- 26 Jul, 2023 1 commit
-
-
Kirthi Shankar Sivamani authored
Signed-off-by:Kirthi Shankar Sivamani <ksivamani@nvidia.com>
-
- 20 Jul, 2023 2 commits
-
-
Shijie authored
* add flash attn tests Signed-off-by:
Shijie Wang <jaywan@nvidia.com> * update flash attn Signed-off-by:
-
Kirthi Shankar Sivamani authored
Signed-off-by:
-
- 19 Jul, 2023 4 commits
-
-
Tian Zheng authored
* 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:
-
Kirthi Shankar Sivamani authored
* FA v2.0 support Signed-off-by:
-
Kirthi Shankar Sivamani authored
Signed-off-by: -
Kirthi Shankar Sivamani authored
* FA does not support head_dim > 64 on Ada Signed-off-by:
-
- 18 Jul, 2023 2 commits
-
-
zlsh80826 authored
* 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:
-
Kirthi Shankar Sivamani authored
* layer number bug fixes Signed-off-by:
-
- 15 Jul, 2023 1 commit
-
-
Tim Moon authored
* Disable TorchDynamo optimizations in PyTorch modules Signed-off-by:
Tim Moon <tmoon@nvidia.com> * Add test for Torch Dynamo Signed-off-by:
Tim Moon <4406448+timmoon10@users.noreply.github.com> Signed-off-by:
-
- 14 Jul, 2023 3 commits
-
-
Kirthi Shankar Sivamani authored
Bug fix for checkpointing Signed-off-by: -
cyanguwa authored
* 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:
-
Kirthi Shankar Sivamani authored
* Deprecate unused APIs Signed-off-by:
-
- 13 Jul, 2023 4 commits
-
-
Kirthi Shankar Sivamani authored
Signed-off-by: -
Neta Zmora authored
* 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 authored
Remove extra buffers Signed-off-by: -
Kirthi Shankar Sivamani authored
* Better dimension assert for FP8 Signed-off-by:
-
- 07 Jul, 2023 1 commit
-
-
Ming-Xu Huang authored
Signed-off-by:Ming Huang <mingh@nvidia.com>
-
- 06 Jul, 2023 1 commit
-
-
Shijie authored
* add more ops Signed-off-by:
-
- 01 Jul, 2023 1 commit
-
-
Tim Moon authored
Signed-off-by:
-
- 30 Jun, 2023 1 commit
-
-
Tejaswin Parthasarathy authored
fix : TE virtuelenv discovery Signed-off-by:tejaswinp <tejaswinp@nvidia.com>
-
- 26 Jun, 2023 2 commits
-
-
Kirthi Shankar Sivamani authored
* Get default dtype from pytorch Signed-off-by:
-
Kirthi Shankar Sivamani authored
Signed-off-by:
-
- 23 Jun, 2023 1 commit
-
-
Neta Zmora authored
* Fix ONNX export of layer_norm ONNX has a spec bug: ConstantOfShape supports all dtypes except for BF16. To WAR we use dtype FP32 and then cast to BF16. Will also issue a PR to the ONNX sig committee to change the spec in opset 20. Signed-off-by:
-
- 22 Jun, 2023 2 commits
-
-
Tian Zheng authored
* 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:
-
cyanguwa authored
* add long sequence support and unify three backends for fused attention Signed-off-by:
-
- 21 Jun, 2023 1 commit
-
-
asfiyab-nvidia authored
Signed-off-by:Asfiya Baig <asfiyab@nvidia.com>
-
- 20 Jun, 2023 3 commits
-
-
Kirthi Shankar Sivamani authored
Signed-off-by: -
zlsh80826 authored
* Enable fused attention dropout Signed-off-by:
-
zlsh80826 authored
* Add self_attn_mask_type and replace attn_type Signed-off-by:
-
- 16 Jun, 2023 2 commits
-
-
Neta Zmora authored
* 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:
-
Kirthi Shankar Sivamani authored
* ReLU ONNX export Signed-off-by:
-
- 13 Jun, 2023 2 commits
-
-
Kirthi Shankar Sivamani authored
Signed-off-by: -
Przemyslaw Tredak authored
* Added ReLU and GLU variants to common Signed-off-by:
Przemek Tredak <ptredak@nvidia.com> * pyTorch changes Signed-off-by:
Przemyslaw Tredak <ptrendx@gmail.com> Signed-off-by:
-
- 12 Jun, 2023 1 commit
-
-
Tim Moon authored
This reverts commit 914f3841 . Signed-off-by:
-
- 11 Jun, 2023 1 commit
-
-
asfiyab-nvidia authored
* fix BF16 onnx export for ort verification Signed-off-by:
asfiyab-nvidia <117682710+asfiyab-nvidia@users.noreply.github.com> --------- Signed-off-by:
-
- 08 Jun, 2023 1 commit
-
-
Kaixi Hou authored
* Only use one gpu for tensorflow tests Signed-off-by:
kaixih <kaixih@nvidia.com> * Simplify the change Signed-off-by:
-
- 07 Jun, 2023 2 commits
-
-
Kirthi Shankar Sivamani authored
* Use torch.compile for version 2.0 and higher Signed-off-by:
-
Frédéric Bastien authored
* 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:
Frédéric Bastien <frederic.bastien@gmail.com> * grammar Signed-off-by:
-
- 06 Jun, 2023 1 commit
-
-
Tian Zheng authored
* 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:
-
