Loading post…
Hugging FaceLuc Georges, LysandreAnnouncement
Safetensors is Joining the PyTorch Foundation
Summary
Safetensors has joined the PyTorch Foundation as a foundation-hosted project under the Linux Foundation, alongside DeepSpeed, Helion, Ray, vLLM, and PyTorch. It began as a Hugging Face project addressing the risk of arbitrary code execution in pickle-based model-weight formats. Its format stores tensor metadata in a JSON header capped at 100MB, followed by raw tensor data, and supports zero-copy mapping and lazy loading. The move places the trademark, repository, and governance with the Linux Foundation, while Luc and Daniel remain on the Technical Steering Committee and lead day-to-day work. Existing users get the same format, APIs, and Hub integration without breaking changes, while contributors gain a documented maintainer path; planned work includes PyTorch core serialization, accelerator-aware loading, parallel loading, and quantization support.
Context
Safetensors was created to provide a way to store and share model weights without the arbitrary-code execution risk associated with pickle-based formats. As open model sharing became central to the ML community and adoption expanded, the project sought governance that represented the broader community rather than a single company.
Approach / What changed
Safetensors is moving its trademark, repository, and governance under the Linux Foundation through the PyTorch Foundation while retaining its existing format, APIs, Hub integration, and day-to-day technical leadership. The project also plans collaboration on PyTorch serialization, device-aware loading, Tensor Parallel and Pipeline Parallel APIs, and support for evolving quantization formats.
Takeaways
The format combines a JSON metadata header limited to 100MB with raw tensor data, enabling zero-copy loading and lazy access to individual weights.
The Linux Foundation now holds Safetensors' trademark, repository, and governance; Luc and Daniel remain on the Technical Steering Committee and continue leading the project day to day.
The transition introduces no breaking changes for users, while contributors receive a formally documented path to maintainer status.
Source and freshness
- Source
- huggingface.co
- Published
- Added to Yomu
- Last updated
Summaries are generated from the original by a language model. Check the original before relying on it.
Keep reading
Related reading
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
The best engineering reads, once a week
One useful email. Unsubscribe anytime.