The PyTorch Foundation announced three new projects joining its open source AI portfolio at PyTorch Conference EU in Paris. Safetensors, originally developed by Hugging Face in 2022, provides a secure tensor serialization format that prevents arbitrary code execution during model sharing. ExecuTorch, initially a Meta internal tool, becomes a PyTorch Core project focused on efficient AI inference on edge devices like mobile phones and AR/VR headsets. Helion, a Python-embedded DSL for authoring ML kernels, aims to simplify kernel development with higher-level abstractions and ahead-of-time autotuning for hardware-portable performance. Together, these additions strengthen the foundation's vendor-neutral infrastructure covering the full AI lifecycle from training to inference.
Table of contents
Safetensors brings secure model distributionFrom ExecuTorch, greater on-demand inference capabilitiesHelion standardizes AI kernel developmentExpanding the open source AI stackSort: