- 16 Jan, 2024 1 commit
-
-
zlsh80826 authored
* Support num_gqa_groups arguments Signed-off-by:
Reese Wang <rewang@nvidia.com> * Add GQA support on the JAX bridge code Signed-off-by:
-
- 12 Jan, 2024 2 commits
-
-
Tian Zheng authored
* Actively free tensor in bwd Signed-off-by:
Tian Zheng (Engrg-Hardware 1) <tizheng@nvidia.com> * - Add inplace support for fp8 casting - Allow skipping weight update in fp8 meta update Signed-off-by:
-
Ming-Xu Huang authored
* Adding Cast custom call Signed-off-by:
Ming Huang <mingh@nvidia.com> * Applying cast to the kernel of layernorm_fp8_dot Signed-off-by:
-
- 11 Jan, 2024 1 commit
-
-
Tian Zheng authored
* Add SP for linear Signed-off-by:
Kirthi Shankar Sivamani <ksivamani@nvidia.com> --------- Signed-off-by:
-
- 10 Jan, 2024 2 commits
-
-
Zhang Haitao authored
* support non-tensor inputs/outputs for checkpoint Signed-off-by:
skydoorkai <htsantaclara@163.com> * better format Signed-off-by:
-
Xiaowei Ren authored
* 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:
cyanguwa <8636796+cyanguwa@users.noreply.github.com>
-
- 08 Jan, 2024 3 commits
-
-
Tim Moon authored
* 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:
-
Jaemin Choi authored
* 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:
-
cyanguwa authored
fix FP8 dims Signed-off-by:
-
- 06 Jan, 2024 1 commit
-
-
Kirthi Shankar Sivamani authored
* Deterministic FA, bump minimum supported version Signed-off-by:
-
- 05 Jan, 2024 1 commit
-
-
Przemyslaw Tredak authored
Signed-off-by:Przemek Tredak <ptredak@nvidia.com>
-
- 03 Jan, 2024 3 commits
-
-
Przemyslaw Tredak authored
Signed-off-by:
-
Sangkug Lym authored
* 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:
-
Przemyslaw Tredak authored
Signed-off-by:
-
- 02 Jan, 2024 1 commit
-
-
Hongbin Liu authored
avoid redundant computation for cu_seqlens Signed-off-by:
Hongbin Liu <hongbinl@nvidia.com> Co-authored-by:
-
- 18 Dec, 2023 1 commit
-
-
Alp Dener authored
* 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:
-
- 16 Dec, 2023 2 commits
-
-
cyanguwa authored
* add sliding window to FA Signed-off-by:
-
Tim Moon authored
* Update fp8_meta amax when copying into Float8Tensor Signed-off-by:
-
- 15 Dec, 2023 3 commits
-
-
Przemyslaw Tredak authored
* Disable dynamo for Fused Attention Signed-off-by:
-
Kunlun Li authored
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>
-
Fabian Joswig authored
[fix] fixed micro batched inference with RoPE Signed-off-by:
Fabian Joswig <fabian.joswig@deepl.com> Co-authored-by:
-
- 14 Dec, 2023 1 commit
-
-
Alp Dener authored
applied Google-advised fix to register custom op primitives with the device dispatch list Signed-off-by:
-
- 13 Dec, 2023 2 commits
-
-
cyanguwa authored
* fix backend selection for sm80 Signed-off-by:
-
Marks101 authored
Signed-off-by:Markus Schnoes <markus.schnoes@gmx.de>
-
- 12 Dec, 2023 2 commits
-
-
cyanguwa authored
disable pylint bare except Signed-off-by: -
cyanguwa authored
* fix onnx/dynamo error Signed-off-by:
Przemyslaw Tredak <ptrendx@gmail.com> Signed-off-by:
-
- 07 Dec, 2023 2 commits
-
-
cyanguwa authored
* Integrate cuDNN frontend v1 to fused attention and miscellaneous fixes Signed-off-by:
-
Tim Moon authored
* Float8Tensor uses cached transpose if available Signed-off-by:
-
- 04 Dec, 2023 2 commits
-
-
Marks101 authored
* [PyTorch] TransformerLayer: add parallel_attention_mlp to support Falcon models Signed-off-by:
-
zlsh80826 authored
Add checkpoint_name Signed-off-by:
-
- 01 Dec, 2023 3 commits
-
-
Tim Moon authored
Fix incorrect variable name in LayerNormMLP backward Signed-off-by: -
LadyRick authored
[PyTorch] fix amax calculate during fp8 calibration Signed-off-by:ladyrick <ladyrick@qq.com>
-
zlsh80826 authored
* Add rng_state output for cross fused attention Signed-off-by:
-
- 30 Nov, 2023 3 commits
-
-
Deepak Narayanan authored
wgrad should be zero'ed out if a weight parameter is shared among multiple layers Signed-off-by:Deepak Narayanan <dnarayanan@nvidia.com>
-
zlsh80826 authored
Support layernorm sm_margin through environment variables Signed-off-by: -
Ming-Xu Huang authored
Use relative idx to ScaledUpperTriangMaskedSoftmaxFwdPrimitive.abstract to support batching. Signed-off-by:
-
- 28 Nov, 2023 3 commits
-
-
Marks101 authored
* [PyTorch] Linear: fix computation for wgrad if sequence_parallel=True Signed-off-by:
-
Deepak Narayanan authored
Getting warnings of the following form using ToT TE: ``` /usr/local/lib/python3.10/dist-packages/transformer_engine/pytorch/attention.py:852: UserWarning: TypedStorage is deprecated. It will be removed in the future and UntypedStorage will be the only storage class. This should only matter to you if you are using storages directly. To access UntypedStorage directly, use tensor.untyped_storage() instead of tensor.storage() data_ptr = grad_outputs[0].storage().data_ptr() ``` Signed-off-by:Deepak Narayanan <2724038+deepakn94@users.noreply.github.com>
-
Tim Moon authored
Use unsigned char for byte dtype Signed-off-by:
-
- 23 Nov, 2023 1 commit
-
-
Tian Zheng authored
* Add async TP comm Signed-off-by:
-
