pytorch-cpu v2.9.1

[regro-cf-autotick-bot]

It is very likely that the current package version for this feedstock is out of date.

Checklist before merging this PR:

• Dependencies have been updated if changed: see upstream
• Tests have passed
• Updated license if changed and license_file is packaged

Information about this PR:

1. Feel free to push to the bot's branch to update this PR if needed.
2. The bot will almost always only open one PR per version.
3. The bot will stop issuing PRs if more than 3 version bump PRs generated by the bot are open. If you don't want to package a particular version please close the PR.
4. If you want these PRs to be merged automatically, make an issue with @conda-forge-admin,please add bot automerge in the title and merge the resulting PR. This command will add our bot automerge feature to your feedstock.
5. If this PR was opened in error or needs to be updated please add the bot-rerun label to this PR. The bot will close this PR and schedule another one. If you do not have permissions to add this label, you can use the phrase @conda-forge-admin, please rerun bot in a PR comment to have the conda-forge-admin add it for you.

Closes: #415
Closes: #433
Closes: #446

Pending Dependency Version Updates

Here is a list of all the pending dependency version updates for this repo. Please double check all dependencies before merging.

Name	Upstream Version	Current Version
cudnn	9.15.1.9
libprotobuf	33.1
protobuf	33.1
pytest	9.0.1

_{This PR was created by the regro-cf-autotick-bot. The regro-cf-autotick-bot is a service to automatically track the dependency graph, migrate packages, and propose package version updates for conda-forge. Feel free to drop us a line if there are any issues! This PR was generated by https://github.com/regro/cf-scripts/actions/runs/19320428618 - please use this URL for debugging.}

[conda-forge-admin]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe/meta.yaml) and found it was in an excellent condition.

I do have some suggestions for making it better though...

For recipe/meta.yaml:

• ℹ️ The magma output has been superseded by libmagma-devel.
• ℹ️ The recipe is not parsable by parser conda-souschef (grayskull). This parser is not currently used by conda-forge, but may be in the future. We are collecting information to see which recipes are compatible with grayskull.
• ℹ️ The recipe is not parsable by parser conda-recipe-manager. The recipe can only be automatically migrated to the new v1 format if it is parseable by conda-recipe-manager.

_{This message was generated by GitHub Actions workflow run https://github.com/conda-forge/conda-forge-webservices/actions/runs/19647270882. Examine the logs at this URL for more detail.}

[h-vetinari]

Good news; #446 has unblocked us on osx 🥳 🚀
Thanks to everyone who made this possible, with the biggest shout-out to @aktech of course! 🙏

I've integrated those changes here now. 93301f0 should obviously move to the ci-setup, but that can happen independently. I want to see if everything works when merging the osx-on-cirun builds; the uploads obviously are not tested in a PR.

[h-vetinari]

Sigh, don't know what's up with the linux-aarch builds here

-- Check for working C compiler: $PREFIX/bin/aarch64-conda-linux-gnu-cc
-- Check for working C compiler: $PREFIX/bin/aarch64-conda-linux-gnu-cc - broken
CMake Error at $PREFIX/share/cmake-4.2/Modules/CMakeTestCCompiler.cmake:67 (message):
  The C compiler

    "$PREFIX/bin/aarch64-conda-linux-gnu-cc"

  is not able to compile a simple test program.

