- update to 2.3.1 with following summarized highlights:
* from 2.0.x:
- torch.compile is the main API for PyTorch 2.0, which wraps your model and
returns a compiled model. It is a fully additive (and optional) feature
and hence 2.0 is 100% backward compatible by definition
- Accelerated Transformers introduce high-performance support for training
and inference using a custom kernel architecture for scaled dot product
attention (SPDA). The API is integrated with torch.compile() and model
developers may also use the scaled dot product attention kernels directly
by calling the new scaled_dot_product_attention() operato
* from 2.1.x:
- automatic dynamic shape support in torch.compile,
torch.distributed.checkpoint for saving/loading distributed training jobs
on multiple ranks in parallel, and torch.compile support for the NumPy
API.
- In addition, this release offers numerous performance improvements (e.g.
CPU inductor improvements, AVX512 support, scaled-dot-product-attention
support) as well as a prototype release of torch.export, a sound
full-graph capture mechanism, and torch.export-based quantization.
* from 2.2.x:
- 2x performance improvements to scaled_dot_product_attention via
FlashAttention-v2 integration, as well as AOTInductor, a new
ahead-of-time compilation and deployment tool built for non-python
server-side deployments.
* from 2.3.x:
- support for user-defined Triton kernels in torch.compile, allowing for
users to migrate their own Triton kernels from eager without
experiencing performance complications or graph breaks. As well, Tensor
Parallelism improves the experience for training Large Language Models
using native PyTorch functions, which has been validated on training
OBS-URL: https://build.opensuse.org/package/show/science:machinelearning/python-torch?expand=0&rev=32
2024-07-19 14:15:19 +02:00
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Thu Jul 11 09:37:17 UTC 2024 - Christian Goll <cgoll@suse.com>
|
|
|
|
|
|
|
|
|
|
|
|
- update to 2.3.1 with following summarized highlights:
|
|
|
|
|
|
* from 2.0.x:
|
|
|
|
|
|
- torch.compile is the main API for PyTorch 2.0, which wraps your model and
|
|
|
|
|
|
returns a compiled model. It is a fully additive (and optional) feature
|
|
|
|
|
|
and hence 2.0 is 100% backward compatible by definition
|
|
|
|
|
|
- Accelerated Transformers introduce high-performance support for training
|
|
|
|
|
|
and inference using a custom kernel architecture for scaled dot product
|
|
|
|
|
|
attention (SPDA). The API is integrated with torch.compile() and model
|
|
|
|
|
|
developers may also use the scaled dot product attention kernels directly
|
|
|
|
|
|
by calling the new scaled_dot_product_attention() operato
|
|
|
|
|
|
* from 2.1.x:
|
|
|
|
|
|
- automatic dynamic shape support in torch.compile,
|
|
|
|
|
|
torch.distributed.checkpoint for saving/loading distributed training jobs
|
|
|
|
|
|
on multiple ranks in parallel, and torch.compile support for the NumPy
|
|
|
|
|
|
API.
|
|
|
|
|
|
- In addition, this release offers numerous performance improvements (e.g.
|
|
|
|
|
|
CPU inductor improvements, AVX512 support, scaled-dot-product-attention
|
|
|
|
|
|
support) as well as a prototype release of torch.export, a sound
|
|
|
|
|
|
full-graph capture mechanism, and torch.export-based quantization.
|
|
|
|
|
|
* from 2.2.x:
|
|
|
|
|
|
- 2x performance improvements to scaled_dot_product_attention via
|
|
|
|
|
|
FlashAttention-v2 integration, as well as AOTInductor, a new
|
|
|
|
|
|
ahead-of-time compilation and deployment tool built for non-python
|
|
|
|
|
|
server-side deployments.
