pytorch (2.1.2+dfsg-4) unstable; urgency=medium * Autopkgtest: do not run python tests -- too slow on our CI. * Autopkgtest: refresh C++ test list. * Autopkgtest: Disable dangling symlink test (may fail due to other package). * rules: Remove the dangling libnnapi_backend.so symlink. * Drop B-D on python3-distutils (Closes: #1065941). -- Mo Zhou Wed, 24 Apr 2024 19:26:04 -0400 pytorch (2.1.2+dfsg-3) unstable; urgency=medium * Remove merged ppc64el FTBFS patch. * Remove the version requirement for libsleef-dev. * Update control file for CUDA variant. (Closes: #1064484) * Require cutlass >= 3.1.0. -- Mo Zhou Wed, 28 Feb 2024 12:06:31 -0500 pytorch (2.1.2+dfsg-2) unstable; urgency=medium [ Petter Reinholdtsen ] * Added patch numbering scheme README and rename patches to follow suit. [ Mo Zhou ] * Correct wrong patch numberings. * Control: Does not compile against sleef from git. * Upload to unstable. (Closes: #1060097) -- Mo Zhou Tue, 13 Feb 2024 13:22:09 -0500 pytorch (2.1.2+dfsg-1) experimental; urgency=medium [ Petter Reinholdtsen ] * Updated Standards-Version from 4.6.0.1 to 4.6.2. No changes needed. * Updated patch metadata based on git history. [ Mo Zhou ] * New upstream version 2.1.2+dfsg * README.Source: update dependency track list for v2.1.2 * Update embedded kineto to 49e854d805d916b2031e337763928d2f8d2e1fbf. - Delete unwanted binary files from embedded kineto. * Patch embedded kineto to use system lib. + Embed a partial copy of dynolog for the embedded kineto. Recursive git submodule is not fun at all. * Remove more unnecessary files from embedded kineto. * We will remove kineto and disable it for PyTorch/v2.2.X. * Update debian/copyright: * d/copyright: Remove mimalloc from third_party dir. (Files-Excluded) * d/copyright: update for third_party/nvfuser. * Remove embedded nvfuser from debian/. We use the one in third_party. * Update patches: * Rebase existing patches. - Auto-regenerate flatbuffers and deprecate the flatbuffers patch. * Patch functorch code to fix FTBFS. * Update debian/rules: * Fix cmake find_package LLVM by hinting with correct path. * d/rules: Update versions and correct the version. * Install libnnapi_backend.so to libtorch2.1. * Update debian/control and other debhelper files: * Bump ABI/SOVERSION from 2.0 to 2.1. Update Breaks+Replaces accordingly. * Require build-dep libcpuinfo-dev >= 0.0~git20230113.6481e8b~. * Require build-dep libonnx-dev >= 1.14.1~. * Require build-dep libgloo-dev (>= 0.0~git20230519.597accf~). * Require build-dep libdnnl-dev (>= 3.1.1~). * Require build-dep libideep-dev (>= 0.0~git20230825.6f4d653~). * Remove build-dep version requirement on libflatbuffers-dev. Flatbuffer files are automatically re-generated during build now. * Specific changes to variant. * Reduce CUDA_ARCH_LIST to 6.1;7.5;8.6 for smaller binary. * Upload to experimental. Prepare for transition. -- Mo Zhou Mon, 01 Jan 2024 00:26:26 -0500 pytorch (2.0.1+dfsg-5) unstable; urgency=medium * Get rid of the embedded copy of clog. * Patch cmake: do not install clog.h (Closes: #1052260) -- Mo Zhou Sun, 01 Oct 2023 10:03:47 -0400 pytorch (2.0.1+dfsg-4) unstable; urgency=medium * Don't install the work-in-progress dh_pytorch for now. * rules: Add more annotations to ease reading. * rules: Sort install commands. * Patch embedded nvfuser to have SOVERSION. * rules: Install the libnvfuser with SOVERSION. This fixes the cannot load libnvfuser.