Add TransformerEngine by weiji14 · Pull Request #26296 · conda-forge/staged-recipes

weiji14 · 2024-05-09T07:55:49Z

A library for accelerating Transformer models on NVIDIA GPUs! Repo at https://github.com/NVIDIA/TransformerEngine

Checklist

conda-forge-webservices · 2024-05-09T07:55:59Z

Hi! This is the friendly automated conda-forge-linting service.

I wanted to let you know that I linted all conda-recipes in your PR (TransformerEngine) and found some lint.

Here's what I've got...

For TransformerEngine:

Please do not delete the example recipe found in recipes/example/meta.yaml.

A library for accelerating Transformer models on NVIDIA GPUs! Repo at https://github.com/NVIDIA/TransformerEngine

conda-forge-webservices · 2024-05-09T07:57:30Z

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipes/TransformerEngine) and found it was in an excellent condition.

weiji14 · 2024-05-09T08:30:14Z

+    - pip
+    - python
+  run:
+    - flash-attn >=2.0.6,<=2.4.2,!=2.0.9,!=2.1.0


Wait for NVIDIA/TransformerEngine#838 and a subsequent release, since conda-forge only has flash-attn=2.5.8 at the moment.

Bumps [TransformerEngine](https://github.com/NVIDIA/TransformerEngine) from 1.5 to 1.6. - [Release notes](https://docs.nvidia.com/deeplearning/transformer-engine/release-notes/index.html) - [Commits](NVIDIA/TransformerEngine@v1.5...v1.6)

weiji14 · 2024-05-15T20:33:48Z

So TransformerEngine v1.6 has a git submodule dependency on cudnn-frontend v1.3.0 (see https://github.com/NVIDIA/TransformerEngine/tree/v1.6/3rdparty). From reading conda-forge/conda-forge.github.io#619, it sounds like the proper way is to actually have a cudnn-frontend package on conda-forge and let transformer-engine depend on that, so opened #26370 for this.

~~Edit: TransformerEngine now uses cudnn-frontend=1.5.2 (NVIDIA/TransformerEngine#975) on its main branch (commit NVIDIA/TransformerEngine@67b6743), maybe wait for TransformerEngine v1.8?~~

~~Edit2: TransformerEngine v1.11 is using cudnn-frontend=1.6.1 (NVIDIA/TransformerEngine#1108). Do the update at conda-forge/python-cudnn-frontend-feedstock#10 first.~~

Edit3: TransformerEngine v1.12 is using cudnn-frontend=1.8.0 (NVIDIA/TransformerEngine#1302). Do the update at conda-forge/python-cudnn-frontend-feedstock#18 first.

Bumps [TransformerEngine](https://github.com/NVIDIA/TransformerEngine) from 1.6 to 1.7. - [Release notes](https://docs.nvidia.com/deeplearning/transformer-engine/release-notes/index.html) - [Commits](NVIDIA/TransformerEngine@v1.6...v1.7)

Bumps [TransformerEngine](https://github.com/NVIDIA/TransformerEngine) from 1.7 to 2.1. - [Release notes](https://docs.nvidia.com/deeplearning/transformer-engine/release-notes/index.html) - [Commits](NVIDIA/TransformerEngine@v1.7...v2.1)

github-actions · 2025-03-28T21:57:32Z

Hi! This is the staged-recipes linter and your PR looks excellent! 🚀

conda-forge-admin · 2025-03-28T21:57:56Z

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipes/TransformerEngine/recipe.yaml) and found it was in an excellent condition.

I do have some suggestions for making it better though...

For recipes/TransformerEngine/recipe.yaml:

ℹ️ No valid build backend found for Python recipe for package transformerengine using pip. Python recipes using pip need to explicitly specify a build backend in the host section. If your recipe has built with only pip in the host section in the past, you likely should add setuptools to the host section of your recipe.