|
|
|
|
|
|
* from 2.3.x:
|
|
|
|
|
|
- support for user-defined Triton kernels in torch.compile, allowing for
|
|
|
|
|
|
users to migrate their own Triton kernels from eager without
|
|
|
|
|
|
experiencing performance complications or graph breaks. As well, Tensor
|
|
|
|
|
|
Parallelism improves the experience for training Large Language Models
|
|
|
|
|
|
using native PyTorch functions, which has been validated on training
|
|
|
|
|
|
runs for 100B parameter models.
|
|
|
|
|
|
- added seperate openmpi4 build
|
|
|
|
|
|
- added sepetate vulcan build, although this functions isn't exposed to python
|
|
|
|
|
|
abi
|
|
|
|
|
|
- For the obs build all the vendored sources follow the pattern
|
|
|
|
|
|
NAME-7digitcommit.tar.gz and not the NAME-COMMIT.tar.gz
|
|
|
|
|
|
- added following patches:
|
|
|
|
|
|
* skip-third-party-check.patch
|
|
|
|
|
|
* fix-setup.patch
|
|
|
|
|
|
- removed patches:
|
|
|
|
|
|
* pytorch-rm-some-gitmodules.patch
|
|
|
|
|
|
* fix-call-of-onnxInitGraph.patch
|
|
|
|
|
|
|
2022-05-24 07:53:14 +02:00
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Thu Jul 22 14:40:45 UTC 2021 - Guillaume GARDET <guillaume.gardet@opensuse.org>
|
|
|
|
|
|
|
|
|
|
|
|
- Fix build on x86_64 by using GCC10 instead of GCC11
|
|
|
|
|
|
https://github.com/google/XNNPACK/issues/1550
|
|
|
|
|
|
|
|
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Thu Jul 22 10:11:03 UTC 2021 - Guillaume GARDET <guillaume.gardet@opensuse.org>
|
|
|
|
|
|
|
|
|
|
|
|
- Update to 1.9.0
|
|
|
|
|
|
- Release notes: https://github.com/pytorch/pytorch/releases/tag/v1.9.0
|
|
|
|
|
|
- Drop upstreamed patch:
|
|
|
|
|
|
* fix-mov-operand-for-gcc.patch
|
|
|
|
|
|
- Drop unneeded patches:
|
|
|
|
|
|
* removed-peachpy-depedency.patch
|
|
|
|
|
|
- Refresh patches:
|
|
|
|
|
|
* skip-third-party-check.patch
|
|
|
|
|
|
* fix-call-of-onnxInitGraph.patch
|
|
|
|
|
|
- Add new patch:
|
|
|
|
|
|
* pytorch-rm-some-gitmodules.patch
|
|
|
|
|
|
|
|
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Thu Jul 22 09:07:31 UTC 2021 - Guillaume GARDET <guillaume.gardet@opensuse.org>
|
|
|
|
|
|
|
|
|
|
|
|
- Add _service file to ease future update of deps
|
|
|
|
|
|
|
|
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Thu Jul 22 08:26:17 UTC 2021 - Guillaume GARDET <guillaume.gardet@opensuse.org>
|
|
|
|
|
|
|
|
|
|
|
|
- Update sleef to fix build on aarch64
|
|
|
|
|
|
|
|
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Fri Apr 23 21:51:36 UTC 2021 - Matej Cepl <mcepl@suse.com>
|
|
|
|
|
|
|
|
|
|
|
|
- Don't build python36-* package (missing pandas)
|
|
|
|
|
|
|
|
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Thu Jan 21 23:28:20 UTC 2021 - Benjamin Greiner <code@bnavigator.de>
|
|
|
|
|
|
|
|
|
|
|
|
- Fix python-rpm-macros usage
|
|
|
|
|
|
|
|
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Wed Oct 7 15:15:28 UTC 2020 - Guillaume GARDET <guillaume.gardet@opensuse.org>
|
|
|
|
|
|
|
|
|
|
|
|
- Use GCC9 to build on aarch64 Tumbleweed to workaround SVE
|
|
|
|
|
|
problem with GCC10 with sleef, see:
|
|
|
|
|
|
https://github.com/pytorch/pytorch/issues/45971
|
|
|
|
|
|
|
|
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Thu Aug 20 09:04:08 UTC 2020 - Martin Liška <mliska@suse.cz>
|
|
|
|
|
|
|
|
|
|
|
|
- Use memoryperjob constraint instead of %limit_build macro.
