Index index by Group index by Distribution index by Vendor index by creation date index by Name Mirrors Help Search

libtorch-openmpi4-2.12.0-11.2 RPM for riscv64

From OpenSuSE Ports Tumbleweed for riscv64

Name: libtorch-openmpi4 Distribution: openSUSE Tumbleweed
Version: 2.12.0 Vendor: openSUSE
Release: 11.2 Build date: Mon Jul 13 09:57:40 2026
Group: Development/Libraries/Python Build host: reproducible
Size: 114665344 Source RPM: python-torch-openmpi4-2.12.0-11.2.src.rpm
Packager: https://bugs.opensuse.org
Url: https://pytorch.org
Summary: Library which used by python-torch-openmpi4
Library which is used by python-torch-openmpi4

Provides

Requires

License

Apache-2.0 AND BSD-2-Clause AND BSD-3-Clause AND MIT AND Zlib AND BSL-1.0

Changelog

* Mon Jul 13 2026 Martin Pluskal <mpluskal@suse.com>
  - Fix Caffe2Targets.cmake / find_package(Torch) referencing
    libc10/libtorch/libtorch_cpu at the pre-move site-packages path,
    which broke every downstream C++ torch consumer: after the
    libraries are moved to %{_libdir} for the separate libtorch
    package, ship compatibility symlinks under torch/lib so the
    relative CMake config (${_IMPORT_PREFIX}/lib and
    ${TORCH_INSTALL_PREFIX}/lib) resolves again.
* Mon Jun 15 2026 Guillaume GARDET <guillaume.gardet@opensuse.org>
  - Backport upstream patch to fix build on aarch64:
    * xnnpack-aarch64-fix.patch
* Mon Jun 08 2026 Christian Goll <cgoll@suse.com>
  - Update to v2.12.0 which uses _service and so does the vendoring via
    git submodules, this leads to the removal of following previosly vendored
    sources and patches:
    * FP16-4dfe081.tar.gz
    * FXdiv-b408327.tar.gz
    * XNNPACK-51a0103.tar.gz
    * add-cstdint.patch
    * cpp-httplib-3b6597b.tar.gz
    * cpuinfo-1e83a2f.tar.gz
    * flatbuffers-01834de.tar.gz
    * fmt-1239137.tar.gz
    * gemmlowp-3fb5c17.tar.gz
    * gloo-5354032.tar.gz
    * gloo-gcc15-fix.patch
    * kineto-a054a4b.tar.gz
    * kleidiai-cca02c2.tar.gz
    * libnop-910b558.tar.gz
    * onnx-b8baa84.tar.gz
    * opentelemetry-cpp-a799f4a.tar.gz
    * pocketfft-9d3ab05.tar.gz
    * psimd-072586a.tar.gz
    * pthreadpool-4fe0e1e.tar.gz
    * pybind11-a2e59f0.tar.gz
    * pytorch-2.7.1.tar.gz
    * sleef-56e1f79.tar.gz
    * tensorpipe-52791a2.tar.gz
    * fix-setup.patch
  - Added disable-py312-intrinsics.patch which disables test for incomaptible python3.12
  - increased limits for build machines
  - Highlights of v2.12.0:
    * Batched linalg.eigh on CUDA is up to 100x faster due to updated cuSolver
      backend selection
    * Added new torch.accelerator.Graph API to unify graph capture and replay
      across CUDA, XPU, and out-of-tree backends
    * Added torch.export.save support for Microscaling (MX) quantization formats
    * Adagrad optimizer now supports fused=True for a single-kernel optimizer
    * torch.cond control flow can now be captured and replayed inside CUDA Graphs
    * ROCm improvements include expandable memory segments, rocSHMEM symmetric
      memory collectives, and FlexAttention pipelining
    * Python 3.14 support: Added experimental support for torch.compile()
  - Backwards Incompatible & Breaking Changes in v2.12.0:
    * C++20 minimum is now enforced in CMake build files
    * SVE Compile Checks strictly validated in FindARM.cmake
    * torch.distributed.nn.functional ops now raise RuntimeError under torch.compile
      (users should migrate to functional collectives API)
    * torchrun port default is now dynamically assigned (OS-allocated free port)
      instead of static port 29500
    * Deprecated CUDA 12.8 builds in favor of CUDA 13.0
    * MAGMA backend deprecated for some operators, dispatching to cuSolver or cuBLAS
  - Summarized highlights and breaking changes from intermediate versions (v2.8.0 to v2.11.0):
