wheelgap tracker

torch wheel coverage

pip install torch on Python 3.14: wheels for the whole tree on 3 of the 6 platforms checked; can't install on Alpine ARM64, macOS Intel and Windows ARM64.

From the wheelgap 0.9.0 scan of 2026-11-16.

torch 2.15.0

#397 by downloads · PyPI

Campaigns: On the to-do list of musllinux aarch64 wheels for the top 500 · On the to-do list of Windows on ARM wheels for the top 500.

Platform3.133.143.14t3.153.15t
Linux x86_64wheelwheelwheelwheelwheel
Linux ARM64wheelwheelwheelwheelwheel
Alpine x86_64can't installcan't installcan't installcan't installcan't install
Alpine ARM64can't installcan't installcan't installcan't installcan't install
macOS Apple siliconwheelwheelwheelwheelwheel
macOS Intelcan't installcan't installcan't installcan't installcan't install
Windows x86_64wheelwheelwheelwheelwheel
Windows ARM64can't installcan't installcan't installcan't installcan't install
Pyodide (browser)can't installcan't installn/an/an/a

Why? uvx wheelgap torch==2.15.0 -p 3.15 -t all --explain

Install coverage with dependencies

Wheelgap checks the scanned release and its dependencies without extras. A target is marked “Not checked” if wheelgap could not resolve the full dependency tree or verify its packages.

TargetInstall result
3.14 linux-x86_64Wheels for the whole tree
3.14 linux-aarch64Wheels for the whole tree
3.14 musl-aarch64Missing distribution
  • cuda-bindings==13.5.0 has no distribution via torch → cuda-bindings
  • nvidia-cublas==13.4.1.3 has no distribution via torch → cuda-toolkit → nvidia-cublas
  • nvidia-cuda-cupti==13.2.86 has no distribution via torch → cuda-toolkit → nvidia-cuda-cupti
  • nvidia-cuda-nvrtc==13.2.86 has no distribution via torch → cuda-toolkit → nvidia-cuda-nvrtc
  • nvidia-cuda-runtime==13.2.86 has no distribution via torch → cuda-toolkit → nvidia-cuda-runtime
  • nvidia-cudnn-cu13==9.26.0.51 has no distribution via torch → nvidia-cudnn-cu13
  • nvidia-cufft==12.2.0.57 has no distribution via torch → cuda-toolkit → nvidia-cufft
  • nvidia-cufile==1.17.1.22 has no distribution via torch → cuda-toolkit → nvidia-cufile
  • nvidia-cusolver==12.2.0.11 has no distribution via torch → cuda-toolkit → nvidia-cusolver
  • nvidia-cusparse==12.7.10.12 has no distribution via torch → cuda-toolkit → nvidia-cusparse
  • nvidia-cusparselt-cu13==0.8.1 has no distribution via torch → nvidia-cusparselt-cu13
  • nvidia-nccl-cu13==2.30.7 has no distribution via torch → nvidia-nccl-cu13
  • nvidia-nvjitlink==13.5.37 has no distribution via torch → cuda-toolkit → nvidia-nvjitlink
  • nvidia-nvshmem-cu13==3.7.2 has no distribution via torch → nvidia-nvshmem-cu13
  • nvidia-nvtx==13.2.86 has no distribution via torch → cuda-toolkit → nvidia-nvtx
  • torch==2.15.0 has no distribution
  • triton==3.9.0 has no distribution via torch → triton
3.14 macos-arm64Wheels for the whole tree
3.14 macos-x86_64Missing distribution
  • torch==2.15.0 has no distribution
3.14 win-arm64Missing distribution
  • torch==2.15.0 has no distribution
3.14t linux-x86_64Wheels for the whole tree
3.14t win-arm64Missing distribution
  • torch==2.15.0 has no distribution

Notes

PyTorch stopped publishing x86_64 macOS wheels after 2.2.2, which has none for Python 3.13 or later (pytorch/pytorch#114602). On an Intel Mac, conda-forge still builds CPU-only PyTorch for osx-64: conda install -c conda-forge pytorch.

Other source · macOS Intel · checked 2026-11-12

github.com/pytorch/pytorch/issues/114602 · anaconda.org/conda-forge/pytorch/files · github.com/conda-forge/pytorch-cpu-feedstock

Know a workaround, an alternative package, a conda-forge build or an upstream issue for torch? Add a note (a pull request on GitHub; the format is in notes/README.md).

Badge for your README

[![wheels](https://img.shields.io/endpoint?url=https://wheelgap.github.io/tracker/badge/torch.json)](https://wheelgap.github.io/tracker/#package=torch)
[![wheels with dependencies](https://img.shields.io/endpoint?url=https://wheelgap.github.io/tracker/badge/torch-tree.json)](https://wheelgap.github.io/tracker/p/torch/)

Package JSON

All 5,000 packages