From Fedora Project Wiki

< SIGs‎ | PyTorch

Line 84: Line 84:
| [https://github.com/intel/ARM_NEON_2_x86_SSE neon2sse] || [https://github.com/intel/ARM_NEON_2_x86_SSE/tree/97a126f08ce318023be604d03f88bf0820a9464a 97a126f] (2018-09-17) ||  ||  ||  ||  ||  ||  ||
| [https://github.com/intel/ARM_NEON_2_x86_SSE neon2sse] || [https://github.com/intel/ARM_NEON_2_x86_SSE/tree/97a126f08ce318023be604d03f88bf0820a9464a 97a126f] (2018-09-17) ||  ||  ||  ||  ||  ||  ||
|-
|-
| [https://github.com/nlohmann/json nlohmann/json] || [https://github.com/nlohmann/json/tree/87cda1d6646592ac5866dc703c8e1839046a6806 87cda1d] (2022-06-18) || ||  || || ||  || between 3.10.5 and 3.11.0 ||
| [https://github.com/nlohmann/json nlohmann/json] || [https://github.com/nlohmann/json/tree/87cda1d6646592ac5866dc703c8e1839046a6806 87cda1d] (2022-06-18) || Packaged? ||  || json || [https://src.fedoraproject.org/rpms/json json] ||  || between 3.10.5 and 3.11.0 ||
|-
|-
| [https://pytorch.org/blog/introducing-nvfuser-a-deep-learning-compiler-for-pytorch/ nvfuser] || [https://github.com/pytorch/pytorch/tree/main/third_party/nvfuser in pytorch tree] ||  ||  ||  ||  ||  || ||
| [https://pytorch.org/blog/introducing-nvfuser-a-deep-learning-compiler-for-pytorch/ nvfuser] || [https://github.com/pytorch/pytorch/tree/main/third_party/nvfuser in pytorch tree] ||  ||  ||  ||  ||  || pytorch tree has no nvfuser ??? ||
|-
|-
| [https://github.com/onnx/onnx onnx] || [https://github.com/onnx/onnx/commits/1014f41f17ecc778d63e760a994579d96ba471ff 1.14.1] || Packaged? ||  || onnx || [https://src.fedoraproject.org/rpms/onnx onnx] ||  ||  ||
| [https://github.com/onnx/onnx onnx] || [https://github.com/onnx/onnx/commits/1014f41f17ecc778d63e760a994579d96ba471ff 1.14.1] || Packaged? ||  || onnx || [https://src.fedoraproject.org/rpms/onnx onnx] ||  ||  ||

Revision as of 13:34, 9 November 2023

PyTorch Dependency Packaging

This page was created for the purpose of tracking the effort for packaging PyTorch in Fedora.

If you start working on a package in this list, please mark it as In Progress and put your name under the packager column. Also, have your review request block the ML-SIG Review Tracker and send a link to the review request in at least the AI-ML matrix room.

If you find errors in this list (package that aren't actually needed or missing dependencies), feel free to add them to this list.

Status Key

When updating the dependency table, please use one of the values in the table below. If the status is really not encompassed by the existing statuses, please add a new value and description for the current state.

Package Status Key
Value Meaning
Packaged? Software is packaged in Fedora but it may not be a version that is compatible with PyTorch
Packaged Software is packaged in Fedora and is compatible with building PyTorch
In Progress Someone is currently working to package an appropriate version of the software
Needs Review Package is ready for review

Dependency List

The upstream links and versions were extracted from pytorch's third_party directory upstream on 2023-09-14.

PyTorch Dependency List
PyTorch Name Version Status Packager Fedora Package Name Fedora Package URL EPEL Notes Accelerator
FP16 4dfe081 (2020-05-14) Packaged trix FP16 FP16
FXdiv b408327 (2020-04-17) Packaged trix fxdiv fxdiv
NNPACK c07e3a0 (2020-12-21) In Progress thunderbirdtr
QNNPACK 7d2a4e9 (2019-08-28) Archived upstream on 2020-10-01
VulkanMemoryAllocator v3.0.1 Packaged Jeremy Newton VulkanMemoryAllocator VulkanMemoryAllocator
benchmark v1.6.1 Packaged? google-benchmark google-benchmark
cpuinfo 6481e8b (2023-01-13) Packaged trix cpuinfo cpuinfo
cub d106ddb (2020-05-12) the linked repo is marked as moved, this could be a mess. equivalent commit in NVIDIA/cub? might be cub 1.10.0 but the tags don't seem to correlate with versions mentioned in commit messages
cudnn_frontend v0.9.2 (2023-07-13) this requires CUDA and is likely not package-able for Fedora
cutlass 3.1 this requires CUDA and is likely not package-able for Fedora. we'll have to figure out how to build pytorch without it.
eigen 3.4.0 (2021-08-18) Packaged? eigen3 eigen3
fbgemm 0.5.0 In Progress aekoroglu asmjit dep. submitted for review
flatbuffers 23.3.3 Packaged? flatbuffers flatbuffers
fmt 10.1.1 Packaged? fmt fmt
foxi c278588 (2021-05-26) master/HEAD, appears to be an extension of ONNX Interface for Framework Integration
gemmlowp 3fb5c17 (2018-11-26) "Needs review" crisdel gemmlowp gemmlowp Needs review
gloo cf1e1ab (2023-07-19) Packaged trix
googletest v1.11.0 (2021-06-11) Packaged? gtest gtest
ideep 13dd1fd master branch appears dead - there appear to be two current branches: ideep_dev and ideep_pytorch and the 13dd1fd commit appearst to only be in the ideep_pytorch branch
ios-cmake 8abaed6 (2017-11-15) fork of ollef/ios-cmake which is a fork of cristeab/ios-cmake
ittapi 5b8a7d7 (2022-04-12) before ittapi's first release
kineto 49e854d (2023-08-08) well after last release (v0.4.0)
mimalloc b66e321 (2023-04-24) Packaged? mimalloc mimalloc
miniz-2.1.0 Forked in PyTorch Upstream Packaged? miniz miniz
nccl 2.18.5-1 (2023-08-23)
neon2sse 97a126f (2018-09-17)
nlohmann/json 87cda1d (2022-06-18) Packaged? json json between 3.10.5 and 3.11.0
nvfuser in pytorch tree pytorch tree has no nvfuser ???
onnx 1.14.1 Packaged? onnx onnx
onnx-tensorrt 6.0 (2019-09-16) this requires CUDA and is likely not package-able for Fedora
pocketfft release_for_eigen (2021-03-12) Packaged trix pocketfft pocketfft
protobuf d1eca4e (2020-10-08) Packaged? protobuf protobuf after v3.13.0, before v3.14.0-rc1
psimd current master/HEAD (2020-05-17) Packaged trix psimd psimd
pthreadpool a134dd5 (2021-04-13) Packaged trix
pybind11 2.11.1 Packaged? pybind11 pybind11
python-peachpy f45429b This is a fork of Maratyszcza/PeachPy
six no specified version Packaged? python-six python-six
sleef e0a003e (2020-12-21) Packaged? sleef sleef
tbb TBB 2018 U6? (2018-10-09) Packaged? tbb tbb
tensorflow_cuda_bazel_build might be part of the pytorch build process
tensorpipe 52791a2 Archived upstream on 2023-07-01
XNNPACK 51a9875 Packaged trix
zstd v1.3.2 + 1 month Packaged? zstd zstd