fr-ru / README.md
AmerAlreyahi's picture
Add OptimaMT fr-ru
fdc2ef7 verified
|
Raw History Blame Contribute Delete
2.53 kB
metadata
license: cc-by-4.0
language:
  - fr
  - ru
pipeline_tag: translation
library_name: transformers.js
base_model: Xenova/opus-mt-fr-ru
tags:
  - translation
  - marian
  - opus-mt
  - onnx

OptimaMT fr → ru (French to Russian)

Drop-in copy of Xenova/opus-mt-fr-ru at revision c2ec732b045a 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 51 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 = 1 539 → 287 MiB (-47 %) 513 → 264 MiB (-49 %) yes
ONNX Runtime defaults 287 → 224 MiB (-22 %) 200 → 168 MiB (-16 %) yes

FLORES-200 devtest (1,012 sentences): the same translations as the original, BLEU 15.07, chrF 44.17 (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/fr-ru", {
  dtype: "q8",
  session_options: { extra: { session: { disable_quant_qdq: "1" } } },
});
const [out] = await translator("Le phare se dresse sur la côte.");
console.log(out.translation_text);

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

python translate.py OptimaMT/fr-ru "Le phare se dresse sur la côte."

Files

Only onnx/decoder_model_merged_quantized.onnx is rebuilt; every other file is the original, byte for byte. Vocabulary: 64,377 tokens.

Licence and credits

  • Weights: OPUS-MT by the University of Helsinki, CC BY 4.0.
  • ONNX export: Xenova/opus-mt-fr-ru at revision c2ec732b045a.
  • Modified: the decoder's output projection was moved out of its two branches (graph change only). CC BY 4.0.