  It fails with the following output:

    Change Dir: '$SRC_DIR/cmake_test/CMakeFiles/CMakeScratch/TryCompile-g1bhHD'
    
    Run Build Command(s): $PREFIX/bin/ninja -v cmTC_21ec8
    [1/2] $PREFIX/bin/aarch64-conda-linux-gnu-cc   -ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O3 -pipe -isystem $PREFIX/include -fdebug-prefix-map=$SRC_DIR=/usr/local/src/conda/libtorch-2.9.1 -fdebug-prefix-map=$PREFIX=/usr/local/src/conda-prefix  -I$PREFIX/targets/sbsa-linux/include -I$BUILD_PREFIX/targets/sbsa-linux/include -o CMakeFiles/cmTC_21ec8.dir/testCCompiler.c.o -c $SRC_DIR/cmake_test/CMakeFiles/CMakeScratch/TryCompile-g1bhHD/testCCompiler.c
    [2/2] : && $PREFIX/bin/aarch64-conda-linux-gnu-cc -ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O3 -pipe -isystem $PREFIX/include -fdebug-prefix-map=$SRC_DIR=/usr/local/src/conda/libtorch-2.9.1 -fdebug-prefix-map=$PREFIX=/usr/local/src/conda-prefix  -I$PREFIX/targets/sbsa-linux/include -I$BUILD_PREFIX/targets/sbsa-linux/include -Wl,-O2 -Wl,--sort-common -Wl,--as-needed -Wl,-z,relro -Wl,-z,now -Wl,--allow-shlib-undefined -Wl,-rpath,$PREFIX/lib -Wl,-rpath-link,$PREFIX/lib -L$PREFIX/lib  -L$PREFIX/targets/sbsa-linux/lib -L$PREFIX/targets/sbsa-linux/lib/stubs -L$BUILD_PREFIX/targets/sbsa-linux/lib -L$BUILD_PREFIX/targets/sbsa-linux/lib/stubs CMakeFiles/cmTC_21ec8.dir/testCCompiler.c.o -o cmTC_21ec8   && :
    FAILED: [code=1] cmTC_21ec8 
    : && $PREFIX/bin/aarch64-conda-linux-gnu-cc -ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O3 -pipe -isystem $PREFIX/include -fdebug-prefix-map=$SRC_DIR=/usr/local/src/conda/libtorch-2.9.1 -fdebug-prefix-map=$PREFIX=/usr/local/src/conda-prefix  -I$PREFIX/targets/sbsa-linux/include -I$BUILD_PREFIX/targets/sbsa-linux/include -Wl,-O2 -Wl,--sort-common -Wl,--as-needed -Wl,-z,relro -Wl,-z,now -Wl,--allow-shlib-undefined -Wl,-rpath,$PREFIX/lib -Wl,-rpath-link,$PREFIX/lib -L$PREFIX/lib  -L$PREFIX/targets/sbsa-linux/lib -L$PREFIX/targets/sbsa-linux/lib/stubs -L$BUILD_PREFIX/targets/sbsa-linux/lib -L$BUILD_PREFIX/targets/sbsa-linux/lib/stubs CMakeFiles/cmTC_21ec8.dir/testCCompiler.c.o -o cmTC_21ec8   && :
    $PREFIX/bin/../lib/gcc/aarch64-conda-linux-gnu/13.4.0/../../../../aarch64-conda-linux-gnu/bin/ld: $PREFIX/bin/../libexec/gcc/aarch64-conda-linux-gnu/13.4.0/liblto_plugin.so: error loading plugin: $PREFIX/bin/../libexec/gcc/aarch64-conda-linux-gnu/13.4.0/liblto_plugin.so: cannot open shared object file: No such file or directory
    collect2: error: ld returned 1 exit status
    ninja: build stopped: subcommand failed.

That library is definitely present at libexec/gcc/aarch64-conda-linux-gnu/13.4.0/liblto_plugin.so in the prefix, based on the deps we're using in the test environment

    gcc_impl_linux-aarch64:             13.4.0-h69010b7_7                  conda-forge
    gcc_linux-aarch64:                  13.4.0-hbf9eca1_14                 conda-forge

CC @conda-forge/ctng-compilers, in case this could be related to recent work on that feedstock.

[mgorny]

Maybe it's just some flake? It would really be weird if it only noticed the missing plugin this late in the build.

[h-vetinari]

Unfortunately it reproduces in the linux-aarch CPU job. Clearly the actual cross-compilation part works during the build. We get the failure during the test-phase when we're emulating linux-aarch64, and there's only a single environment.

