2024 has been a year of incredible growth for PyTorch. As that continues in 2025, the PyTorch Foundation has made important steps towards evolving the governance of the project under the Linux Foundation’s vendor-neutral umbrella.
An important piece of governance for PyTorch is represented by the Technical Advisory Council (TAC). The TAC acts as a bridge between the industry, including but not limited to the PyTorch Foundation members, the community, and the PyTorch core development team.
Operating with transparency and inclusivity, the TAC gathers input, facilitates collaboration, and drives initiatives that enhance the experience for everyone who relies on PyTorch.
In 2025, the TAC will focus on four key areas:
- Build Open, Multi-Cloud Continuous Integration (CI): Building on the groundwork from 2024, the TAC will oversee the transition to an open, community-driven CI infrastructure. In addition to ensuring the extremely high bar for correctness that PyTorch has, PyTorch’s CI is complex with a high-quality bar including many automated functional and performance daily test runs. In 2025, PyTorch’s CI infrastructure will be fully open sourced and extended to support multiple compute providers, enabling broader contribution and participation to the effort from organizations benefitting from PyTorch.
- Support more Accelerators: The TAC is committed to creating a level playing field for the growing landscape of AI accelerators. By gathering industry players and PyTorch developers, the TAC will facilitate efforts towards third-party device support and provide levels of integration of external CI systems with the main PyTorch CI. This will make it easier for emerging hardware to gain adoption within the PyTorch ecosystem, and for users to experiment with diverse compute options for training and inference.
- Create a High-Quality, User-Centric Ecosystem: A big focus for the TAC in early 2025 is on improving the experience and discoverability of the PyTorch ecosystem. With many projects growing organically, users often face challenges navigating projects of different scope and quality within the rapidly changing AI landscape. To solve this, a newly curated ecosystem landscape tool will be launched soon on the PyTorch website. We will also introduce lightweight, open processes to improve projects and ensure users a predictable, high-quality experience. In many ways, the experience with PyTorch is as good as its ecosystem.
- Gather Feedback from Industry and the Community: PyTorch has widespread adoption across research labs, startups, and enterprises. Striking the right balance between expressiveness and performance across the board is a very challenging task, so the TAC set out to be one of the several ways the Core development team receives signals. During our monthly TAC meetings, we provide the opportunity to PyTorch Foundation members from industry and academia, as well as non-member organizations to present their use case, their challenges and discuss them directly with appropriate members of the Core team. This feedback loop helps prioritize improvements, ensuring the framework stays relevant in a fast-evolving AI landscape.
By focusing on these priorities, the TAC aims to maintain PyTorch’s position as the leading deep learning framework, while ensuring it remains open, accessible, and responsive to the needs of its diverse community.
As members of the TAC, we’re extremely excited to contribute to the success of PyTorch and to the impact it’s having in the real world. If you are a PyTorch user or developer, consider participating in our monthly calls (they are open to everyone, and the recordings are available here). Also, if you develop or maintain a project based on PyTorch, consider contributing it to the new PyTorch ecosystem (instructions).