ovos-plugin-linguonnx

OVOS language plugins built on linguonnx. One package, two plugins, both CPU-only on onnxruntime, both fully offline after the models are cached, neither needs torch:

Install

pip install ovos-plugin-linguonnx

Models download from HuggingFace on first use and are cached under ~/.cache/linguonnx/models/<model_id>/. Constructing either plugin does not trigger a download; the first real call does. On a server, prefetch instead — see Prefetching.

Usage

from ovos_plugin_linguonnx import LinguONNXLangDetectPlugin, LinguONNXTranslatePlugin

det = LinguONNXLangDetectPlugin()
det.detect("bom dia")   # 'pt'

tx = LinguONNXTranslatePlugin()
tx.translate("Good morning", target="gl", source="en")

Documentation

• docs/language-detection.md — detection config, model choice, and language varieties.
• docs/translation.md — translation config, routing, and prefetching.
• docs/configuration.md — the full translation config reference, including the size/fallback routing budget.
• docs/docker.md — running both plugins as a server image.
• docs/errors.md — what translate() raises and how to map it to HTTP status codes.

Credits

• linguonnx — the detection and translation engine this package wraps.
• TigreGotico/glotlid-onnx — the ONNX export of GlotLID that linguonnx runs for detection.