As such, flags like -L$BUILD_PREFIX/targets/sbsa-linux/lib/stubs make no sense in the test environment, because there's no $BUILD_PREFIX anymore. Do you know the precedence of repeated -L arguments off the top of your head? It happens that the the above is the last -L for the compiler invocation. Beyond that grain of a suspicion, I have nothing currently

[h-vetinari]

The last passing build two weeks ago had

    gcc_impl_linux-aarch64:             13.4.0-h69010b7_7                  conda-forge
    gcc_linux-aarch64:                  13.4.0-hbf9eca1_13                 conda-forge
    gxx_impl_linux-aarch64:             13.4.0-hf6d83cf_7                  conda-forge
    gxx_linux-aarch64:                  13.4.0-h2dde472_13                 conda-forge
    [...]
    sysroot_linux-aarch64:              2.34-h96c1060_2                    conda-forge

That's the only difference between then and now from a compiler POV is conda-forge/ctng-compiler-activation-feedstock#168 AFAICT, but that PR only affects the bootstrap compilers, which we're not using here (also the newest run still has a 2.34 sysroot). There's also conda-forge/binutils-feedstock#101, but that looks benign. In total, the environment differences are as follows:

diff of test environment from before (passing) to after (failing)

@@ -3,12 +3,12 @@ The following NEW packages will be INSTALLED:
     _openmp_mutex:                      4.5-6_kmp_llvm                     conda-forge
     arm-variant:                        1.2.0-sbsa                         conda-forge
     attr:                               2.5.1-h4e544f5_1                   conda-forge
-    binutils_impl_linux-aarch64:        2.45-ha36da51_0                    conda-forge
-    binutils_linux-aarch64:             2.45-hf1166c9_0                    conda-forge
+    binutils_impl_linux-aarch64:        2.45-default_1234567_3             conda-forge
+    binutils_linux-aarch64:             2.45-default_1234567_3             conda-forge
     bzip2:                              1.0.8-h4777abc_8                   conda-forge
     c-ares:                             1.34.5-h86ecc28_0                  conda-forge
     ca-certificates:                    2025.11.12-hbd8a1cb_0              conda-forge
-    cmake:                              4.1.2-hc9d863e_0                   conda-forge
+    cmake:                              4.2.0-hc9d863e_0                   conda-forge
     cuda-cccl_linux-aarch64:            12.9.27-h579c4fd_0                 conda-forge
     cuda-crt-dev_linux-aarch64:         12.9.86-h579c4fd_2                 conda-forge
     cuda-crt-tools:                     12.9.86-h579c4fd_2                 conda-forge
@@ -34,17 +34,17 @@ The following NEW packages will be INSTALLED:
     cudnn:                              9.10.2.21-h32c1c63_0               conda-forge
     fmt:                                12.0.0-h416241a_0                  conda-forge
     gcc_impl_linux-aarch64:             13.4.0-h69010b7_7                  conda-forge
-    gcc_linux-aarch64:                  13.4.0-hbf9eca1_13                 conda-forge
+    gcc_linux-aarch64:                  13.4.0-hbf9eca1_14                 conda-forge
     gxx_impl_linux-aarch64:             13.4.0-hf6d83cf_7                  conda-forge
-    gxx_linux-aarch64:                  13.4.0-h2dde472_13                 conda-forge
+    gxx_linux-aarch64:                  13.4.0-he64a8ad_14                 conda-forge
     kernel-headers_linux-aarch64:       5.14.0-h05a177a_2                  conda-forge
     keyutils:                           1.6.3-h86ecc28_0                   conda-forge
     krb5:                               1.21.3-h50a48e9_0                  conda-forge
-    ld_impl_linux-aarch64:              2.45-hd32f0e1_0                    conda-forge
+    ld_impl_linux-aarch64:              2.45-default_1234567_3             conda-forge
     libabseil:                          20250512.1-cxx17_h201e9ed_0        conda-forge
-    libblas:                            3.9.0-38_haddc8a3_openblas         conda-forge
+    libblas:                            3.11.0-2_haddc8a3_openblas         conda-forge
     libcap:                             2.77-h68e9139_0                    conda-forge
