Script 'mail_helper' called by obssrc Hello community, here is the log from the commit of package python-torchao for openSUSE:Factory checked in at 2026-08-25 13:20:01 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ Comparing /work/SRC/openSUSE:Factory/python-torchao (Old) and /work/SRC/openSUSE:Factory/.python-torchao.new.1258 (New) ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Package is "python-torchao" Tue Aug 25 13:20:01 2026 rev:2 rq:1373473 version:0.18.0 Changes: -------- --- /work/SRC/openSUSE:Factory/python-torchao/python-torchao.changes 2026-08-24 15:43:44.282965214 +0200 +++ /work/SRC/openSUSE:Factory/.python-torchao.new.1258/python-torchao.changes 2026-08-25 13:20:12.543210035 +0200 @@ -1,0 +2,23 @@ +Mon Aug 24 18:03:54 UTC 2026 - Martin Pluskal <[email protected]> + +- Update to 0.18.0: + * Minimum supported PyTorch is now 2.11; support for older + releases was removed from the code + * Removed the deprecated v1 tensor-subclass and layout system + (AffineQuantizedTensor, AQTTensorImpl, Layout, PlainLayout, + TensorCoreTiledLayout, LinearActivationQuantizedTensor and + friends), completing the migration to v2 tensor subclasses + * Renamed Int4WeightOnlyOpaqueTensorConfig to + PrototypeInt4WeightOnlyConfig and moved NF4Tensor to + torchao.quantization.quantize_.workflows + * New prototype NVFP4 training support for dense linears + * Added MXFP8Linear plus blockwise FP8 MoE support + * Float8 softmax quantization now uses a fixed scale instead of + an observer + * Building under a free-threaded (no-GIL) interpreter is now + supported + * Fixed Python 3.14 compatibility of quantization/pt2e + * Numerous QAT, sparsity and pruning fixes +- Require python-torch >= 2.11 to match the new upstream floor + +------------------------------------------------------------------- Old: ---- python-torchao-0.17.0.tar.gz New: ---- python-torchao-0.18.0.tar.gz ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ Other differences: ------------------ ++++++ python-torchao.spec ++++++ --- /var/tmp/diff_new_pack.SivpxH/_old 2026-08-25 13:20:14.846291354 +0200 +++ /var/tmp/diff_new_pack.SivpxH/_new 2026-08-25 13:20:14.854291636 +0200 @@ -18,7 +18,7 @@ %{?sle15_python_module_pythons} Name: python-torchao -Version: 0.17.0 +Version: 0.18.0 Release: 0 Summary: PyTorch native quantization and sparsity # Legal-Review-Notice: root LICENSE is BSD-3-Clause. prototype/paretoq @@ -31,13 +31,14 @@ Source: https://github.com/pytorch/ao/archive/refs/tags/v%{version}.tar.gz#/%{name}-%{version}.tar.gz BuildRequires: %{python_module pip} BuildRequires: %{python_module setuptools} -BuildRequires: %{python_module torch} +# upstream dropped support for PyTorch older than 2.11 in 0.18.0 +BuildRequires: %{python_module torch >= 2.11} BuildRequires: %{python_module wheel} BuildRequires: fdupes BuildRequires: python-rpm-macros # setup.py does not declare install_requires; pythondistdeps sees no # Requires-Dist in the wheel METADATA -Requires: python-torch +Requires: python-torch >= 2.11 BuildArch: noarch %python_subpackages ++++++ python-torchao-0.17.0.tar.gz -> python-torchao-0.18.0.tar.gz ++++++ /work/SRC/openSUSE:Factory/python-torchao/python-torchao-0.17.0.tar.gz /work/SRC/openSUSE:Factory/.python-torchao.new.1258/python-torchao-0.18.0.tar.gz differ: char 13, line 1
