rust-bert

Why we included this project

Most Rust teams would rather not stand up a separate Python service just to run a transformer model. rust-bert ports the Hugging Face Transformers stack to native Rust and gives you ready-to-use pipelines for things like question answering, translation, summarization, and named entity recognition, so a few lines of code calling predict() replace hand-wiring model weights, tokenizers, and tensor ops. It runs on tch-rs or ONNX Runtime, with multi-threaded tokenization and GPU inference, which suits latency-sensitive or embedded use where you want the model close to the rest of your code. One thing to know: this is a component library, not a runnable application, so you integrate it into your own service or tooling.