login-eos01.eos.clusters.nvidia.com>
Co-authored-by: Przemyslaw Tredak <ptredak@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

fix failing tests due to PR #557
Signed-off-by: Sudhakar Singh <sudhakars@nvidia.com>
Co-authored-by: cyanguwa <8636796+cyanguwa@users.noreply.github.com>

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

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

* make TransformerLayer accept a `bshd` or `sbhd` tensor format
Signed-off-by: Sudhakar Singh <sudhakars@nvidia.com>

* Fixes from feedback
Signed-off-by: Sudhakar Singh <sudhakars@nvidia.com>

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

* remove incorrect info from docstring
Signed-off-by: Sudhakar Singh <sudhakars@nvidia.com>

* fix from feedback
Signed-off-by: Sudhakar Singh <sudhakars@nvidia.com>

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

* Implemented deferred initialization via `device='meta'` option for te.Linear and added new PyTorch example to demonstrate its use with FullyShardedDataParallel execution.
Signed-off-by: Alp Dener <adener@nvidia.com>

* correcting Float8Tensor initialization and fixing linting errors
Signed-off-by: Alp Dener <adener@nvidia.com>

* removed duplicate code from upstream rebase, local tests passing
Signed-off-by: Alp Dener <adener@nvidia.com>

* improved comments/documentation for FSDP example
Signed-off-by: Alp Dener <adener@nvidia.com>

* converted reset_parameters() into a base module function
Signed-off-by: Alp Dener <adener@nvidia.com>

* fixed Float8Tensor creation with deferred init, all tests passing locally
Signed-off-by: Alp Dener <adener@nvidia.com>

* extended deferred initialization to all TE modules
Signed-off-by: Alp Dener <adener@nvidia.com>

* fixed linting errors
Signed-off-by: Alp Dener <adener@nvidia.com>

* removed unnecessary reference to the parent module of parameter, added clarifying comments in parameter reset
Signed-off-by: Alp Dener <adener@nvidia.com>

---------
Signed-off-by: Alp Dener <adener@nvidia.com>

* support non-tensor inputs/outputs for checkpoint
Signed-off-by: skydoorkai <htsantaclara@163.com>

* better format
Signed-off-by: skydoorkai <htsantaclara@163.com>

* modify to avoid python loops
Signed-off-by: skydoorkai <htsantaclara@163.com>

---------
Signed-off-by: skydoorkai <htsantaclara@163.com>

* try to use cuDNN fused attention for context parallelism
Signed-off-by: xren <xren@nvidia.com>

* assert CP is only supported with NVTE_F16_arbitrary_seqlen
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* port fused attn api to context parallelism
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* add one more assert
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* assert CP does not support padded tokens
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* add qkv_format into CP implementation
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* remove qkv_format from CP function
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

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

* fix bwd error with FA v2
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* make cp implementation support non-causal masking
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

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

* remove redundant asserts for CP
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* minor assert information change
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* assert core attn bias has not been supported with CP yet
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* make CP work with window_sizes of [-1, -1] and [-1, 0]
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* add draft code for fa test with cp
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* move fused attn test to a specific folder
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* add assert_close to flash attn cp test
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* add more tests for CP
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* add optional arguments for FA v2.4+
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* minor change
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* add skip condition for CP test
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* class and function naming fix
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

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

* do not use fused attn if backend does not work with CP
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* create a separate folder for CP test as it needs multi-GPUs
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

* add attn_mask_type check in attn_forwrad_func_with_cp
Signed-off-by: Xiaowei Ren <xren@nvidia.com>

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

---------
Signed-off-by: xren <xren@nvidia.com>
Signed-off-by: Xiaowei Ren <xren@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: cyanguwa <8636796+cyanguwa@users.noreply.github.com>

* Refactor parameter split in Linear module

Remove module state from noop_cat. Support arbitrary names in parameter split. Handle tensor parallelism.
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Make noop_cat a standalone operation
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Update parameter splits in LayerNormLinear
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Debug case without bias

Fix pylint complaints.
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Remove unused import
Signed-off-by: Tim Moon <tmoon@nvidia.com>

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

* Use jit_fuser for bias-dropout-add fusion
Signed-off-by: Jaemin Choi <jaeminc@nvidia.com>

* Use jit_fuser for CP FA kernel
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

* Deterministic FA, bump minimum supported version
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

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

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

* Address review comments
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>

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

* Provide pre-computed max sequence to remove unnecessary kernels and D2H copies
Signed-off-by: Sangkug Lym <slym@nvidia.com>

* Tweak comments
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

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

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

avoid redundant computation for cu_seqlens
Signed-off-by: Hongbin Liu <hongbinl@nvidia.com>
Co-authored-by: Hongbin Liu <hongbinl@nvidia.com>
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

* Linear and LayerNormLinear weight and bias buffer cleanup at the end of init when there is no parameter split
Signed-off-by: Alp Dener <adener@nvidia.com>

* fixed typo in tensor name
Signed-off-by: Alp Dener <adener@nvidia.com>

* fixed typo in tensor name
Signed-off-by: Alp Dener <adener@nvidia.com>

---------
Signed-off-by: Alp Dener <adener@nvidia.com>

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

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

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

* change bert test to causal as unfused does not support padding
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix FlashAttention for v2-2.3 versions
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* verify FA swa works
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix mask related restrictions and duplicate code after merge
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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

* move repeated code into a function
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* add determinism filter and fix FA warning message
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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

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

* fix indent
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>

* Update fp8_meta amax when copying into Float8Tensor
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Avoid amax when copying between Float8Tensors with fp8_metas
Signed-off-by: Tim Moon <tmoon@nvidia.com>

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

* Disable dynamo for Fused Attention
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>