* error message during import. * Do not install libnnapi_backend.so (not built). * Autopkgtest: make dangling symlink check permissive. (Closes: #1051828) -- Mo Zhou Wed, 13 Sep 2023 15:01:04 -0400 pytorch (2.0.1+dfsg-3) unstable; urgency=medium * rules: Copy googletest source instead of symlinking. * Add llvm-dev to B-D and toggle USE_LLVM. * Install the libnnapi_backend.so shlib. * rules: Automatically switch to mold/lld when they are installed. * Take back third_party/valgrind-headers. * Import the nvfuser thirdparty library. - remove doc directory from the embedded nvfuser. * Copyright: add entry for the embedded debian/nvfuser source code. * Variant: * Add the missing B-D libcupti-dev. * Specify USE_MAGMA = OFF. * Patch: fix cudnn_frontend include path. * Cherry-pick upstream commit to fix CUDA12 NVCC OOM. * rules: Update shlib list for install targets. * README.Source: update dependency tracking list -- Mo Zhou Fri, 08 Sep 2023 19:42:31 -0400 pytorch (2.0.1+dfsg-2) unstable; urgency=medium * Cherry-pick upstream fix for ppc64el. (Closes: #1051258) * The libcaffe2* symlinks should no longer exist as caffe2 is disabled (due to upstream deprecation) and no longer built. (Closes: #1035458) * Patch cmake template to fix including of build path. (Closes: #1031687) -- Mo Zhou Wed, 06 Sep 2023 15:50:27 -0400 pytorch (2.0.1+dfsg-1) unstable; urgency=medium * Control: add the missing Conflicts+Replaces packages. * Control: specify version requirement for pytorch-specific dependencies. * Files-Excluded: do not exclude third_party/nvfuser. We will need it for the CUDA version. * Upload to unstable. PyTorch has been broken in Sid for a while. -- Mo Zhou Mon, 04 Sep 2023 17:17:32 -0400 pytorch (2.0.1+dfsg-1~exp1) experimental; urgency=medium * New upstream version 2.0.1+dfsg (Closes: #1024903, #1037824, #1050175) * README.Source: update dependency tracking list * Copyright: - Exclude third party embedded copy from upstream source. + Go through all source files again and update the d/copyright, as the upstream made massive code base change in the 2.0 release. * Files-Excluded: Do not remove valgrind-headers. These files will be included at the next gbp-import-orig. * Packages: * Bump SOVERSION from 1.13 to 2.0. * Patch work updates: * Rebase existing patches, and remove unused ones. + Add patch to fix ftbfs with GCC-13. * Build modifications: + Specify -Wno-dangling-reference compiler flag. + rules: Disable valgrind for build. * Patch setup.py to skip installing valgrind headers. * rules: Export PYTORCH_BUILD_VERSION. * rules: No longer install the dangling include/TH symlink. * Embedded source copies: * Upgrade embedded kineto to new snapshot. + Patch embedded kineto to use system lib. + Patch the embedded kineto to fix GCC-13 ftbfs. - delete some unneeded files from embedded kineto. - remove th_plugin directory from libkineto. * Import dynolog partial snapshot for libkineto third_party. * Upgrade embedded expecttest to 0.1.6. (76f4af2667186271162e03817f2bbda782ec773d) * Upgrade embedded foxi to c278588e34e535f0bb8f00df3880d26928038cad. * Autopkgtest updates: * Autopkgtest: make the embedded expecttest module available. * Autopkgtest: correctly import embedded expecttest. * Autopkgtest: add the