|
|
|
|
|
|
|
|
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Tue Jun 23 15:28:57 UTC 2020 - Christian Goll <cgoll@suse.com>
|
|
|
|
|
|
|
|
|
|
|
|
- updated to new stable release 1.5.1 which has following changes:
|
|
|
|
|
|
This release includes several major new API additions and improvements. These
|
|
|
|
|
|
include new APIs for autograd allowing for easy computation of hessians and
|
|
|
|
|
|
jacobians, a significant update to the C++ frontend, ‘channels last’ memory
|
|
|
|
|
|
format for more performant computer vision models, a stable release of the
|
|
|
|
|
|
distributed RPC framework used for model parallel training, and a new API
|
|
|
|
|
|
that allows for the creation of Custom C++ Classes that was inspired by
|
|
|
|
|
|
PyBind. Additionally torch_xla 1.5 is now available and tested with the
|
|
|
|
|
|
PyTorch 1.5 release providing a mature Cloud TPU experience.
|
|
|
|
|
|
* see release.html for detailed information
|
|
|
|
|
|
- added patches:
|
|
|
|
|
|
* fix-call-of-onnxInitGraph.patch for API mismatch in onnx
|
|
|
|
|
|
* fix-mov-operand-for-gcc.patch for aarch64 operands
|
|
|
|
|
|
|
|
|
|
|
|
- removed sources:
|
|
|
|
|
|
* cpuinfo-89fe1695edf9ee14c22f815f24bac45577a4f135.tar.gz
|
|
|
|
|
|
* gloo-7c541247a6fa49e5938e304ab93b6da661823d0f.tar.gz
|
|
|
|
|
|
* onnx-fea8568cac61a482ed208748fdc0e1a8e47f62f5.tar.gz
|
|
|
|
|
|
* psimd-90a938f30ba414ada2f4b00674ee9631d7d85e19.tar.gz
|
|
|
|
|
|
* pthreadpool-13da0b4c21d17f94150713366420baaf1b5a46f4.tar.gz
|
|
|
|
|
|
- added sources:
|
|
|
|
|
|
* cpuinfo-0e6bde92b343c5fbcfe34ecd41abf9515d54b4a7.tar.gz
|
|
|
|
|
|
* gloo-113bde13035594cafdca247be953610b53026553.tar.gz
|
|
|
|
|
|
* onnx-9fdae4c68960a2d44cd1cc871c74a6a9d469fa1f.tar.gz
|
|
|
|
|
|
* psimd-10b4ffc6ea9e2e11668f86969586f88bc82aaefa.tar.gz
|
|
|
|
|
|
* pthreadpool-d465747660ecf9ebbaddf8c3db37e4a13d0c9103.tar.gz
|
|
|
|
|
|
|
|
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Tue Jun 23 09:25:06 UTC 2020 - Christian Goll <cgoll@suse.com>
|
|
|
|
|
|
|
|
|
|
|
|
- updated to bugfix release 1.4.1 and added _multibuild file so
|
|
|
|
|
|
that cuda versions can be build on commandline
|
|
|
|
|
|
|
|
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Thu Apr 23 14:30:22 UTC 2020 - Tomáš Chvátal <tchvatal@suse.com>
|
|
|
|
|
|
|
|
|
|
|
|
- Make sure to pull py2/py3 package from the devel pkg
|
|
|
|
|
|
|
|
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Thu Apr 23 09:54:25 UTC 2020 - Tomáš Chvátal <tchvatal@suse.com>
|
|
|
|
|
|
|
|
|
|
|
|
- Do not pull in python2 only dependencies
|
|
|
|
|
|
|
2020-02-26 14:47:31 +01:00
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Wed Feb 26 13:07:14 UTC 2020 - Simon Lees <sflees@suse.de>
|
|
|
|
|
|
|
|
|
|
|
|
- Exclude i586 builds for now, they fail with a cryptic return
|
|
|
|
|
|
code of 1 from cmake from python.
