- 18 Feb, 2026 1 commit
-
-
Przemek Tredak authored
Signed-off-by:Przemek Tredak <ptredak@nvidia.com>
-
- 17 Feb, 2026 2 commits
-
-
jberchtold-nvidia authored
* initial debug of inspect ffi Signed-off-by:
Jeremy Berchtold <jberchtold@nvidia.com> * writing binary dumps of tensors works Signed-off-by:
pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
-
Hemil Desai authored
- Include `build_tools/` in the source distribution via `MANIFEST.in` so that cached builds from `uv` (and `pip`) can resolve `setup.py`'s top-level imports `setup.py` imports from `build_tools` at the top level: ```python from build_tools.build_ext import CMakeExtension, get_build_ext from build_tools.te_version import te_version from build_tools.utils import cuda_archs, cuda_version, ... ``` The `__legacy__` build backend in `pyproject.toml` adds the source root to `sys.path`, so these imports work when building directly from the source tree. However, `build_tools/` is not included in the sdist because: 1. `MANIFEST.in` did not list it 2. `build_tools/` is not discovered by `find_packages()` (it's a standalone directory at the repo root, not under `transformer_engine/`) When `uv` caches the sdist and later builds a wheel from it, the sdist is extracted to a temporary directory where `build_tools/` is absent, causing a `ModuleNotFoundError`. Passing `--no-cache` to `uv` works around this by forcing a fresh build from the full source tree. Added `build_tools` to `MANIFEST.in`: ```diff recursive-include transformer_engine/common/include *.* +recursive-include build_tools *.py *.txt ``` - [x] `python setup.py sdist` produces a tarball that contains `build_tools/` ``` $ tar tzf dist/transformer_engine-*.tar.gz | grep build_tools transformer_engine-2.13.0.dev0+82f7ebeb/build_tools/ transformer_engine-2.13.0.dev0+82f7ebeb/build_tools/VERSION.txt transformer_engine-2.13.0.dev0+82f7ebeb/build_tools/__init__.py transformer_engine-2.13.0.dev0+82f7ebeb/build_tools/build_ext.py transformer_engine-2.13.0.dev0+82f7ebeb/build_tools/jax.py transformer_engine-2.13.0.dev0+82f7ebeb/build_tools/pytorch.py transformer_engine-2.13.0.dev0+82f7ebeb/build_tools/te_version.py transformer_engine-2.13.0.dev0+82f7ebeb/build_tools/utils.py ``` Signed-off-by:
Hemil Desai <hemild@nvidia.com> Co-authored-by:
Claude Opus 4.6 <noreply@anthropic.com>
-
- 13 Feb, 2026 2 commits
-
-
Teddy Do authored
* [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * adding more stuff missing from cherry picky jeremy PR for inspecting Signed-off-by:
tdophung <tdophung@nvidia.com> * fix some tracing issues when intergating to maxtext Signed-off-by:
JAX Toolbox <jax@nvidia.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * remove Jeremy's PR for inspect ffi Signed-off-by:
-
Kirthi Shankar Sivamani authored
* Make grouped weights opt-in Signed-off-by:
Kirthi Shankar Sivamani <ksivamani@nvidia.com> * Change varname Signed-off-by:
-
- 12 Feb, 2026 6 commits
-
-
vcherepanov-nv authored
* Remove nvshmem usage Signed-off-by:
Vladimir Cherepanov <vcherepanov@nvidia.com> * Renamings Signed-off-by:
-
Harikrishna KP authored
fix: correct copy-paste error messages in FusedSGD Signed-off-by:Mr-Neutr0n <64578610+Mr-Neutr0n@users.noreply.github.com>
-
Kim, Jin (Jay@SKT) authored
* Add sigmoid GLU Signed-off-by:
Kim, Jin <jinn.kim@sk.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci Signed-off-by:
Tim Moon <tmoon@nvidia.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Fix incorrect reshape Signed-off-by:
Tim Moon <4406448+timmoon10@users.noreply.github.com> * Apply suggestion from @timmoon10 Signed-off-by:
-
Tim Moon authored
* Add ops for MoE grouped MLP Signed-off-by:
-
Oleg Goncharov authored
* Added GEMM-ready preswizzling option Signed-off-by:
Oleg Goncharov <ogoncharov@nvidia.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Signed-off-by:
-
Santosh Bhavani authored
fix: handle nvidia namespace packages where __file__ is None Signed-off-by:
Santosh Bhavani <santosh.bhavani@live.com> Co-authored-by:
-
- 11 Feb, 2026 4 commits
-
-
Kirthi Shankar Sivamani authored
* NVFP4 GroupedQuantize Signed-off-by:
Zhongbo Zhu <zhongboz@nvidia.com> Co-authored-by:
-
Kirthi Shankar Sivamani authored
* PyTorch-Python GroupedTensor Signed-off-by:
greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com> Signed-off-by:
-
Lifu Zhang authored
* Fix on TE to support Mcore Vision Encoder CUDA Graph Signed-off-by:
Lifu Zhang <lifuz@login-lyris02.lyris.clusters.nvidia.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * refactoring code Signed-off-by:
-
Faradawn Yang authored
* fix broken link of quickstart guide Signed-off-by:
Faradawn Yang <73060648+faradawn@users.noreply.github.com> * Update README.rst Co-authored-by:
Paweł Gadziński <62263673+pggPL@users.noreply.github.com> Signed-off-by:
-
- 10 Feb, 2026 2 commits
-
-
Jacket authored
Signed-off-by:Kaining Zhong <kainingz@nvidia.com>
-
Przemyslaw Tredak authored
* Fix the compilation warnings for the PyTorch extension Signed-off-by:
Przemyslaw Tredak <ptrendx@gmail.com> --------- Signed-off-by:
-
- 09 Feb, 2026 3 commits
-
-
Pingtian Li authored
* add grad reduce api for cuda graph hook Signed-off-by:
Pingtian Li <pingtianl@nvidia.com> * fix code consistency Signed-off-by:
-
Paweł Gadziński authored
fix Signed-off-by:Pawel Gadzinski <pgadzinski@nvidia.com>
-
jberchtold-nvidia authored
* expand troubleshooting docs Signed-off-by:
jberchtold-nvidia <158520091+jberchtold-nvidia@users.noreply.github.com> * Update README.rst Co-authored-by:
-
- 07 Feb, 2026 1 commit
-
-
Charlene Yang authored
bucket max_b with more granularity when >512 Signed-off-by:Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
-
- 06 Feb, 2026 2 commits
-
-
Oleg Goncharov authored
* Rebased to main Signed-off-by:
vthumbe1503 <vthumbe@nvidia.com>
-
Jacket authored
Signed-off-by:
-
- 04 Feb, 2026 1 commit
-
-
jberchtold-nvidia authored
* Update README.rst Signed-off-by:
-
- 03 Feb, 2026 3 commits
-
-
Paweł Gadziński authored
* init Signed-off-by:
-
Oleg Goncharov authored
* Fixed scaling-factor computation for FP32 to match the reference implementation. Signed-off-by:
-
Vadim Markovtsev authored
* Support building with headers from nvidia wheels There are two changes: 1. `import nvidia` returns a namespace package with `__file__` equal to `None` 2. Add the way to force headers from nvidia wheels. Without that envvar, it's practically impossible with CUDA installed system-wide. I successfully built the package with torch using the following `uv` configuration: ``` [tool.uv.extra-build-dependencies] "transformer-engine-torch" = [ "ninja", "nvidia-cuda-crt==13.0.88", "nvidia-cuda-cccl==13.0.85", { requirement = "torch", match-runtime = true }, { requirement = "pytorch-triton", match-runtime = true }, { requirement = "nvidia-cusolver", match-runtime = true }, { requirement = "nvidia-curand", match-runtime = true }, { requirement = "nvidia-cublas", match-runtime = true }, { requirement = "nvidia-cusparse", match-runtime = true }, { requirement = "nvidia-cudnn-cu13", match-runtime = true }, { requirement = "nvidia-nvtx", match-runtime = true }, { requirement = "nvidia-cuda-nvrtc", match-runtime = true }, { requirement = "nvidia-cuda-runtime", match-runtime = true }, ] ``` Signed-off-by:Vadim Markovtsev <vadim@poolside.ai> * Apply suggestion from @ksivaman Signed-off-by:
-
- 02 Feb, 2026 1 commit
-
-
Paweł Gadziński authored
* Code drop: Update recipes documentation and remove custom recipes from low precision training Signed-off-by:
-
- 30 Jan, 2026 1 commit
-
-
Paweł Gadziński authored
* version change Signed-off-by:
-
- 29 Jan, 2026 1 commit
-
-
Paweł Gadziński authored
fix wheel Signed-off-by:
-
- 28 Jan, 2026 2 commits
-
-
Paweł Gadziński authored
* code drop Signed-off-by:
-
Paweł Gadziński authored
* jjit bug fix Signed-off-by:
-
- 27 Jan, 2026 1 commit
-
-
jberchtold-nvidia authored
* Use "nyu-mll/glue" instead of "glue" for encoder datasets to fix 404 error Signed-off-by:
-
- 26 Jan, 2026 1 commit
-
-
Santosh Bhavani authored
* fix(examples): te_llama compatibility with HuggingFace transformers >= 4.57 The te_llama.py example was failing with HuggingFace transformers 4.57+ due to API changes in how decoder layer outputs are handled. Changes: - Handle case where hidden_states is passed as a tuple (older HF versions) - Return tensor directly instead of wrapped in tuple (HF 4.57+ expects this) - Fix regex pattern to use raw string (fixes SyntaxWarning) Error fixed: AttributeError: 'tuple' object has no attribute 'contiguous' Tested with: - transformer_engine 2.5.0 - transformers 4.57.3 - PyTorch container nvcr.io/nvidia/pytorch:25.08-py3 Signed-off-by:
-
- 25 Jan, 2026 1 commit
-
-
Tim Moon authored
* Expose option for custom op fusions Refactor fusion functions to remove index bookkeeping. Refactor fused ops to use consistent operation order. Signed-off-by:
-
- 23 Jan, 2026 2 commits
-
-
Oleg Goncharov authored
* Disabled the tuned NVFP4 kernels Signed-off-by:
-
Teddy Do authored
* Use correct block size for workspace in row id map creation, also shard workspace correctly based on 2nd dim of routing_map/row_id map Signed-off-by:
DoubleCheeseCheetos <hanhdp99@gmail.com> * reduce size of largest test case on single_GPU scenario to fit on L40 and A100 in CI line up Signed-off-by:
-
- 22 Jan, 2026 3 commits
-
-
Sudhakar Singh authored
* SWA (left, right) with FusedAttention changes cherry-picked from https://github.com/NVIDIA/TransformerEngine/pull/1369 Signed-off-by:
Sudhakar Singh <sudhakars@nvidia.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix test_kv_cache failures Signed-off-by:
-
Chen Cui authored
* Update THD sink attention logic for newer cudnn versions THD Sink attention is supported in 9.18.0 Signed-off-by:
Chen Cui <chcui@nvidia.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * update thd sink attention logic for cp>1 Signed-off-by:
-
Teddy Do authored
return tokens_per_experts always Signed-off-by:
-