missing python3-scipy runtime deps. * Autopkgtest: add test cases to detect dangling symlinks * Autopkgtest: refresh test list using gencontrol.sh * Autopkgtest: add all python tests into permissive list. * Upload to experimental. -- Mo Zhou Sat, 02 Sep 2023 12:43:53 -0400 pytorch (1.13.1+dfsg-6) UNRELEASED; urgency=medium * cuda: minor fix to the sed pattern. * Enable Gloo and XNNPACK for riscv64 as well. * Cover more CUDA architectures. * run-cpptest: support pytorch-cuda as well. * Add run-pytest-core.sh * Embed https://github.com/ezyang/expecttest v0.1.4 for python testing. * run-{cpp,py}test.sh: improve scripts * run-pytest.sh: support --findall option * run-pytest.sh: do not exit upon fail * Autopkgtest: add back to core python tests. * Remove shebang.patch. Done automatically by sed in d/rules. * lintian: reduce duplicated overrides -- Mo Zhou Fri, 18 Aug 2023 13:33:53 -0400 pytorch (1.13.1+dfsg-5) experimental; urgency=medium * Overhaul the packaging for preparing different flavors of pytorch, which includes cpu-only (default), cuda, and rocm (not implemented yet). * The cuda version of control file is located at debian/control.cuda Lots of build dependencies are added. The cuda variant of pytorch must enter contrib due to non-free dependencies from Nvidia. * The script debian/cudabuild.sh is provided to convert the debian directory for the cuda build. * cuda: require cudnn >= 8.7.0.84~cuda11.8 for automatic build. * cuda: Build for CUDA architectures 6.0 7.0 8.0. * No longer build Caffe2 (it is already deprecated). * Update debian/patches to enable HAVE_SOVERSION for cuda shlibs. * Let the CPU/CUDA version conflicts/replaces each other. * Disale+Remove gflags and google-glog dependencies. * Update Lintian overrides. * watch: set dversionmangle as auto * copyright: Justify src:pytorch-cuda is the same source as src:pytorch. * rules: toggle USE_KINETO=ON * copyright: update * Patch embedded kineto to prevent further git submodule inclusion. * Toggle USE_DISTRIBUTED on any architecture. * Autopkgtest: support --cuda arg in test script. -- Mo Zhou Sat, 25 Mar 2023 23:15:08 -0400 pytorch (1.13.1+dfsg-4) unstable; urgency=medium [ Aron Xu ] * Disable CI on salsa, unable to complete a build [ Mo Zhou ] * Switch back to Generic BLAS against correctly build against it. We have update-alternatives for the BLAS/LAPACK backends. https://wiki.debian.org/DebianScience/LinearAlgebraLibraries * Only in this way can both Numpy and Torch consistently load the same BLAS implementation, lest bad things such as threading library collision happen. -- Mo Zhou Tue, 14 Mar 2023 13:21:37 -0400 pytorch (1.13.1+dfsg-3) unstable; urgency=medium [ Mo Zhou ] * fix cmake for Closes: #997931 [ Aron Xu ] * d/rules: more cmake fixes * d/control: build with libopenblas-pthread-dev instead of libblas-dev -- Aron Xu Wed, 08 Feb 2023 01:54:15 +0800 pytorch (1.13.1+dfsg-2) unstable; urgency=medium * Team upload. * d/tests/control: disable never passed py-tests -- Aron Xu Mon, 30 Jan 2023 21:54:24 +0800 pytorch (1.13.1+dfsg-1) unstable; urgency=medium [ Aron Xu ] * Team upload. * New upstream version 1.13.1 (Closes: #1027851) Fixes CVE-2022-45907. * d/rules: remove only build/ directory * d/rules: USE_ITT = OFF * Bump SOVERSION to 1.13 * d/control: add versioned build-dep to ideep * debian: carry kineto 