PyTorch is an open-source framework for training, optimising and running machine learning models, released under the BSD-3-Clause licence. According to the PyTorch Foundation, it is the core of a wider set of open-source AI projects, including vLLM, DeepSpeed, Ray and Safetensors. It is developed by a community of contributors and is used by developers, researchers and enterprises. PyTorch is a library rather than a hosted service, so each organisation chooses where it runs and where its data is processed.
The PyTorch Foundation is hosted by the Linux Foundation, a non-profit organisation that describes itself as a 501(c)(6) non-profit and lists a San Francisco, California address. The foundation is overseen by a governing board under a published charter, and the technical governance of PyTorch is defined by its maintainers. The foundation's public pages describe no hosted service of its own, so the hosting decision rests with the organisation that deploys the framework, whether on EU-based or on-premise systems.