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

---------
Signed-off-by: Przemek Tredak <ptredak@nvidia.com>
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

Avoid redeclaration error of "nv_bfloat16" when Compiling in CUDA12.1.
Error log: /usr/local/cuda/include/cuda_fp16.hpp(2724): error: invalid redeclaration of type name "nv_bfloat16" (declared at line 2837 of /usr/local/cuda/include/cuda_bf16.hpp)
Signed-off-by: Kunlun Li <94586211+kunlunl@users.noreply.github.com>

[fix] fixed micro batched inference with RoPE
Signed-off-by: Fabian Joswig <fabian.joswig@deepl.com>
Co-authored-by: cyanguwa <8636796+cyanguwa@users.noreply.github.com>

Signed-off-by: Markus Schnoes <markus.schnoes@gmx.de>

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

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

* move changes to pytorch/__init__ using try/except
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* Update transformer_engine/pytorch/__init__.py
Co-authored-by: Przemyslaw Tredak <ptrendx@gmail.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: Przemyslaw Tredak <ptrendx@gmail.com>

* Integrate cuDNN frontend v1 to fused attention and miscellaneous 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 jax/paddle for unit tests
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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

* fix and/or logic in get_backend
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix flag_max512 and test_numerics
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove v.contiguous() since get_qkv_layout covers it
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* skip fp8 tests for sm89
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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

* revert mask type to comma-separated list
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 last two commits
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* integrate v1/pre-release-5
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* cleanup prerelease5 integration and fix FA2.1 commit
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* force dropout to 0 if not training
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* testing bias/alibi and padding+causal; add alibi to unfused DPA
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* set flag_arb to false when non determinism is not allowed
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* followup on prev commit; remove redundant python env var setting
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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

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

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

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

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

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

* remove use of non-deter var for backend selection
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* fix workspace size in bwd and uncomment bias test
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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

* remove workspace_opt from FADescriptor_v1
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* disable arbitrary backend + post scale bias in Jax; waiting on PR 525
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* clean up bhsd order
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* swap bias/rng_state order in aux_ctx_tensor and add bias to aux_ctx_tensor in _qkvpacked/_kvpacked API
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove support for padding_causal + cross for max512
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* change alibi bias to float32 for bias_1_4/5 tests
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* fix thd fwd output shape for FlashAttention and add backend info for DPA
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix definition of workspace limit when dbias is present
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* further tweak DP_WORKSPACE_LIMIT definition
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* disallow alibi+no_mask for sdpa flash and update alibi tests
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* update jax/paddle after PR525 and fix DP_WORKSPACE_LIMIT for dbias Jax tests
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* disable dbias for non-hopper archs
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

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

* remove build dir in setup.py
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* change selection logic to prefer fused attn on sm90
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* fix h and s order in header
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* update to cudnn fe v1 branch
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove manual setting of workopt path due to dbias after v1 update
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

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

* add post_scale_bias and alibi to sdpa flash support matrix
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix support matrix in header files
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* move headers back to .cu and change seed/offset to int64
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* update Megatron commit in L1 test and remove all prints in fused attn test
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix L1 Megatron test
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix fp8 arg in L1 Megatron script
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* print only when debug flag is on
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove checkpointing loading to avoid loading other tests results
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>

* Float8Tensor uses cached transpose if available
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Fix bug with non-2D transpose
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Custom pickling for Float8Tensor
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Debug test for pickling Float8Tensor
Signed-off-by: Tim Moon <tmoon@nvidia.com>

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

* Review suggestions from @sudhakarsingh27

Avoid FP8 casts when copying between Float8Tensors. Make make_like a class function.
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* Add unit test for checkpointing model with FP8 params

Debugged pickling and copy functions.
Signed-off-by: Tim Moon <tmoon@nvidia.com>

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

* [PyTorch] TransformerLayer: add parallel_attention_mlp to support Falcon models
Signed-off-by: Markus Schnoes <markus.schnoes@gmx.de>

* [PyTorch] add test for parallel_attention_mlp to test_numerics
Signed-off-by: Markus Schnoes <markus.schnoes@gmx.de>

* [PyTorch] TorchGPT: fix dropout for parallel_attention_mlp

Now uses nn.functional.dropout because depending on the path there are one or two dropouts.
Signed-off-by: Markus Schnoes <markus.schnoes@gmx.de>

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

* [PyTorch] test_gpt_accuracy: fix spelling in construction of TorchGPT
Signed-off-by: Markus Schnoes <markus.schnoes@gmx.de>

---------
Signed-off-by: Markus Schnoes <markus.schnoes@gmx.de>
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

Fix incorrect variable name in LayerNormMLP backward
Signed-off-by: Tim Moon <tmoon@nvidia.com>

[PyTorch] fix amax calculate during fp8 calibration
Signed-off-by: ladyrick <ladyrick@qq.com>

wgrad should be zero'ed out if a weight parameter is shared among multiple layers
Signed-off-by: Deepak Narayanan <dnarayanan@nvidia.com>

* [PyTorch] Linear: fix computation for wgrad if sequence_parallel=True
Signed-off-by: Markus Schnoes <markus.schnoes@gmx.de>

* Remove buggy gather_along_last_dim
Signed-off-by: Tim Moon <tmoon@nvidia.com>

* [PyTorch] Linear: fix line length
Signed-off-by: Markus Schnoes <markus.schnoes@gmx.de>

* Simplify logic for saving input tensor for Linear backward
Signed-off-by: Tim Moon <tmoon@nvidia.com>

---------
Signed-off-by: Markus Schnoes <markus.schnoes@gmx.de>
Signed-off-by: Tim Moon <tmoon@nvidia.com>
Co-authored-by: Tim Moon <tmoon@nvidia.com>