20210224.0703c78 for header-only use * dirtyhack.patch: refresh and add hack of kineto * 80272.patch: merged upstream, removing * cherry-pick upstream py3.11 compatibility patches * Update autopkgtest control file * d/upstream/metadata: add Bug-Database [ Andreas Tille ] * Fix watch file [ Mo Zhou ] * Add fmtlib-revert.patch * Update flatbuffers patch -- Aron Xu Sat, 28 Jan 2023 23:31:42 +0800 pytorch (1.12.1-1) unstable; urgency=medium * New upstream version 1.12.1 * Rebase and refresh existing patches. -- Mo Zhou Fri, 30 Sep 2022 20:50:32 -0400 pytorch (1.12.0-1) unstable; urgency=medium * New upstream version 1.12.0 (June 28 2022) (Closes: #1004782, #1009200) * Autopkgtest: use openblas or blis instead of reference blas. * Patch to fix wrong shebang in several .py files. * Refresh existing patches. * Update lintian overrides. * Upload to unstable. -- Mo Zhou Mon, 12 Sep 2022 17:16:34 -0400 pytorch (1.12.0~rc1-1) experimental; urgency=medium * New upstream version 1.12.0~rc1 (May 2022) * Remove pretrained networks via Files-Excluded. * Cherry-pick upstream PR 80272 for pybind11 in cmake. * Embed pocketfft ea778e37710c07723435b1be58235996d1d43a5a. * Revise debian/{rules,control} to reflect the following changes: - No plan to enable FBGEMM, NNPACK, etc. - Specify USE_KINETO=OFF * Reflect upstream SOVERSION bump to 1.12 + Build-depends on flatbuffers and re-generate cpp from .fbs file. (use --cpp --gen-all --gen-mutable for flatc). * Rebase existing patches and add new patches: * Rebase dirtyhack.patch * Rebase mkldnn.patch, also disable ONEDNN_GRAPH. * Rebase shebang.patch * Rebase pytorch_glog_update.patch * Change include path for pocketfft. + Add patch for flatbuffers v2.0.6 migration * Autopkgtest: update tests. - Temprarily disable python autopkgtests due to missing dep. + Autopkgtest: add a simple sequential model test. * Update d/copyright and some lintian overrides. -- Mo Zhou Mon, 27 Jun 2022 20:45:06 -0700 pytorch (1.8.1-5) unstable; urgency=medium * Patch for compatibility with glog 0.6.0+. (Closes: #1007200) -- Mo Zhou Tue, 22 Mar 2022 15:56:16 -0400 pytorch (1.8.1-4) unstable; urgency=medium * B-D on python3-dev instead of -all-dev. (Closes: #1004331) We only build with one Py3 version due to the buildsystem complexity. * Explicitly fix python and shell shebang in d/rules. * Fix superfluous-file-pattern in d/copyright. * Fix duplicate-short-description in d/control. * Bump Standards-Version to 4.6.0.1 (no change). * Update lintian overrides. -- Mo Zhou Tue, 01 Feb 2022 20:11:00 -0500 pytorch (1.8.1-3) unstable; urgency=medium * Add comments in d/rules on package maintainence. * Mask python spectral_ops test which requires MKL. (Closes: #995360) * Mask python binary_ufuncs autopkgtest. * Only build on modern 64-bit architectures. (Closes: #994423) * Add missing Dep libprotobuf-dev for libtorch-dev. * d/watch: Track releases instead of tags. -- Mo Zhou Sun, 23 Jan 2022 09:14:58 -0500 pytorch (1.8.1-2) unstable; urgency=medium * Upload to unstable. * Let python3-torch Depend on libtorch-test to fix broken link. (Closes: #988778) -- Mo Zhou Mon, 06 Sep 2021 00:22:04 -0400 pytorch (1.8.1-1) experimental; urgency=medium [ Michael R. Crusoe ] * Mark libtorch1.7 as Multi-Arch: same [ Mo Zhou ] * New upstream version 1.8.1 * Upgrade embedded foxi to 