- 31 Aug, 2025 1 commit
-
-
vasunvidia authored
* Add dropout kernel with 8-bit RNG Co-authored-by:
Vasudevan Rengasamy <vrengasamy@nvidia.com> Co-authored-by:
Tim Moon <tmoon@nvidia.com> Signed-off-by:
pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
-
- 29 Aug, 2025 3 commits
-
-
buptzyb authored
* support bf16+fp8 model Signed-off-by:
Robin Zhang <robinz@nvidia.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * update Signed-off-by:
Tim Moon <4406448+timmoon10@users.noreply.github.com>
-
Daniel Stokes authored
-
oliver könig authored
* build: pull cached wheels Signed-off-by:
oliver könig <okoenig@nvidia.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update setup.py Signed-off-by:
Kirthi Shankar Sivamani <ksivamani@nvidia.com>
-
- 28 Aug, 2025 3 commits
-
-
Zhongbo Zhu authored
[PyTorch][MOE] Tentative Fix For Replacing from_blob with empty for experts receiving zero tokens (#2134) use torch empty for empty shape instead of from_blob Signed-off-by:
zhongboz <zhongboz@nvidia.com> Co-authored-by:
-
Paweł Gadziński authored
* Compute amax in normalization forward in current scaling in untuned kernels Signed-off-by:
Jan Bielak <jbielak@nvidia.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix Signed-off-by:
Pawel Gadzinski <pgadzinski@nvidia.com> * fix Signed-off-by:
-
Charlene Yang authored
* disable determinism for sm100+ and cudnn<9.14 Signed-off-by:
Charlene Yang <8636796+cyanguwa@users.noreply.github.com> * fix remaining CI failures Signed-off-by:
-
- 27 Aug, 2025 1 commit
-
-
Kshitij Lakhani authored
Signed-off-by:Kshitij Lakhani <klakhani@nvidia.com>
-
- 26 Aug, 2025 1 commit
-
-
Xin Yao authored
expose more activation functions Signed-off-by:Xin Yao <xiny@nvidia.com>
-
- 23 Aug, 2025 1 commit
-
-
Tim Moon authored
* Return dummy wgrad tensors when requested by Mcore Signed-off-by:
Jan Bielak <janekb04@icloud.com> Signed-off-by:
-
- 20 Aug, 2025 3 commits
-
-
Tim Moon authored
Avoid garbage collection when capturing a CUDA Graph Signed-off-by: -
Paweł Gadziński authored
* code drop Signed-off-by:
-
Jan Bielak authored
* Add rmsnorm_bwd_add Signed-off-by:
-
- 18 Aug, 2025 1 commit
-
-
Xin Yao authored
* check if the given recipe is supported in fp8_autocast Signed-off-by:
-
- 16 Aug, 2025 1 commit
-
-
jomitchellnv authored
fix: fixes multi head attention for context parallel: rotary embedding to use padded cu_seq_lens (#2077) fix: fixes mha to use padded cu_seq_lens during cp Signed-off-by:Jonathan Mitchell <jomitchell@nvidia.com>
-
- 15 Aug, 2025 1 commit
-
-
Jan Bielak authored
* Add `nvte_cublas_gemm_scaled` Signed-off-by:
-
- 14 Aug, 2025 5 commits
-
-
Kirthi Shankar Sivamani authored
Add launch bounds to swizzle kernel, use empty scale inv Signed-off-by: -
Tim Moon authored
* Unfused impl for dbias-quantize Signed-off-by:
-
Tim Moon authored
Avoid registering FP8 recipe update in ops without backward pass Signed-off-by:
-
Tim Moon authored
* Register weight/bias params in linear op Signed-off-by:
-
Kshiteej K authored
* fix: update grad_output quant to avoid redundant work Signed-off-by:
kshitij12345 <kshitijkalambarkar@gmail.com> * add test Signed-off-by:
-
- 13 Aug, 2025 2 commits
-
-
Kate Cheng authored
* Remove if-else and torch.tensor to meet cudagraph requirement Signed-off-by:
Kate Cheng <yunhsuanc@nvidia.com> * Add is_cg_capturable flag to guard the if-else statement Signed-off-by:
-
Paweł Gadziński authored