    * Highlights from v2.11.0:
    - Differentiable Collectives support for functional collectives in distributed
      training
    - Expanded operator support for Metal Performance Shaders (MPS) on XPU Graph support
      for Intel XPU backends
    * Backwards Incompatible & Breaking Changes from v2.11.0:
    - DLPack objects updated to align with the latest DLPack release,
      potentially breaking low-level integrations using torch.utils.dlpack
    * Highlights from v2.10.0:
    - Full support for torch.compile() on Python 3.14, with experimental support
      for free-threaded builds (Python 3.14t)
    - New DebugMode tool for tracking dispatched calls and debugging numerical
      divergence; horizontal fusion of combo-kernels to reduce launch overhead
      in torchinductor
    * Backwards Incompatible & Breaking Changes from v2.10.0:
    - Sampler data_source argument removed from DataLoader, and legacy grouping
      datapipes imports removed (use sharding datapipes instead)
    - fallback=False is now the default in torch.onnx.export (defaults to
      dynamo=True without fallback)
    - Nested Jagged Tensor support removed from flex_attention
    - TorchScript (torch.jit) is officially deprecated and not guaranteed to work
      on Python 3.14
    * Highlights from v2.9.0:
    - Symmetric Memory feature enabling high-performance programming of multi-GPU
      kernels
    - Stable libtorch ABI updates with torch::stable::Tensor APIs and device/stream
      utilities
    - Arbitrary toggle between error or resume on graph breaks in torch.compile for
      debuggability
    * Backwards Incompatible & Breaking Changes from v2.9.0:
    - Minimum supported Python version updated to 3.10 (Python 3.9 support dropped)
    - DLPack upgraded to 1.0, breaking torch.utils.dlpack low-level APIs
    - Outputs of custom operators registered via torch.library are no longer allowed
      to share storage with input Tensors (can cause undefined behavior under compile)
    - Exception type changes where torch.cat and unsupported dtypes raise specific
      errors (e.g., ValueError, IndexError, NotImplementedError) instead of RuntimeError
    * Highlights from v2.8.0:
    - Introduced limited stable libtorch ABI subset for third-party extensions
    - Native support for high-performance quantized LLM inference on Intel CPUs
    - Inductor CUTLASS backend support for torch.compile and AOTInductor
    * Backwards Incompatible & Breaking Changes from v2.8.0:
    - Dropped Maxwell (sm50) and Pascal (sm60, sm61) GPU support in CUDA 12.8+ builds
      (older GPUs must use CUDA 12.6)
    - Stricter SVE validation in CMake and header cleanup removing implicit
      torch/types.h from Dispatcher.h
    - Error triggered if custom autograd.Function mutates a view of a leaf requiring
      gradients
* Thu Jul 10 2025 Guillaume GARDET <guillaume.gardet@opensuse.org>
  - Add kleidiai dep (vendored)
  - Workaround build error on aarch64 with -Wno-incompatible-pointer-types
* Fri Jul 04 2025 Christian Goll <cgoll@suse.com>
  - Updated to 2.7.1 with following fiex:
    * Fix assertion error due to inductor permuting inputs to flex attention (#151959)
    * Fix performance regression on nanogpt speedrun (#152641)
    * Fix improve PyTorch Wheel size due to introduction of addition of 128 bit vectorization (#148320) (#152396)
    * Fix fmsub function definition (#152075)
    * Fix Floating point exception in torch.mkldnn_max_pool2d (#151848)
    * Fix ONNX decomposition does not preserve custom CompositeImplicitAutograd ops (#151826)
    * Fix error with dynamic linking of libgomp library (#150084)
    * Fix segfault in profiler with Python 3.13 (#153848)
  - Changes from 2.7.0:
