OptimaMT uk → ru (Ukrainian to Russian)

Drop-in copy of Xenova/opus-mt-uk-ru at revision cfb76ccd2077 with a lighter merged decoder: ONNX Runtime's fast mode (disable_quant_qdq = 1, 2.3–2.5× faster) at half the memory. Same weights, same tokens. Part of OptimaMT: all 49 pairs, named OptimaMT/<src>-<tgt>.

Used in production:

  • OpenCircle, a desktop screen translator for Windows and Linux, ships these models since v0.1.4.
Memory after load Linux Windows Identical tokens
disable_quant_qdq = 1 494 → 268 MiB (-46 %) 471 → 243 MiB (-48 %) yes
ONNX Runtime defaults 265 → 208 MiB (-21 %) 188 → 158 MiB (-16 %) yes

FLORES-200 devtest (1,012 sentences): the same translations as the original, BLEU 23.28, chrF 51.04 (greedy).

ONNX Runtime 1.31.1, CPU, 4 threads (Linux; Windows in a VM). Interactive results: OptimaMT/results · method: REPORT.md.

Use

import { pipeline } from "@huggingface/transformers";

const translator = await pipeline("translation", "OptimaMT/uk-ru", {
  dtype: "q8",
  session_options: { extra: { session: { disable_quant_qdq: "1" } } },
});
const [out] = await translator("Маяк стоїть на березі моря.");
console.log(out.translation_text);

Python (translate.py: onnxruntime, sentencepiece, numpy):

python translate.py OptimaMT/uk-ru "Маяк стоїть на березі моря."

Files

Only onnx/decoder_model_merged_quantized.onnx is rebuilt; every other file is the original, byte for byte. Vocabulary: 57,777 tokens.

Licence and credits

  • Weights: OPUS-MT by the University of Helsinki, CC BY 4.0.
  • ONNX export: Xenova/opus-mt-uk-ru at revision cfb76ccd2077.
  • Modified: the decoder's output projection was moved out of its two branches (graph change only). CC BY 4.0.
Downloads last month
37
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for OptimaMT/uk-ru

Quantized
(1)
this model

Space using OptimaMT/uk-ru 1

Collection including OptimaMT/uk-ru