- 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:
pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
-
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:
Pawel Gadzinski <pgadzinski@nvidia.com> * fix 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:
-
- 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:
-
- 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:
-
- 24 Jul, 2025 2 commits
-
-
Jan Bielak authored
* Mark output tensors as not deletable in backward Signed-off-by:
-
Evgeny Tsykunov authored
* Increase intermediate precision and reuse tensors from fwd Signed-off-by:
-
- 23 Jul, 2025 1 commit
-
-
Charlene Yang authored
* fix current device for cuDNN/cuBLAS handles Signed-off-by:
Charlene Yang <8636796+cyanguwa@users.noreply.github.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * add unit test Signed-off-by:
-
- 22 Jul, 2025 1 commit
-
-
Jan Bielak authored
* Refactor _OperationFuserAutogradFunction.forward to use less parameters Signed-off-by:
-
- 21 Jul, 2025 1 commit
-
-
Charlene Yang authored
* exclude 9.10.0/.1 for certain configs Signed-off-by:
-
- 17 Jul, 2025 1 commit
-
-
hx authored
* save original input Signed-off-by:
-
- 16 Jul, 2025 1 commit
-
-
Paweł Gadziński authored
* some initial code Signed-off-by:
Paweł Gadziński <62263673+pggPL@users.noreply.github.com> * fix Signed-off-by:
root <root@prenyx0221.a51.clusters.nvidia.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * lint fix Signed-off-by:
Pawel Gadzinski <pgadzinski@gmail.com> * Update build.yml Signed-off-by:
-
- 14 Jul, 2025 2 commits
-
-
Autumn1998 authored
* fix underterminsic problem in CI Signed-off-by:
-
hx authored
* optimize permute Signed-off-by:
-
- 12 Jul, 2025 1 commit
-
-
Jan Bielak authored
* Fix clearing tensor data in backward removing is_first_op Signed-off-by:
-
- 10 Jul, 2025 2 commits
-
-
Autumn1998 authored
* add router fusion Signed-off-by:
-
Paweł Gadziński authored
* push Signed-off-by:
-
- 09 Jul, 2025 1 commit
-
-
Tim Moon authored
* Add tests for loading previously-generated checkpoints Signed-off-by:
-
- 08 Jul, 2025 1 commit
-
-
Jan Bielak authored
Add test for `LayerNormMLP` implementation using `te.ops.Sequential` to `test_fusible_ops.py` (#1924) * Add e2e test for LayerNormMLP implemented with te.Sequential Signed-off-by:
-
- 04 Jul, 2025 1 commit
-
-
Paweł Gadziński authored
* fix Signed-off-by:
-
- 01 Jul, 2025 1 commit
-
-
Jan Bielak authored
* Replace `is_float8_tensor` with `is_quantized_tensor` Replace free function `is_float8_tensor` with `is_quantized_tensor` in `_common.py` and use it throughout the `ops` codebase to check if a tensor is a (possibly internal) quantized tensor Signed-off-by:
-
- 28 Jun, 2025 1 commit
-
-
yuzhongw-nvidia authored
* fix: (1) UT ignores MLA; (2) bshd format runtime error. Ban fp8 mla attn + cp due to correctness problem Signed-off-by:
Yuzhong Wang <yuzhongw@nvidia.com> * only disable FP8 CP for MLA Signed-off-by:
-
- 26 Jun, 2025 2 commits
-
-
Charlene Yang authored
* skip kv cache for sm89, cudnn < 9.12 Signed-off-by:
-
Paweł Gadziński authored
* fixed the bug Signed-off-by:
-
- 19 Jun, 2025 1 commit
-
-
Tim Moon authored
* Use FP16 tols for tests with TF32 Signed-off-by:
-
- 16 Jun, 2025 1 commit
-
-
Li Tao authored
* Fix an issue when mcore uses te fusion ce implementation Signed-off-by:
lit <lit@nvidia.com> * simplify unit test code Signed-off-by:
-
- 13 Jun, 2025 4 commits
-
-
Charlene Yang authored
* add support for head dim > 128 Signed-off-by:
Charlene Yang <charleney@nvidia.com> * remove debugging Signed-off-by:
-
Tim Moon authored
* Add FP8 current scaling to te.Sequential tests Signed-off-by:
-
Tim Moon authored
* Do not initialize quantized weights with column-wise usage in inference mode Signed-off-by:
-
Daniel Stokes authored
* Add support for overlapping wgrad NCCL AG with dgrad GEMM Signed-off-by:
dastokes <dastokes@dastokes-dvt-01.nvidia.com>
-
- 12 Jun, 2025 1 commit
-
-
Evgeny Tsykunov authored
* Support L2Norm basic op Signed-off-by:
-