_{This message was generated by GitHub Actions workflow run https://github.com/conda-forge/conda-forge-webservices/actions/runs/14261263322. Examine the logs at this URL for more detail.}

Done using `conda-recipe-manager convert recipes/TransformerEngine/meta.yaml > recipes/TransformerEngine/recipe.yaml`

Fix `fatal error: cuda_runtime.h: No such file or directory`

Fix `fatal error: crt/host_config.h: No such file or directory`

Fixes ``` CMake Error at CMakeLists.txt:44 (include): include could not find requested file: $SRC_DIR/transformer_engine/common/../../3rdparty/cudnn-frontend/cmake/cuDNN.cmake ```

weiji14 · 2025-04-30T20:11:36Z

Cmake build error from https://dev.azure.com/conda-forge/feedstock-builds/_build/results?buildId=1216048&view=logs&j=df9408f7-e795-500f-143d-aa7735a0ed8b&t=47addc80-ccbf-566b-ee3a-66cdb8436937&l=1351:

Building CMake extension transformer_engine
Running command $BUILD_PREFIX/bin/cmake -S $SRC_DIR/transformer_engine/common -B $SRC_DIR/build/cmake -DPython_EXECUTABLE=$PREFIX/bin/python -DPython_INCLUDE_DIR=$PREFIX/include/python3.9 -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=$SRC_DIR/build/lib.linux-x86_64-cpython-39 -DCMAKE_CUDA_ARCHITECTURES=70;80;89;90 -Dpybind11_DIR=$PREFIX/lib/python3.9/site-packages/pybind11/share/cmake/pybind11 -GNinja
CMake Error at $BUILD_PREFIX/share/cmake-4.0/Modules/CMakeDetermineCompilerId.cmake:868 (message):
  Compiling the CUDA compiler identification source file
  "CMakeCUDACompilerId.cu" failed.
  Compiler:
  $PREFIX/bin/nvcc
  Build flags:
  Id flags: --keep;--keep-dir;tmp -v
  The output was:
  127
  nvcc warning : Support for offline compilation for architectures prior to
  '<compute/sm/lto>_75' will be removed in a future release (Use
  -Wno-deprecated-gpu-targets to suppress warning).
  #$ NVCC_PREPEND_FLAGS="
  -ccbin=$BUILD_PREFIX/bin/x86_64-conda-linux-gnu-c++"
  #$ NVCC_APPEND_FLAGS="
  -I$PREFIX/targets/x86_64-linux/include
  -L$PREFIX/targets/x86_64-linux/lib
  -L$PREFIX/targets/x86_64-linux/lib/stubs"
  #$ _NVVM_BRANCH_=nvvm
  #$ _SPACE_=
  #$ _CUDART_=cudart
  #$
  _HERE_=$PREFIX/bin
  #$
  _THERE_=$PREFIX/bin
  #$ _TARGET_SIZE_=
  #$ _TARGET_DIR_=
  #$ _TARGET_DIR_=targets/x86_64-linux
  #$
  TOP=$PREFIX/bin/../targets/x86_64-linux
  #$
  CICC_PATH=$PREFIX/bin/../targets/x86_64-linux/nvvm/bin
  #$
  NVVMIR_LIBRARY_DIR=$PREFIX/bin/../targets/x86_64-linux/nvvm/libdevice
  #$
  LD_LIBRARY_PATH=$PREFIX/bin/../targets/x86_64-linux/lib:
  #$
  PATH=$PREFIX/bin/../targets/x86_64-linux/bin:$PREFIX/bin/../targets/x86_64-linux/nvvm/bin:$PREFIX/bin/../targets/x86_64-linux/../../bin:$PREFIX/bin/../targets/x86_64-linux/../../nvvm/bin:$BUILD_PREFIX/bin:$PREFIX/bin:/opt/conda/bin:/opt/conda/condabin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/home/conda/bin
  #$
  INCLUDES="-I$PREFIX/bin/../targets/x86_64-linux/include"
  #$ LIBRARIES=
  "-L$PREFIX/bin/../targets/x86_64-linux/lib/stubs"
  "-L$PREFIX/bin/../targets/x86_64-linux/lib"
  #$ CUDAFE_FLAGS=
  #$ PTXAS_FLAGS=
  #$ rm tmp/a_dlink.reg.c
  #$
  "$BUILD_PREFIX/bin"/x86_64-conda-linux-gnu-c++
  -D__CUDA_ARCH_LIST__=520 -D__NV_LEGACY_LAUNCH -E -x c++ -D__CUDACC__
  -D__NVCC__
  -I"$PREFIX/targets/x86_64-linux/include"
  "-I$PREFIX/bin/../targets/x86_64-linux/include"