-    libcblas:                           3.9.0-38_hd72aa62_openblas         conda-forge
+    libcblas:                           3.11.0-2_hd72aa62_openblas         conda-forge
     libcublas:                          12.9.1.4-he38c790_1                conda-forge
     libcudnn:                           9.10.2.21-hd88968f_0               conda-forge
     libcudnn-dev:                       9.10.2.21-h76cf850_0               conda-forge
@@ -57,17 +57,17 @@ The following NEW packages will be INSTALLED:
     libcusparse:                        12.5.10.65-h8f3c8d4_2              conda-forge
     libedit:                            3.1.20250104-pl5321h976ea20_0      conda-forge
     libev:                              4.33-h31becfc_2                    conda-forge
-    libexpat:                           2.7.1-hfae3067_0                   conda-forge
+    libexpat:                           2.7.3-hfae3067_0                   conda-forge
     libffi:                             3.5.2-hd65408f_0                   conda-forge
     libgcc:                             15.2.0-he277a41_7                  conda-forge
     libgcc-devel_linux-aarch64:         13.4.0-hd10b1b9_107                conda-forge
     libgcc-ng:                          15.2.0-he9431aa_7                  conda-forge
     libgfortran:                        15.2.0-he9431aa_7                  conda-forge
     libgfortran5:                       15.2.0-h87db57e_7                  conda-forge
-    libglib:                            2.86.1-he84ff74_2                  conda-forge
+    libglib:                            2.86.2-hf53f6bf_1                  conda-forge
     libgomp:                            15.2.0-he277a41_7                  conda-forge
     libiconv:                           1.18-h90929bb_2                    conda-forge
-    liblapack:                          3.9.0-38_h88aeb00_openblas         conda-forge
+    liblapack:                          3.11.0-2_h88aeb00_openblas         conda-forge
     liblzma:                            5.8.1-h86ecc28_2                   conda-forge
     libmagma:                           2.9.0-hef847a9_3                   conda-forge
     libnghttp2:                         1.67.0-ha888d0e_0                  conda-forge
@@ -83,17 +83,17 @@ The following NEW packages will be INSTALLED:
     libstdcxx-devel_linux-aarch64:      13.4.0-hd10b1b9_107                conda-forge
     libstdcxx-ng:                       15.2.0-hf1166c9_7                  conda-forge
     libsystemd0:                        258.2-h8c5a66d_1                   conda-forge
-    libtorch:                           2.9.1-cuda129_generic_h5314ce0_200 local      
+    libtorch:                           2.9.1-cuda129_generic_hd20f0ec_200 local      
     libudev1:                           258.2-h8c5a66d_1                   conda-forge
     libuv:                              1.51.0-he30d5cf_1                  conda-forge
     libzlib:                            1.3.1-h86ecc28_2                   conda-forge
-    llvm-openmp:                        21.1.5-he40846f_2                  conda-forge
+    llvm-openmp:                        21.1.6-he40846f_0                  conda-forge
     nccl:                               2.28.9.1-heee7246_0                conda-forge
     ncurses:                            6.5-ha32ae93_3                     conda-forge
-    ninja:                              1.13.1-hdc560ac_0                  conda-forge
+    ninja:                              1.13.2-hdc560ac_0                  conda-forge
     nvtx-c:                             3.3.0-h7ac5ae9_0                   conda-forge
     openssl:                            3.6.0-h8e36d6e_0                   conda-forge
-    pcre2:                              10.46-h15761aa_0                   conda-forge
+    pcre2:                              10.47-hf841c20_0                   conda-forge
     pkg-config:                         0.29.2-hce167ba_1009               conda-forge
     pybind11-abi:                       11-hc364b38_1                      conda-forge
     rdma-core:                          60.0-he839754_0                    conda-forge

