- 09 May, 2023 1 commit
-
-
Neta Zmora authored
* ONNX export refactoring * Remove infer_ort (to enable more testing) * Add BF16 ORT tests for Q/DQ ops and GELU. * Use FP32 i/o instead of BF16 (because ORT doesn't support BF16 i/o) and add casts from FP32 to BF16 (this is only for subgraph inputs and outputs). * We'll need to add more BF16 testing. * GEMM: * Add cast after DQ to achieve better performance (matmul at sub-fp32 precisions). * Fold bias into Gemm operation (=> smaller graphs) * Wrap GEMM-GELU with FP32 (TE implements GELU in FP32) * Enable tests for cross attention (test_export_multihead_attention) * Reduce test thresholds for test_export_layernorm_mlp, test_export_layernorm_linear, test_export_layernorm Signed-off-by:
Neta Zmora <nzmora@nvidia.com> * Loosen MHA export validation thresholds for FP16 Signed-off-by:
-
- 03 May, 2023 3 commits
-
-
Przemyslaw Tredak authored
* Use NVIDIA Sans font in Transformer Engine documentation Signed-off-by:
Przemek Tredak <ptredak@nvidia.com> * Use NV font for the names in the API Signed-off-by:
-
galagam authored
Signed-off-by:
Gal Hubara Agam <ghubaraagam@nvidia.com> Co-authored-by:
Kirthi Shankar Sivamani <ksivamani@nvidia.com>
-
Kirthi Shankar Sivamani authored
DDP support for no-bias option Signed-off-by:
-
- 02 May, 2023 3 commits
-
-
cyanguwa authored
* move dbias from input list to output list for bwd Signed-off-by:
Charlene Yang <8636796+cyanguwa@users.noreply.github.com> * split asserts into three for bias checks Signed-off-by:
Tim Moon <4406448+timmoon10@users.noreply.github.com> Signed-off-by:
-
Sangkug Lym authored
* using different strems for pushsend and pushrecv Signed-off-by:
Sangkug Lym <slym@nvidia.com> * fix stream dependency Signed-off-by:
-
Kirthi Shankar Sivamani authored
Use only built-ins for setup Signed-off-by:
-
- 01 May, 2023 1 commit
-
-
Kirthi Shankar Sivamani authored
Fixes in nested autocast Signed-off-by:
-
- 30 Apr, 2023 1 commit
-
-
Kirthi Shankar Sivamani authored
Remove distutils Signed-off-by:
-
- 29 Apr, 2023 2 commits
-
-
galagam authored
* Fixes to test_onnx_export when saving input and output tensors - Allow saving i/o tensors when onnxruntime inference is skipped - Support saving multiple outputs Signed-off-by:
-
cyanguwa authored
correct cuDNN version requirement Signed-off-by:Charlene Yang <charleney@nvidia.com>
-
- 28 Apr, 2023 4 commits
-
-
Neta Zmora authored
* iFix LN ONNX export When exporting LayerNorm make sure that the weights and bias inputs have the same type as the LN input. Also: * Add a regression test. * Add environment variable to override directory of generated test artifacts Signed-off-by:
-
Kirthi Shankar Sivamani authored
Fix bug in Sequence Parallelism Signed-off-by: -
Ming-Xu Huang authored
* Adjust Module Structure. 1. Collect Flax related modules to a sub-folder, flax. 2. Add a function to unify scale_init for zero-centered-gamma LN. Signed-off-by:
Ming Huang <mingh@nvidia.com> * Make changes be compatible to previous versions. Signed-off-by:
-
Kirthi Shankar Sivamani authored
Signed-off-by:
-
- 27 Apr, 2023 1 commit
-
-
Przemyslaw Tredak authored
* Remove the nonexistent parameter from fused attention documentation Signed-off-by:
-
- 26 Apr, 2023 2 commits
-
-
Kirthi Shankar Sivamani authored
* Recursively checkout submodules Signed-off-by:
-
Przemyslaw Tredak authored
* Faster split of QKV for FlashAttention Signed-off-by:
-
- 25 Apr, 2023 1 commit
-
-
Kirthi Shankar Sivamani authored
* Add guide to build from source Signed-off-by:
-
- 22 Apr, 2023 2 commits
-
-
cyanguwa authored
remove used function ternary_pw_op_create Signed-off-by:
-
asfiyab-nvidia authored
add cast for BF16 input/output types for Q/DQ ONNX ops Signed-off-by:
Asfiya Baig <asfiyab@nvidia.com> Co-authored-by:
-
- 21 Apr, 2023 4 commits
-
-
Kirthi Shankar Sivamani authored
Signed-off-by: -
Sangkug Lym authored
Signed-off-by:
-
cyanguwa authored
* Add FP8 fused attention to TE for PyTorch Signed-off-by:
-
Kirthi Shankar Sivamani authored
* Initial refactor; linker error Signed-off-by:
Tim Moon <tmoon@nvidia.com> * Lint Signed-off-by:
-
- 20 Apr, 2023 2 commits
-
-
Ming-Xu Huang authored
* Allow update_collections and update_fp8_metas to return both Dict and FrozenDict. Signed-off-by:
-
Frédéric Bastien authored
* Clean up the installation instruction. We where telling to install the dev version in the README. Signed-off-by:
Frederic Bastien <fbastien@nvidia.com> * Typos Signed-off-by:
-
- 19 Apr, 2023 1 commit
-
-
Kirthi Shankar Sivamani authored
* Port initial changes Co-authored-by:
Vasudevan Rengasamy <vrengasamy@nvidia.com> Signed-off-by:
-
- 18 Apr, 2023 3 commits
-
-
Frédéric Bastien authored
Signed-off-by: -
Sangkug Lym authored
* amax reduction internval Signed-off-by:
-
Tim Moon authored
Signed-off-by:
-
- 17 Apr, 2023 3 commits
-
-
Przemek Tredak authored
Signed-off-by: -
Kirthi Shankar Sivamani authored
* Add tests for cuda graph capture Signed-off-by:
-
Kirthi Shankar Sivamani authored
* use upstream flash-attn Signed-off-by:
-
- 15 Apr, 2023 1 commit
-
-
Kirthi Shankar Sivamani authored
* Fix gencode flags based on cuda version Signed-off-by:
-
- 14 Apr, 2023 1 commit
-
-
Kirthi Shankar Sivamani authored
Signed-off-by:
-
- 13 Apr, 2023 3 commits
-
-
Neta Zmora authored
* Fix model load exception when state resides on GPU - Whenever converting a torch.tensor to numpy, we need to first migrate the tensor storage to the host CPU. - Add a warning not to do contant-folding when exporting to ONNX. This is due to a torch.onnx export bug. - Refactor compare_outputs Signed-off-by:
-
zlsh80826 authored
* Add zero_center_gamma/functional pass Signed-off-by:
Reese Wang <rewang@nvidia.com> * Add zero_centered_gamma for fp8_ln_mlp Signed-off-by:
-
Kaixi Hou authored
Remove the autocast_variable Signed-off-by:kaixih <kaixih@nvidia.com>
-
- 11 Apr, 2023 1 commit
-
-
Trevor Morris authored
* Fix pybind11 install doc Signed-off-by:
Trevor Morris <tmorris@nvidia.com> * Set CMAKE_PREFIX_PATH for TF to find pybind11 Signed-off-by:
-
