Infrastructure / Core ML frameworks & libraries

Safetensors

Safetensors is a file format and library for safely storing and loading tensors, designed as a secure alternative to Python pickle that avoids arbitrary code execution and supports fast zero-copy loading. It is implemented in Rust with Python bindings and works across PyTorch, TensorFlow, and other frameworks. Maintained by Hugging Face, it has become the default weight format for models on the Hugging Face Hub.

Verified live 2026-06-22 via primary sources. LICENSE body is verbatim Apache License 2.0; full Rust/Python source public.

Openness

5 high confidence

5.0

license: Apache-2.0(OSI)
source: public
core-gated: ungated

LICENSE body is verbatim Apache License 2.0; full Rust/Python source public.

https://github.com/huggingface/safetensors/blob/main/LICENSE recorded 2026-06-22
LICENSE file is the verbatim Apache License Version 2.0

Adoption

5 high confidence

5.0

PyPI last-month downloads of 78,250,137 exceed the 50M+ threshold.

https://pypistats.org/api/packages/safetensors/recent recorded 2026-06-22
last_month = 78,250,137 PyPI downloads

Capability

4 high confidence

4.0

Narrow scope (storage format) but central as the default Hub weight format.

https://github.com/huggingface/safetensors recorded 2026-06-22
README describes safe, zero-copy, multi-framework tensor format

Unchanged since 2026-07-30 (last edited, not re-checked)