Not sure though with the CMake logs if the sanity checks were actually skipped or the (re)computation isn't shown in the logs; the passing run had

-- The C compiler identification is GNU 13.4.0
-- The CXX compiler identification is GNU 13.4.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: $PREFIX/bin/aarch64-conda-linux-gnu-cc - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: $PREFIX/bin/aarch64-conda-linux-gnu-c++ - skipped

while now we have

-- The C compiler identification is GNU 13.4.0
-- The CXX compiler identification is GNU 13.4.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - failed
-- Check for working C compiler: $PREFIX/bin/aarch64-conda-linux-gnu-cc
-- Check for working C compiler: $PREFIX/bin/aarch64-conda-linux-gnu-cc - broken

[traversaro]

This is only based on my intuition, but I would not exclude the possibility of a cmake 4.1.2 --> 4.2.0 regression.

[isuruf]

Same issue as conda-forge/numpy-feedstock#241. Will be fixed by conda-forge/binutils-feedstock#105

[h-vetinari]

Same issue as conda-forge/numpy-feedstock#241. Will be fixed by conda-forge/binutils-feedstock#105

Thank you very much! Could you please explain a little what was the issue? Ostensibly we had the right binutils builds (latest build number, non-bootstrap, right platform), so while I noticed conda-forge/binutils-feedstock#104, that looked like a cosmetic issue to me.

Also the numpy issue you're referencing is really old, and not immediately obviously related; the only hint is the title of your PR that fixed things at the time: "make sure packages have unique build strings for cross compiling".

Which packages here had coinciding build strings, and how did that break the compiler detection only during the test phase? It's an impressive deduction to me - I have no idea what's going on. Would appreciate it if you could enlighten us.

[isuruf]

conda extracts linux-64/binutils-impl_linux-aarch64-2.45.0-default_1234567_3 and linux-aarch64/binutils-impl_linux-aarch64-2.45.0-default_1234567_3 to the same place pkgs/binutils-impl_linux-aarch64-2.45.0-default_1234567_3 and conflicts with each other.

[h-vetinari]

Ah, right, so it's a function of the package cache, and therefore not visible in the test environment (where we're not using the linux-64/ variant at all). The obvious question that arises is why conda{,-build} does not distinguish the package cache between subfolders?!

[h-vetinari]

Sigh; I had merged after the issue had been resolved for the CPU build, but it appears there's yet another issue for the aarch + CUDA builds:

$PREFIX/bin/cmake: cannot execute binary file: Exec format error

This shouldn't really happen in emulation... 🤔

I see some clobber errors during environment creation, but not sure if they're related.

[h-vetinari]

For win+CUDA:

$ gh run download 19654421211 --repo conda-forge/pytorch-cpu-feedstock --name conda_artifacts_19654421211_win_64_channel_targetsconda-forge_maincu_hca575dce
$ unzip pytorch-cpu-feedstock_conda_artifacts_.zip
$ cd bld/win-64
$ rm current_repodata.json index.html repodata*
$ ls
libtorch-2.9.1-cuda128_mkl_hbf96477_300.conda       pytorch-gpu-2.9.1-cuda128_mkl_hc88b545_300.conda
pytorch-2.9.1-cuda128_mkl_py310_hcc6672c_300.conda  pytorch-tests-2.9.1-cuda128_mkl_py310_h6b46e55_300.conda
pytorch-2.9.1-cuda128_mkl_py311_h7c65ee9_300.conda  pytorch-tests-2.9.1-cuda128_mkl_py311_ha41f3a0_300.conda
pytorch-2.9.1-cuda128_mkl_py312_h003053b_300.conda  pytorch-tests-2.9.1-cuda128_mkl_py312_hd7dd76b_300.conda
pytorch-2.9.1-cuda128_mkl_py313_h7f80487_300.conda  pytorch-tests-2.9.1-cuda128_mkl_py313_h976a8e3_300.conda
$ ls | xargs anaconda upload
$ DELEGATE=h-vetinari
PACKAGE_VERSION=2.9.1
for package in libtorch pytorch pytorch-gpu pytorch-tests; do
  anaconda copy --from-label main --to-label main --to-owner conda-forge ${DELEGATE}/${package}/${PACKAGE_VERSION}
done