6a4e19a2aaf7ae4b9fa9597526e65b395d5e79ad. * Rebase dirtyhack.patch. * Add libnuma-dev to Build-Depends. * SOVERION bump for binary packages. * Refresh shebang.patch and other existing patches. * Refresh autopkgtest control file. -- Mo Zhou Fri, 21 May 2021 14:20:58 +0800 pytorch (1.7.1-7) unstable; urgency=medium * Remove B-D libunwind. * Update autopkgtest. -- Mo Zhou Thu, 11 Feb 2021 15:13:27 +0800 pytorch (1.7.1-6) unstable; urgency=medium * Make B-D:libunwind-dev specific to amd64,arm64,armhf,ppc64el. -- Mo Zhou Mon, 01 Feb 2021 14:35:46 +0800 pytorch (1.7.1-5) unstable; urgency=medium * Add libunwind-dev to avoid weird FTBFS (due to missing header). * Require libonnx-dev (>= 1.7.0+dfsg-3) to avoid FTBFS (Closes: #980636) -- Mo Zhou Wed, 27 Jan 2021 18:38:00 +0800 pytorch (1.7.1-4) unstable; urgency=medium * Allow more autopkgtest cases to fail on armhf. -- Mo Zhou Sat, 02 Jan 2021 17:17:42 +0800 pytorch (1.7.1-3) unstable; urgency=medium * Build against generic BLAS/LAPACK instead of OpenBLAS. -- Mo Zhou Sun, 27 Dec 2020 19:47:18 +0800 pytorch (1.7.1-2) unstable; urgency=medium * Update autopkgtest scripts. -- Mo Zhou Sun, 20 Dec 2020 13:28:22 +0800 pytorch (1.7.1-1) unstable; urgency=medium * New upstream version 1.7.1 * Disable OpenMPI in the distributed functionality of pytorch. + Regenerate Autopkgtest control file after removing OpenMPI support. + Explicitly specify USE_MPI=OFF in d/rules. * Add missing runtime deps to autopkgtest and refresh. - Autopkgtest: allow several tests to fail. (Closes: #974210) + Revamp of the python part of autopkgtest and add the missing B-Ds. * Get rid of the symbols control file. (Closes: #977533) -- Mo Zhou Wed, 16 Dec 2020 20:58:19 +0800 pytorch (1.7.0-2) unstable; urgency=medium * Patch cmake to fix ppc64el and s390x FTBFS, really disabling XNNPACK. * d/rules: Only enable QNNPACK on amd64. * Refresh symbols control file for amd64 to fix amd64 FTBFS. -- Mo Zhou Tue, 24 Nov 2020 20:10:34 +0800 pytorch (1.7.0-1) unstable; urgency=medium * Make B-D:libxnnpack-dev amd64 and arm64 arch-specific. * Make USE_DISTRIBUTED and the corresponding Build-Depends specific to the 64-bit architectures: amd64 arm64 ppc64el mips64el s390x. * Upload to unstable. -- Mo Zhou Tue, 24 Nov 2020 12:59:42 +0800 pytorch (1.7.0-1~exp1) experimental; urgency=medium * New upstream version 1.7.0 * Add missing python deps for the autopkgtest :: python cases. * Mark libideep-dev/libdnnl-dev & USE_MKLDNN as amd64,arm64,ppc64el-specific. * Patch cmake to strip third-party directory includes/checks. * Turn off USE_PYTORCH_QNNPACK on non-amd64 architectures. * Update embedded foxi to snapshot 4aba696ec8f31794fd42880346dc586486205e0a. * Rebase and refresh existing patches. * Rename binary packages following the SOVERSION bump. * Update dirtyhack.patch and add new B-D libnop-dev. * Refresh symbols control file for amd64. * Autopkgtest: refresh C++ unit test list. * Autopkgtest: refresh the list of python unit tests. * Refresh d/copyright for the new upstream release. -- Mo Zhou Sun, 22 Nov 2020 14:30:28 +0800 pytorch (1.6.0-5) unstable; urgency=medium * Revert introduction of unnecessary python3 dependencies. * Mute dh_auto_configure which triggers incorrect mixed python{3.8,3.9} build. * Refresh symbols control file. * Fix autopkgtest regenerator and refresh d/tests/control. * Let