- 09 Mar, 2026 2 commits
-
-
wenjh authored
Signed-off-by:wenjh <wenjh@sugon.com>
-
wuyf1 authored
Signed-off-by:
Wuyufan <Wuyf1@sugon.com> 解决了如下问题： 1. FA 48个单测报错问题（fA输入:[B,S,H,D] reshape to [blocknums,blocksize,H,D]）. 2. 在解决FA 报错的时候一个kernel launch error. 3. norm_mlp问题暂时用rest_rng_state解决 See merge request dcutoolkit/deeplearing/TransformerEngine!77 Co-authored-by:
Tangao <2205747538@qq.com> Co-authored-by:
wuyufffan <1095978552@qq.com>
-
- 23 Feb, 2026 2 commits
-
-
wenjh authored
Signed-off-by: -
wenjh authored
Signed-off-by:
-
- 09 Jan, 2026 1 commit
-
-
wenjh authored
Signed-off-by:
-
- 30 Oct, 2025 1 commit
-
-
zhaochao authored
Signed-off-by:zhaochao <zhaochao1@sugon.com>
-
- 23 Oct, 2025 1 commit
-
-
zhaochao authored
Signed-off-by:
-
- 20 Oct, 2025 6 commits
-
-
zhaochao authored
Signed-off-by: -
tabuchixiangcai3 authored
[DCU]Fix MPI root support, enable int8 simulation and batched_inear to access non-existent. main_grad Signed-off-by: -
zhaochao authored
Signed-off-by: -
zhaochao authored
Signed-off-by: -
zhaochao authored
Signed-off-by: -
zhaochao authored
Signed-off-by:
-
- 17 Oct, 2025 1 commit
-
-
tabuchixiangcai3 authored
Signed-off-by:
-
- 16 Oct, 2025 1 commit
-
-
tabuchixiangcai3 authored
Signed-off-by:
-
- 18 Sep, 2025 1 commit
-
-
wenjh authored
Signed-off-by:
-
- 09 Sep, 2025 1 commit
-
-
wenjh authored
Signed-off-by:
-
- 03 Sep, 2025 1 commit
-
-
wenjh authored
Signed-off-by:
-
- 28 Aug, 2025 1 commit
-
-
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:
pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
-
- 27 Aug, 2025 1 commit
-
-
Paweł Gadziński authored
* code drop Signed-off-by:
Pawel Gadzinski <pgadzinski@nvidia.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix Signed-off-by:
-
- 26 Aug, 2025 1 commit
-
-
vcherepanov-nv authored
* Bump cuDNN FE to 1.14.0 Signed-off-by:
Vladimir Cherepanov <vcherepanov@nvidia.com> * Change submodule hash Signed-off-by:
-
- 25 Aug, 2025 2 commits
- 21 Aug, 2025 1 commit
-
-
yuguo authored
-
- 15 Aug, 2025 1 commit
-
-
Jan Bielak authored
* Add `nvte_cublas_gemm_scaled` Signed-off-by:
Jan Bielak <jbielak@nvidia.com> * Support use of `alpha` and `beta` in `tex.generic_gemm` Signed-off-by:
Tim Moon <4406448+timmoon10@users.noreply.github.com>
-
- 14 Aug, 2025 3 commits
-
-
Tim Moon authored
* Unfused impl for dbias-quantize Signed-off-by:
Tim Moon <tmoon@nvidia.com> * Unfused impl for dact-dbias-quantize Signed-off-by:
-
Xin Yao authored
* reduce driver calls Signed-off-by:
Xin Yao <xiny@nvidia.com> * reduce driver calls 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:
Kirthi Shankar Sivamani <ksivamani@nvidia.com> --------- Signed-off-by:
-
- 13 Aug, 2025 1 commit
-
-
Paweł Gadziński authored
* code drop 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 2 commits
-
-
Paweł Gadziński authored
* turn on userbuffers for layers without debug Signed-off-by:
-
yuguo authored
-
- 06 Aug, 2025 1 commit
-
-
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:
-
- 01 Aug, 2025 1 commit
-
-
Autumn1998 authored
* fix bug if all values<0 Signed-off-by:
tongliu <tongliu@nvidia.com> * minor fix Signed-off-by:
-
- 31 Jul, 2025 1 commit
-
-
yuguo authored
-
- 29 Jul, 2025 2 commits
-
-
Tim Moon authored
* remove reciprocal op Signed-off-by:
zhongboz <zhongboz@nvidia.com> * Refactor Quantizer::create_tensor function Signed-off-by:
-
Kirthi Shankar Sivamani authored
* Add verbosity only for failing tests Signed-off-by:
-
- 25 Jul, 2025 3 commits
-
-
Tim Moon authored
* Add ops for dropout and constant scale Signed-off-by:
-
Kirthi Shankar Sivamani authored
* Remove deprecated device arg Signed-off-by:
-
Evgeny Tsykunov authored
* Support RMSNorm for QK Signed-off-by:
Evgeny <etsykunov@nvidia.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * rms -> RMSNorm, l2 -> L2Normalization (align with current pattern) Signed-off-by:
-