  -D__CUDACC_VER_MAJOR__=12 -D__CUDACC_VER_MINOR__=8
  -D__CUDACC_VER_BUILD__=93 -D__CUDA_API_VER_MAJOR__=12
  -D__CUDA_API_VER_MINOR__=8 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1
  -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include "cuda_runtime.h" -m64
  "CMakeCUDACompilerId.cu" -o "tmp/CMakeCUDACompilerId.cpp4.ii"
  #$ cudafe++ --c++17 --gnu_version=130300 --display_error_number
  --orig_src_file_name "CMakeCUDACompilerId.cu" --orig_src_path_name
  "$SRC_DIR/build/cmake/CMakeFiles/4.0.0/CompilerIdCUDA/CMakeCUDACompilerId.cu"
  --allow_managed --m64 --parse_templates --gen_c_file_name
  "tmp/CMakeCUDACompilerId.cudafe1.cpp" --stub_file_name
  "CMakeCUDACompilerId.cudafe1.stub.c" --gen_module_id_file
  --module_id_file_name "tmp/CMakeCUDACompilerId.module_id"
  "tmp/CMakeCUDACompilerId.cpp4.ii"
  #$
  "$BUILD_PREFIX/bin"/x86_64-conda-linux-gnu-c++
  -D__CUDA_ARCH__=520 -D__CUDA_ARCH_LIST__=520 -D__NV_LEGACY_LAUNCH -E -x c++
  -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__
  -I"$PREFIX/targets/x86_64-linux/include"
  "-I$PREFIX/bin/../targets/x86_64-linux/include"
  -D__CUDACC_VER_MAJOR__=12 -D__CUDACC_VER_MINOR__=8
  -D__CUDACC_VER_BUILD__=93 -D__CUDA_API_VER_MAJOR__=12
  -D__CUDA_API_VER_MINOR__=8 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1
  -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include "cuda_runtime.h" -m64
  "CMakeCUDACompilerId.cu" -o "tmp/CMakeCUDACompilerId.cpp1.ii"
  #$ "$CICC_PATH/cicc" --c++17 --gnu_version=130300 --display_error_number
  --orig_src_file_name "CMakeCUDACompilerId.cu" --orig_src_path_name
  "$SRC_DIR/build/cmake/CMakeFiles/4.0.0/CompilerIdCUDA/CMakeCUDACompilerId.cu"
  --allow_managed -arch compute_52 -m64 --no-version-ident -ftz=0 -prec_div=1
  -prec_sqrt=1 -fmad=1 --include_file_name "CMakeCUDACompilerId.fatbin.c"
  -tused --module_id_file_name "tmp/CMakeCUDACompilerId.module_id"
  --gen_c_file_name "tmp/CMakeCUDACompilerId.cudafe1.c" --stub_file_name
  "tmp/CMakeCUDACompilerId.cudafe1.stub.c" --gen_device_file_name
  "tmp/CMakeCUDACompilerId.cudafe1.gpu" "tmp/CMakeCUDACompilerId.cpp1.ii" -o
  "tmp/CMakeCUDACompilerId.ptx"
  sh: line 1:
  $PREFIX/bin/../targets/x86_64-linux/nvvm/bin/cicc:
  No such file or directory
  # --error 0x7f --
  Compiling the CUDA compiler identification source file
  "CMakeCUDACompilerId.cu" failed.
  Compiler:
  $PREFIX/bin/nvcc
  Build flags:
  -Wl,--version-script=$SRC_DIR/transformer_engine/common/libtransformer_engine.version
  Id flags: --keep;--keep-dir;tmp -v
  The output was:
  1
  nvcc fatal : Unknown option
  '-Wl,--version-script=$SRC_DIR/transformer_engine/common/libtransformer_engine.version'
Call Stack (most recent call first):
  $BUILD_PREFIX/share/cmake-4.0/Modules/CMakeDetermineCompilerId.cmake:8 (CMAKE_DETERMINE_COMPILER_ID_BUILD)
  $BUILD_PREFIX/share/cmake-4.0/Modules/CMakeDetermineCompilerId.cmake:53 (__determine_compiler_id_test)
  $BUILD_PREFIX/share/cmake-4.0/Modules/CMakeDetermineCUDACompiler.cmake:131 (CMAKE_DETERMINE_COMPILER_ID)
  CMakeLists.txt:27 (project)
CMake Error: CMAKE_CXX_COMPILER not set, after EnableLanguage
-- Configuring incomplete, errors occurred!

weiji14 · 2025-05-01T22:07:59Z

Closing since #29403 is merged 🚀 Feedstock is up at https://github.com/conda-forge/transformer-engine-feedstock