python3-torch also Recommend build-essential (for JIT functionality). -- Mo Zhou Wed, 21 Oct 2020 00:16:57 +0800 pytorch (1.6.0-4) unstable; urgency=medium [ Steffen Möller ] * Added d/u/metadata file [ Mo Zhou ] * Address regression bug in dependency data. (Closes: #972540) * Add LaTeX bibliography information to upstream/metadata. -- Mo Zhou Tue, 20 Oct 2020 17:59:38 +0800 pytorch (1.6.0-3) unstable; urgency=medium * Change maintainer mail address into debian-ai@l.d.o * d/rules: Explicitly specify build with default python3. * Refresh symbols control file. * Use those defined-but-unused control variables. * Upload to unstable. -- Mo Zhou Mon, 19 Oct 2020 16:12:46 +0800 pytorch (1.6.0-2) experimental; urgency=medium * Let python3-torch recommend libtorch-dev (= ${binary:Version}). * Let python3-torch recommend ninja-builds (Building PyTorch C++ Extension). * Add debian/tests/run-cpptest.sh for manually running tests. * Refresh test program list for gencontrol-pytest.sh and add run-pytest.sh * Filter out CUDA tests from gencontrol-pytest.sh. * Remove the unnecessary USE_OPENCV=ON override from the patch * Enable DNNL/ideep integration and refresh symbols control file. * Remove "-" signs from rules and add symlink in python3-torch. * Fix the stronger-dependency-implies-weaker problem for python3-torch. * Fix the globbing-patterns-out-of-order problem for d/copyright. * Refresh the shebang patch. * Sort tests in gencontrol-cpptest.sh and regenerate tests/control. * Toggle USE_ZSTD=ON and add the corresponding patches and B-Ds. -- Mo Zhou Mon, 07 Sep 2020 14:27:37 +0800 pytorch (1.6.0-1) experimental; urgency=medium * New upstream version 1.6.0 * Refresh shebang.patch. * Significantly rebase dirtyhack.patch as most changes have been upstreamed. * Remove dont-break-dev-null.patch. fixed upstream. * Add some missing Depends for libtorch-dev. * Stripping the quotes in ONNX_NAMESPACE to fix build failure. * Add libtensorpipe-dev to Build-Depends. * Rename libtorch1 -> libtorch1.6 and refresh symbols control file. * dirtyhack.patch: Deal with the fmt and tensorpipe libraries. * rules: use SOVERSION variable and update the install target. * Update d/copyright and apply wrap-and-sort. * Put proper symlinks to usr/lib/python3*/dist-packages/torch/lib. -- Mo Zhou Mon, 31 Aug 2020 19:41:58 +0800 pytorch (1.5.1-1) experimental; urgency=medium * New upstream version 1.5.1 * Fix a small part of copyright that lintian complains about. * Rewrite autopkgtest: use script to automatically generate control file. * Automatically generate autopkgtest cases for upstream python tests. * Add the ordinal number to the name of autopkgtest cases. * Rules: add option to allow users to rebuild to -march=native. * Patch a python test case lest it break the /dev/null file. * Add build-essential and libtorch-dev to autopkgtest deps. * Refresh symbols control file. -- Mo Zhou Sun, 26 Jul 2020 21:28:04 +0800 pytorch (1.5.0-2) experimental; urgency=medium * Install NOTICE to the binary files * Fix the path globs in d/copyright. * Add lintian overrides. * Track the shlibs symbols of binary package libtorch1. * Patch the usr/bin/env shabang. -- Mo Zhou Wed, 29 Apr 2020 18:14:25 +0800 pytorch (1.5.0-1) experimental; urgency=medium * Initial release. (Closes: #853923) -- Mo Zhou Wed, 29 Apr 2020 13:11:46 +0800