|
|
|
|
|
|
|
2020-01-14 15:16:11 +01:00
|
|
|
|
-------------------------------------------------------------------
|
2020-02-21 16:50:33 +01:00
|
|
|
|
Fri Feb 21 14:15:00 UTC 2020 - Christian Goll <cgoll@suse.com>
|
2020-01-14 15:16:11 +01:00
|
|
|
|
|
2020-02-21 16:50:33 +01:00
|
|
|
|
- updated to stable release 1.4.0, which has as Highlights:
|
|
|
|
|
|
* Distributed Model Parallel Training
|
|
|
|
|
|
* Pruning functionalities have been added to PyTorch
|
|
|
|
|
|
- New Features:
|
|
|
|
|
|
* torch.optim.lr_scheduler now support “chaining.”
|
|
|
|
|
|
* torch.distributed.rpc is a newly introduced package
|
|
|
|
|
|
- full Changelog listed in relases file or under
|
|
|
|
|
|
https://github.com/pytorch/pytorch/releases
|
2020-02-25 15:08:41 +01:00
|
|
|
|
and in the releases.hml file
|
2020-02-21 16:50:33 +01:00
|
|
|
|
- added files:
|
|
|
|
|
|
* skip-third-party-check.patch which is a patch to skip
|
|
|
|
|
|
the check of disabled dependencies
|
|
|
|
|
|
* QNNPACK-7d2a4e9931a82adc3814275b6219a03e24e36b4c.tar.gz
|
|
|
|
|
|
which is part of pytorch but developed in different repo
|
2020-02-25 15:08:41 +01:00
|
|
|
|
* releases.html which is the downloaded releases file
|
2020-02-21 16:50:33 +01:00
|
|
|
|
- removed patch files:
|
|
|
|
|
|
* fix-build-options.patch
|
|
|
|
|
|
* honor-PSIMD-env.patch
|
|
|
|
|
|
* removed-some-tests.patch
|
2020-01-14 15:16:11 +01:00
|
|
|
|
|
2020-02-21 16:50:33 +01:00
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Tue Jan 14 13:01:33 UTC 2020 - Guillaume GARDET <guillaume.gardet@opensuse.org>
|
|
|
|
|
|
|
|
|
|
|
|
- Requires python-PeachPy on x86_64 only, as it is optional
|
|
|
|
|
|
and available on x86_64 only
|
|
|
|
|
|
|
2020-02-25 13:33:19 +01:00
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Wed Jan 8 10:47:18 UTC 2020 - Christian Goll <cgoll@suse.com>
|
|
|
|
|
|
|
|
|
|
|
|
- updated the requirement for examples and converters
|
|
|
|
|
|
|
2019-08-28 13:58:32 +02:00
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Wed Jun 12 11:17:34 UTC 2019 - Christian Goll <cgoll@suse.com>
|
|
|
|
|
|
|
|
|
|
|
|
- Updated to stable version 1.1.0, which needed also updates of
|
|
|
|
|
|
following dependend sources:
|
|
|
|
|
|
* onnx-1.4.1.tar.gz ->
|
|
|
|
|
|
onnx-22662bfd4dcc6baebf29e3b823a051676f991001.tar.gz
|
|
|
|
|
|
- Removed following sources:
|
|
|
|
|
|
* FBGEMM-f65f0ebe54f0512d8f42ee10025b596e3f42e0b8.tar.gz
|
|
|
|
|
|
- Added following sources:
|
|
|
|
|
|
* foxi-8f74bc4df3a4cfc69b1a3eadf62aa29d9961c72d.tar.gz
|
|
|
|
|
|
- Changed patch
