The PyTorch Foundation has taken on two new projects: Helion, a tool for writing machine learning kernels contributed by Meta, and Safetensors, a secure model file format contributed by Hugging Face.
Both were announced at PyTorch Conference Europe in Paris, and the two now join DeepSpeed, Ray, vLLM, and PyTorch itself as foundation-hosted projects.
Moreover, the foundation has confirmed that ExecuTorch, Meta's solution for running PyTorch models on edge and on-device environments, is being merged into PyTorch Core.
If you were looking for the why, it is fairly straightforward.
Both moves come as AI teams increasingly focus on getting models into production rather than just training them. Running kernels efficiently across different hardware and keeping model files safe to load are two problems the ecosystem has been dealing with for a while.
Talking about Helion joining up, Matt White, the CTO of the PyTorch Foundation, added that:
Helion gives engineers a much more productive path to writing high-performance kernels, including autotuning across hundreds of candidate implementations for a single kernel.
As part of the PyTorch Foundation community, this project strengthens the foundation for an open AI stack that is more portable and significantly easier for the community to build on.
Luc Georges, Chief Open Source Officer, Hugging Face echoed similar excitement:
Safetensors joining the PyTorch Foundation is an important step towards using a safe serialization format everywhere by default. The new ecosystem and exposure the library will gain from this move will solidify its security guarantees and usability. Safetensors is a well-established project, adopted by the ecosystem at large, but we're still convinced we're at the very beginning of its lifecycle: the coming months will see significant growth, and we couldn't think of a better home for that next chapter than the PyTorch Foundation.
What does this mean?
The PyTorch Foundation is a Linux Foundation-hosted organization that acts as the vendor-neutral home for PyTorch and a growing set of open source AI projects. The main goal here is to keep governance and technical direction community-driven rather than tied to any single company's whims.
The Linux Foundation is the broader stewardship body behind over 1,000 open source projects, covering everything from the Linux kernel and Kubernetes to OpenSSF. The PyTorch Foundation sits under that umbrella, giving its projects access to LF's governance infrastructure and oversight.
Helion comes in as a tool that makes writing the low-level code that runs AI models on GPUs significantly less painful. It handles a lot of the tedious groundwork automatically, and finds the best configuration for your hardware on its own.
Whereas Safetensors is a file format for storing and sharing AI model weights that doesn't come with the security baggage of older formats.
