September 12, 2022
To accelerate progress in AI, PyTorch is moving to a new, independent PyTorch Foundation, under the Linux Foundation umbrella.
The project will join the Linux Foundation with a diverse governing board composed of representatives from AMD, Amazon Web Services, Google Cloud, Meta, Microsoft Azure, and Nvidia, with the intention to expand over time. The PyTorch Foundation will act as a responsible steward for the technology and support PyTorch through conferences, training courses, and other initiatives.
The foundation’s mission is to drive adoption of AI tooling by fostering and sustaining an ecosystem of open source, vendor-neutral projects with PyTorch. It will democratize state-of-the-art tools, libraries, and other components to make these innovations accessible to everyone.
The PyTorch Foundation will focus on the business and product marketing of PyTorch and the related ecosystem. The transition will not entail any changes to PyTorch’s code and core project, including its separate technical governance structure.
Meta will continue to invest in PyTorch and use it as our primary framework for AI research and production applications at the company.
Since we partnered with the AI community to create the PyTorch framework for AI research six years ago, open collaboration has been essential to its success. With some 2,400 contributors who have built more than 150,000 projects on the framework, PyTorch has become one of the leading platforms for AI research as well as commercial production use. Today, we are excited to announce the next step for PyTorch: The project will transition to a newly launched PyTorch Foundation, which will be part of the nonprofit Linux Foundation, a technology consortium whose core mission is the collaborative development of open source software.
The PyTorch Foundation will boast a wide-ranging governing board composed of representatives from AMD, Amazon Web Services, Google Cloud, Meta, Microsoft Azure, and Nvidia, with the intention to expand further over time. The board will prioritize the continued growth of PyTorch’s vibrant community, the driving force behind the project’s success. Releases, features, and technical direction will still be driven by the maintainers, committers, and contributors to the project. The creation of the PyTorch Foundation ensures that decisions will be made in a transparent and open manner by a diverse group of members for many years to come.
It’s hard to overstate how quickly PyTorch has grown. In 2016, a group of Meta AI researchers embarked quietly on a couple of projects. They were determined to create a single, simple, standardized interface for their end-to-end workflow. They also harbored hopes of fixing the tedious, complicated research-to-production pipeline of the AI field. In fact, this pipeline was more of a labyrinth, involving multiple steps and tools, fragmented processes, and navigation between different frameworks that were optimized for either research or production, but not both. The Meta team experimented with machine learning (ML) frameworks such as Theano and Torch, and with advanced concepts from Lua Torch, Chainer, and HIPS Autograd, but stayed focused on one thing above all for their new framework: usability.
Just two years later, Meta announced PyTorch 1.0, a dynamic, interactive framework that allowed developers to not only experiment rapidly but also seamlessly transition to graph-based modes for deployment.
Their efforts have paid off. PyTorch has since grown into the lingua franca of AI research. Today, more than 80 percent of researchers who submit their work at major ML conferences, such as NeurIPS or ICML, harness the framework. We have built libraries that support some of the principal domains of the AI field, such as torchvision, which powers most of the world’s modern computer vision research. The framework will continue to be a part of Meta’s AI research and engineering work. PyTorch is also a foundation of the AI research and products built by Amazon Web Services, Microsoft Azure, OpenAI, and many other companies and research institutions.
Meta has worked steadily to nurture the community-driven growth that has fueled PyTorch’s success. We’ve committed hundreds of engineers to the framework and strongly supported product development and community outreach.
But the time is right for a new home for PyTorch.
PyTorch was built from the ground up with an open source, community-first philosophy, and that will not change. When researchers and developers open-source their code, others around the world can share their work, learn from one another’s advances, and then contribute back to the community.
We have always had a clear technical governance structure in place for PyTorch, with core maintainers setting the vision for the framework and the best AI researchers and engineers across industry building its modules. That will be even more important for the future of the framework, as Meta and other contributors have a three-year vision to expand its functionality, modularity, and diversity of code ownership.
Going forward, the framework’s contributors will benefit from the robust governance, diverse leadership, and additional investments provided by the new PyTorch Foundation partners. The PyTorch Foundation will strive to adhere to four principles: remaining open, maintaining neutral branding, staying fair, and forging a strong technical identity. One of the foundation’s main priorities is to maintain a clear separation between the business and technical governance of PyTorch.
Meta stays fully committed to PyTorch. We will continue to invest in the framework, and use it as the primary framework for our AI research and production. The transition will not entail any changes to PyTorch’s code and core project, and developer operating models, including contributor guidelines and licensing, will also remain unchanged. All releases, features, and technical direction will continue to be driven by PyTorch’s community: from individual code contributors, those who review and commit changes, to the module maintainers. We’re lucky to have strong support from leadership at Meta, which is fully behind the move to take PyTorch to its next logical phase.
Meta has put open science at the core of our work in AI, whether it’s releasing code for large language models, self-supervised computer vision systems, innovative new datasets, embodied AI platforms, and much more. We believe this approach enables the fastest progress in building and deploying new systems that will address real-world needs and answer fundamental questions about the nature of intelligence. PyTorch has been a core part of this work, and now, with the creation of the PyTorch Foundation, the entire AI community is positioned to push the field forward in countless exciting new ways.