AI Potluck
Contribute
Back to Gap Map Infrastructure / Core ML frameworks & libraries

PyTorch

PyTorch Foundation (Linux Foundation)
open source / Overall score: 5.0(leading)

PyTorch is a Python package for tensor computation with GPU acceleration and for deep neural networks built on a tape-based autograd system. It works alongside Python libraries such as NumPy, SciPy and Cython, and provides distributed training across multiple GPUs and machines.

Openness

5 high confidence
5.0
license
BSD-3-Clause(OSI)
source
public
core features withheld
no

PyTorch is released under a BSD-3-Clause license carrying copyright lines from Meta and other contributing institutions, and the full source is public in this unarchived repository. The README describes the whole library with no paid, enterprise or hosted tier beside it, so the core is ungated.

  • https://cdn.jsdelivr.net/gh/pytorch/pytorch@main/LICENSE recorded 2026-09-29

    The LICENSE file in pytorch/pytorch carries the BSD 3-Clause conditions, including "Redistributions of source code must retain the above copyright" and "Neither the names of Facebook, Deepmind Technologies, NYU, NEC Laboratories America".

  • https://github.com/pytorch/pytorch/blob/main/LICENSE recorded 2026-08-13

    LICENSE body carries the three BSD 3-Clause conditions and the 'Neither the names ... may be used to endorse' clause, under per-contributor copyright lines (Facebook, Idiap, Deepmind, NEC, NYU and others)

  • https://pypi.org/pypi/torch/json recorded 2026-09-29

    PyPI page text carries the project README: "PyTorch has a BSD-style license, as found in the [LICENSE](LICENSE) file." and the license expression lists "BSD-3-Clause" among the bundled component licenses.

  • https://pypi.org/pypi/torch/json recorded 2026-09-29

    The README build guide tells readers to fetch the public source tree with "git clone https://github.com/pytorch/pytorch". Searched the body for paid tier, enterprise edition and hosted tier: the only "Enterprise" is Visual Studio Enterprise in the Windows build notes.

Adoption

5 high confidence
5.0

Adoption is measured by PyPI downloads of the torch package, though conda installs and PyTorch's own wheel index are not counted, so the figure undercounts total installs.

Capability

5 high confidence
5.0

Defines the model-training layer alongside TensorFlow.

  • https://docs.pytorch.org/docs/2.14/index.html recorded 2026-09-29

    The 2.14 documentation index lists reference pages for "torch.autograd", "torch.nn", "torch.compiler" and "Distributed communication package - torch.distributed".

  • https://github.com/pytorch/pytorch recorded 2026-08-13

    README still describes tensor computation with strong GPU acceleration and deep neural networks on a tape-based autograd system

  • https://pypi.org/pypi/torch/json recorded 2026-09-29

    PyPI page for torch carries the project README, which says "PyTorch is a Python package that provides two high-level features" including "Tensor computation (like NumPy) with strong GPU acceleration" and "Deep neural networks built on a tape-based autograd system".

  • https://pytorch.org/tutorials/beginner/dist_overview.html recorded 2026-09-29

    Distributed overview tutorial advises "Use DistributedDataParallel (DDP), if your model fits in a single GPU but you want to easily scale up training using multiple GPUs" and links guides for FSDP2, tensor parallel and pipeline parallelism.

Verified 2026-09-24