README / README.md
AmerAlreyahi's picture
OptimaMT organization card
22cdea4 verified
|
Raw History Blame Contribute Delete
1.31 kB
metadata
title: OptimaMT
emoji: 🌐
short_description: Lighter OPUS-MT translation models for ONNX Runtime
colorFrom: blue
colorTo: indigo
sdk: static
pinned: false

OptimaMT

ONNX Runtime's fast mode at half the memory: 2.3–2.5× faster than the default, same translations.

51 OPUS-MT translation models for ONNX Runtime and transformers.js: drop-in copies of the int8 exports with a lighter merged decoder. Same weights, same tokens.

Memory after load (median; Linux 51 pairs, Windows 49) Linux Windows
disable_quant_qdq = 1 525 → 281 MiB (-46 %) 500 → 257 MiB (-49 %)
ONNX Runtime defaults 277 → 217 MiB (-22 %) 196 → 166 MiB (-16 %)

Used in production:

  • OpenCircle, a desktop screen translator for Windows and Linux, ships these models since v0.1.4.

All models · repo ids OptimaMT/<src>-<tgt>, e.g. OptimaMT/en-ar · interactive results · method and code

Models CC BY 4.0 (OPUS-MT, University of Helsinki; ONNX exports by Xenova); code MIT.