ort
A Rust interface for hardware-accelerated inference and training on ONNX-format machine learning models, built mainly as a wrapper around Microsoft's ONNX Runtime with support for other pure-Rust runtimes.
ort is a Rust library for running and training machine learning models in the Open Neural Network Exchange (ONNX) format, with hardware acceleration across a wide range of accelerators. It is built primarily as a wrapper around Microsoft's ONNX Runtime, successor to the now-inactive onnxruntime-rs crate, and also supports other pure-Rust runtimes.
It targets deploying a model originally trained in PyTorch, TensorFlow, Keras, scikit-learn or PaddlePaddle either on a user's device or in a datacenter, and is light enough to run on end-user hardware. Other open-source projects use it for text-embedding inference, file-type detection, OCR, transformer inference at the edge, text-to-speech, semantic search and local speech-to-text.