[h-vetinari]

Interestingly, the win+CUDA build passed upon restarting, but the upload section is completely empty

2025-11-26T06:09:17.8469388Z ##[group]Uploading packages
2025-11-26T06:09:18.1238157Z D:\Miniforge\Lib\site-packages\binstar_client\__init__.py:16: UserWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html. The pkg_resources package is slated for removal as early as 2025-11-30. Refrain from using this package or pin to Setuptools<81.
2025-11-26T06:09:18.1239898Z   from pkg_resources import parse_version as pv
2025-11-26T06:09:19.2869603Z Found git SHA db89c38872c8b396a17910897dbd4d564889bf0c for this build!
2025-11-26T06:09:19.2870179Z Using STAGING_BINSTAR_TOKEN for anaconda.org uploads to cf-staging.
2025-11-26T06:09:19.3555402Z WARNING: No numpy version specified in conda_build_config.yaml.  Falling back to default numpy value of 1.26
2025-11-26T06:09:19.3613760Z Adding in variants from internal_defaults
2025-11-26T06:09:19.3614466Z Adding in variants from C:\Users\runnerx\_work\pytorch-cpu-feedstock\pytorch-cpu-feedstock\recipe\conda_build_config.yaml
2025-11-26T06:09:19.3615669Z Adding in variants from C:\Users\runnerx\_work\pytorch-cpu-feedstock\pytorch-cpu-feedstock\.ci_support\win_64_channel_targetsconda-forge_maincuda_compiler_version12.8is_rcFalse.yaml
2025-11-26T06:09:25.8873525Z ##[endgroup]

I can only infer that it didn't even attempt to upload stuff because it found artefacts with exactly matching names already on the CDN, but I still would have expected some logs. However, I checked this on another feedstock where pushing an LTS branch means CI for some past commit on main gets rerun (and thus the packages get reproduced exactly, if not toooooo much time has passed), and there as well it looks exactly like that

2025-10-29T00:15:20.8970834Z ##[group]Uploading packages
2025-10-29T00:15:21.2151644Z C:\Miniforge\Lib\site-packages\binstar_client\__init__.py:16: UserWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html. The pkg_resources package is slated for removal as early as 2025-11-30. Refrain from using this package or pin to Setuptools<81.
2025-10-29T00:15:21.2153127Z   from pkg_resources import parse_version as pv
2025-10-29T00:15:22.2859250Z Found git SHA 7f0f076256913eb3c332f9437c39fa5071e8a928 for this build!
2025-10-29T00:15:22.2860415Z Using STAGING_BINSTAR_TOKEN for anaconda.org uploads to cf-staging.
2025-10-29T00:15:22.3652976Z WARNING: No numpy version specified in conda_build_config.yaml.  Falling back to default numpy value of 1.26
2025-10-29T00:15:22.3712769Z Adding in variants from internal_defaults
2025-10-29T00:15:22.3713539Z Adding in variants from D:\a\1\s\recipe\conda_build_config.yaml
2025-10-29T00:15:22.3714320Z Adding in variants from D:\a\1\s\.ci_support\win_64_cuda_compiler_version11.8.yaml
2025-10-29T00:15:30.0493236Z ##[endgroup]

[h-vetinari]

Thankfully, the aarch+CUDA build passed on main. That means we'll be able to start the migration without much further delay! 🥳