ExecuTorch-rs

ExecuTorch-rs provides a high-level Rust API over ExecuTorch's C++ library for running PyTorch models on mobile, embedded and edge devices. A model is created and exported in Python as usual, then loaded and executed from Rust using this crate.

The crate requires the C++ ExecuTorch library to be compiled separately, since the needed modules, backends and operations depend on the target configuration, and each release of the Rust crate is tied to a specific ExecuTorch C++ version. Optional Cargo features add a data-loader module for additional input formats, a higher-level module API for loading and running models, and a tensor pointer type that manages a tensor's buffer and metadata together.