Local execution of AI on edge devices is important for low latency and offline operation. However, deploying models on diverse hardware remains fragmented, often requiring model conversion or complete reimplementation outside the PyTorch ecosystem where the model was originally authored. We introduce ExecuTorch, a unified PyTorch-native deployment framework for edge AI. ExecuTorch enables seamless deployment of machine learning models across heterogeneous compute environments. It scales from embedded microcontrollers to complex system-on-chips (SoCs) with dedicated accelerators, powering devices ranging from wearables and smartphones to large compute clusters. ExecuTorch preserves PyTorch semantics while allowing customization, support for optimizations like quantization, and pluggable execution "backends". These features together enable fast experimentation, allowing researchers to validate deployment behavior entirely within PyTorch, bridging the gap between research and production.
@article{arxiv.2605.08195,
title = {ExecuTorch -- A Unified PyTorch Solution to Run AI Models On-Device},
author = {Mergen Nachin and Digant Desai and Sicheng Stephen Jia and Chen Lai and Mengwei Liu and Jacob Szwejbka and Raziel Alvarez and RJ Ascani and Dave Bort and Manuel Candales and Andrew Caples and Yanan Cao and Zhengxu Chen and Soumith Chintala and Gregory Comer and Tanvir Islam and Songhao Jia and Tarun Karuturi and Jack Khuu and Abhinay Kukkadapu and Tugsbayasgalan Manlaibaatar and Andrew Or and Kimish Patel and Siddartha Pothapragada and Lucy Qiu and Supriya Rao and Orion Reblitz-Richardson and Max Ren and Scott Roy and Anthony Shoumikhin and Scott Wolchok and Guang Yang and Angela Yi and Martin Yuan and Hansong Zhang and Jack Zhang and Jerry Zhang and Shunting Zhang and C. Cagatay Bilgin},
journal= {arXiv preprint arXiv:2605.08195},
year = {2026}
}