Add TransformerEngine

9bad097

A library for accelerating Transformer models on NVIDIA GPUs! Repo at https://github.com/NVIDIA/TransformerEngine

weiji14 force-pushed the add-transformer-engine branch from bd391e7 to 9bad097 Compare May 9, 2024 07:57

weiji14 mentioned this pull request May 9, 2024

Add flash-attn #26239

Merged

10 tasks

weiji14 commented May 9, 2024

View reviewed changes

weiji14 added 3 commits May 9, 2024 18:47

Add pytorch as host dependency and set cuda compiler

59625d9

Bump TransformerEngine from 1.5 to 1.6

4ba579d

Bumps [TransformerEngine](https://github.com/NVIDIA/TransformerEngine) from 1.5 to 1.6. - [Release notes](https://docs.nvidia.com/deeplearning/transformer-engine/release-notes/index.html) - [Commits](NVIDIA/TransformerEngine@v1.5...v1.6)

Add cmake and ninja as build dependency, cudnn as host dependency

92f51d6

weiji14 added 3 commits June 15, 2024 12:24

Bump TransformerEngine from 1.6 to 1.7

6df918a

Bumps [TransformerEngine](https://github.com/NVIDIA/TransformerEngine) from 1.6 to 1.7. - [Release notes](https://docs.nvidia.com/deeplearning/transformer-engine/release-notes/index.html) - [Commits](NVIDIA/TransformerEngine@v1.6...v1.7)

Merge branch 'main' into add-transformer-engine

4233339

Bump TransformerEngine from 1.7 to 2.1

ea332d9

Bumps [TransformerEngine](https://github.com/NVIDIA/TransformerEngine) from 1.7 to 2.1. - [Release notes](https://docs.nvidia.com/deeplearning/transformer-engine/release-notes/index.html) - [Commits](NVIDIA/TransformerEngine@v1.7...v2.1)

weiji14 mentioned this pull request Mar 28, 2025

Add NVIDIA TransformerEngine #29403

Merged

10 tasks

weiji14 added 5 commits April 4, 2025 16:40

Merge branch 'main' into add-transformer-engine

9c47561

Convert to recipe.yaml

6df0484

Done using `conda-recipe-manager convert recipes/TransformerEngine/meta.yaml > recipes/TransformerEngine/recipe.yaml`

Add cuda-cuart-dev to host deps

40c31b0

Fix `fatal error: cuda_runtime.h: No such file or directory`

Add cuda-crt-dev_linux-64 to host deps

590ed11

Fix `fatal error: crt/host_config.h: No such file or directory`

Add python-cudnn-frontend and symlink headers to 3rdparty/cudnn-frontend

a4b517f

Fixes ``` CMake Error at CMakeLists.txt:44 (include): include could not find requested file: $SRC_DIR/transformer_engine/common/../../3rdparty/cudnn-frontend/cmake/cuDNN.cmake ```

weiji14 closed this May 1, 2025

Navigation Menu

Uh oh!

Uh oh!

Add TransformerEngine - #26296

Add TransformerEngine#26296
weiji14 wants to merge 12 commits into
conda-forge:mainfrom
weiji14:add-transformer-engine

weiji14 commented May 9, 2024

Uh oh!

conda-forge-webservices Bot commented May 9, 2024

Uh oh!

conda-forge-webservices Bot commented May 9, 2024

Uh oh!

weiji14 May 9, 2024

Uh oh!

weiji14 commented May 15, 2024 •

edited

Loading

Uh oh!

github-actions Bot commented Mar 28, 2025

Uh oh!

conda-forge-admin commented Mar 28, 2025 •

edited

Loading

Uh oh!

weiji14 commented Apr 30, 2025

Uh oh!

weiji14 commented May 1, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Uh oh!

Conversation

weiji14 commented May 9, 2024

Uh oh!

conda-forge-webservices Bot commented May 9, 2024

Uh oh!

conda-forge-webservices Bot commented May 9, 2024

Uh oh!

weiji14 May 9, 2024

Choose a reason for hiding this comment

Uh oh!

weiji14 commented May 15, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Mar 28, 2025

Uh oh!

conda-forge-admin commented Mar 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

weiji14 commented Apr 30, 2025

Uh oh!

weiji14 commented May 1, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

weiji14 commented May 15, 2024 •

edited

Loading

conda-forge-admin commented Mar 28, 2025 •

edited

Loading