Are we learning yet?
Running a model that was trained somewhere else is the strongest thing Rust does in machine learning, and the community says so itself: training stays in Python, and low-latency inference is where Rust may be the right callSource 1: users.rust-lang.org.
The usual route is an ONNX export loaded with ortort · ONNX & general runtimesNo stable release: the newest version is 2.0.0-rc.13.version2.0.0-rc.13downloads6.8M / 90dstars2.5klicenseMIT OR Apache-2.0last commit2026-09-14published2026-07-28dependents31 owners · 32 repos
Open ort in the catalogLast scraped on Sep 14, 2026, with pure-Rust
engines below it if you would rather not link a C++ runtime.
The exception is the small end: the 2023 generation of browser and microcontroller runtimes has been archived or abandoned, and what works there is a pure-Rust engine compiled to WASM, on the CPU.
- users.rust-lang.org, 2025-07-24 — 62 likes Back to reference 1 in the text