    * torch.onnx.dynamo_export now uses the ExportedProgram logic path (#137296)
      Users using the torch.onnx.dynamo_export API may see some ExportOptions
      become unsupported due to an internal switch to use torch.onnx.export(...,
      dynamo=True): diagnostic_options, fake_context and onnx_registry are
      removed/ignored by ExportOptions. Only dynamic_shapes is retained.
    * Finish deprecation of LRScheduler.print_lr() along with the verbose kwarg
      to the LRScheduler constructor. (#147301)
    * libtorch_python.so symbols are now invisible by default on all platforms (#142214)
    * Please use torch.export.export instead of capture_pre_autograd_graph to
      export the model for pytorch 2 export quantization (#139505)
    * New interface for
      torch.fx.passes.graph_transform_observer.GraphTransformObserver to enable
      Node Level provenance tracking (#144277)
    * torch.ao.quantization.pt2e.graph_utils.get_control_flow_submodules is no
      longer public (#141612)
    * torch.onnx.dynamo_export is deprecated (#146425, #146639, #146923)
    * XNNPACKQuantizer is deprecated in PyTorch and moved to ExecuTorch, please
      use it from executorch.backends.xnnpack.quantizer.xnnpack_quantizer instead
      of torch.ao.quantization.quantizer.xnnpack_quantizer. (#144940)
  - Changes from 2.6.0
    * [Beta] torch.compiler.set_stance
    * [Beta] torch.library.triton_op
    * [Beta] torch.compile support for Python 3.13
    * [Beta] New packaging APIs for AOTInductor
    * [Beta] New packaging APIs for AOTInductor
    * [Beta] AOTInductor: minifier
    * [Beta] AOTInductor: ABI-compatible mode code generation
    * [Beta] FP16 support for X86 CPUs (both eager and Inductor modes)
    * FlexAttention support on X86 CPU for LLMs
  - Added patches for gcc15 compatibilty:
    * add-cstdint.patch
    * gloo-gcc15-fix.patch
  - Updated vendored sources:
    * kineto-d975313.tar.gz -> kineto-a054a4b.tar.gz
    * onnx-3bf92c0.tar.gz -> onnx-b8baa84.tar.gz
    * pybind11-7c33cdc.tar.gz -> pybind11-a2e59f0.tar.gz
    * sleef-60e76d2.tar.gz -> sleef-56e1f79.tar.gz
    * XNNPACK-fcbf55a.tar.gz -> XNNPACK-51a0103.tar.gz
    * cpuinfo-fa1c679.tar.gz -> cpuinfo-1e83a2f.tar.gz
    * fmt-e69e5f9.tar.gz -> fmt-1239137.tar.gz
* Tue Dec 17 2024 Andreas Schwab <schwab@suse.de>
  - Use oneDNN only on x86_64 aarch64 ppc64le
* Fri Oct 18 2024 Guillaume GARDET <guillaume.gardet@opensuse.org>
  -  Update to 2.5.0:
    * https://github.com/pytorch/pytorch/releases/tag/v2.5.0
* Fri Oct 04 2024 Guillaume GARDET <guillaume.gardet@opensuse.org>
  - Add patch to fix build with oneDNN:
    * pytorch-patch-onednn.patch
* Tue Oct 01 2024 Guillaume GARDET <guillaume.gardet@opensuse.org>
  - Update to 2.4.1:
    * https://github.com/pytorch/pytorch/releases/tag/v2.4.1
  - Skip update to 2.4.0:
    * https://github.com/pytorch/pytorch/releases/tag/v2.4.0
  - Remove _service since 'osc mr download_files' is easier to use
    and maintain
  - Drop config vars not used anymore: BUILD_CAFFE2, USE_LEVELDB, USE_LMDB,
    USE_OPENCV, USE_TBB
  - Remove examples package since code has been removed upstream
  - Refresh pacth:
    * skip-third-party-check.patch
* Thu Aug 29 2024 Guang Yee <gyee@suse.com>
  - Enable sle15_python_module_pythons.
  - GCC 9.3 or newer is required, regardless if CUDA is enabled.
    See https://github.com/pytorch/pytorch/blob/v2.3.1/CMakeLists.txt#L48
    Therefore, for SLE15 we went with GCC 11 as it seems to be the most
    common one.
  - Use %gcc_version macro for Tumbleweed.
* Thu Jul 11 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

Files

/usr/lib64/libc10.so
/usr/lib64/libtorch.so
/usr/lib64/libtorch_cpu.so


Generated by rpm2html 1.8.1

Fabrice Bellet, Sun Aug 2 03:56:04 2026