* code drop Signed-off-by:
-
- 12 Aug, 2025 2 commits
-
-
Tim Moon authored
Fix bug when deducing dtype in linear functional API Signed-off-by: -
Tim Moon authored
* Update amax pointers when reallocating amax history in fusible ops Signed-off-by:
-
- 11 Aug, 2025 1 commit
-
-
Kirthi Shankar Sivamani authored
* Fix high-precision dtype for MXFP8 AG Signed-off-by:
-
- 09 Aug, 2025 1 commit
-
-
Daniel Stokes authored
* fix: Add stream synchronization before destroying MPI communicator (#1979) Signed-off-by:
djns99 <40156487+djns99@users.noreply.github.com> * feat: Implement column-wise userbuffer overlap for comm+GEMM operations Add support for overlapping column-wise allgather communication with GEMM operations to improve training performance: * **Core infrastructure changes:** - Update bulk_overlap_columnwise_ag() to accept explicit stream parameter - Modify userbuffers send/recv loops to use rank-ordered iteration - Add userbuffers_send_all/recv_all function declarations * **Python integration:** - Add bulk_overlap_ag_with_external_gemm() C++ extension function - Expose new overlap function via pybind11 bindings - Update overlap method configurations to include more ring_exchange ops * **LayerNorm MLP optimization:** - Enable column-wise quantization for FC2 gradient output - Implement overlap of allgather communication with FC2 DGRAD GEMM - Use fill_userbuffers_buffer_for_all_gather for efficient buffering This optimization allows overlapping communication and computation phases more effectively, reducing training wall-clock time by hiding allgather latency behind GEMM execution. Signed-off-by:
-
- 08 Aug, 2025 1 commit
-
-
Paweł Gadziński authored
* turn on userbuffers for layers without debug Signed-off-by:
-
- 07 Aug, 2025 1 commit
-
-
ldl authored
[PyTorch] Fix numeric overflow caused by int-type parameters and return value in the roundup function (#2034) Signed-off-by:
lvdunlin <lvdunlin@xiaomi.com> Co-authored-by:
-
- 06 Aug, 2025 3 commits
-
-
hx authored
[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor (#1978) * fix input_quantizer in save_original_input bwd Signed-off-by:
Hongxiao Bai <hongxiaob@nvidia.com> * fix get shape of blockwise tensor with only compact colwise data Signed-off-by:
-
Xin Yao authored
* for loop Signed-off-by:
-
xiaoxi-wangfj authored
Signed-off-by:xiaoxi-wangfj <690912414@qq.com>
-
- 04 Aug, 2025 2 commits
-
-
buptzyb authored
* fix graph static grad input reuse Signed-off-by:
-
Tim Moon authored
* Add basic kernel for swapping first two tensor dims Signed-off-by:
-
- 01 Aug, 2025 1 commit
-
-
Hongbin Liu authored
* disable wgrad accumulation and reduce in backward() And manually launch it in backward_dw() Signed-off-by:
Hongbin Liu <hongbinl@nvidia.com> * format Signed-off-by:
-
- 31 Jul, 2025 1 commit
-
-
Jan Bielak authored
* Compute amax in normalization kernels as long as the pointer is provided, even if using non quantized output Signed-off-by:
-
- 30 Jul, 2025 2 commits
-
-
Jan Bielak authored
Refactor normalization.cpp to use quantizer logic introduced in #1952 instead of manual quantization Signed-off-by: -
Tim Moon authored
Fix merge conflict bug with clearing op outputs Signed-off-by:
-
- 29 Jul, 2025 2 commits
-
-
Jan Bielak authored
* Compute amax in activation kernels when the output pointer is provided, even for non-fp8 outputs Signed-off-by:
-
Jan Bielak authored
Signed-off-by:
-