|
|
|
|
|
|
* fix-build-options.patch to work with new buid system and
|
|
|
|
|
|
exclude FBGEMM
|
|
|
|
|
|
- Added patch:
|
|
|
|
|
|
* honor-PSIMD-env.patch, which makes depend sources of pytorch
|
|
|
|
|
|
to use the source of psimd
|
|
|
|
|
|
|
|
|
|
|
|
-------------------------------------------------------------------
|
|
|
|
|
|
Tue Mar 26 09:33:11 UTC 2019 - Christian Goll <cgoll@suse.com>
|
|
|
|
|
|
|
|
|
|
|
|
- Inital commit of pytorch/caffe2 which is an opensource
|
|
|
|
|
|
machineleraning platform. This is the stable release 1.0.1
|
|
|
|
|
|
including like other tools a lot of third party sources,
|
|
|
|
|
|
which could not be used from the base system due to messy
|
|
|
|
|
|
build system. Additional sources are
|
|
|
|
|
|
* gloo, a communitcation library for GPUs as
|
|
|
|
|
|
gloo-670b4d4aa46886cc66874e2a4dc846f5cfc2a285.tar.gz
|
|
|
|
|
|
* fbgemm, a low precission, high peformance matrix lib
|
|
|
|
|
|
FBGEMM-f65f0ebe54f0512d8f42ee10025b596e3f42e0b8.tar.gz
|
|
|
|
|
|
* cpuinfo, a cross platform cpu information tool
|
|
|
|
|
|
cpuinfo-89fe1695edf9ee14c22f815f24bac45577a4f135.tar.gz
|
|
|
|
|
|
* sleef, a function for elementary functions
|
|
|
|
|
|
sleef-191f655caa25526ae226cf88dd2529265176014a.tar.gz
|
|
|
|
|
|
* pytbind11, which exposes C/C++ headers to pythob, but
|
|
|
|
|
|
the source code of this library is deeply integrated into
|
|
|
|
|
|
pytorch, so we need
|
|
|
|
|
|
pybind11-25abf7efba0b2990f5a6dfb0a31bc65c0f2f4d17.tar.gz
|
|
|
|
|
|
* onnx, which is an format for exchaning neural networks as
|
|
|
|
|
|
onnx-1.4.1.tar.gz
|
|
|
|
|
|
* pthreadpool, a pthread based thread tool implementation, which
|
|
|
|
|
|
can be used when omp is not available
|
|
|
|
|
|
pthreadpool-13da0b4c21d17f94150713366420baaf1b5a46f4.tar.gz
|
|
|
|
|
|
* FXdiv, a Header-only library for division via fixed-point
|
|
|
|
|
|
multiplication by inverse, which has no stable API atm, so
|
|
|
|
|
|
FXdiv-b742d1143724d646cd0f914646f1240eacf5bd73.tar.gz
|
|
|
|
|
|
* psimd, portable 128-bit SIMD intrinsics
|
|
|
|
|
|
psimd-90a938f30ba414ada2f4b00674ee9631d7d85e19.tar.gz
|
|
|
|
|
|
* fp16, a numeric conversion library
|
|
|
|
|
|
FP16-febbb1c163726b5db24bed55cc9dc42529068997.tar.gz
|
|
|
|
|
|
* gemmlowp, self-contained low-precision GEMM library as
|
|
|
|
|
|
gemmlowp-8416bab644641a5c0a81ecf91a5cda804af0aee1.tar.gz
|
|
|
|
|
|
* fix-build-options.patch, which points pytorch to system libs
|
|
|
|
|
|
* removed-peachpy-depedency.patch, which forces to use system
|
|
|
|
|
|
peachpy
|
|
|
|
|
|
|
