--- license: cc-by-4.0 language: - vi - en pipeline_tag: translation library_name: transformers.js base_model: Xenova/opus-mt-vi-en tags: - translation - marian - opus-mt - onnx --- # OptimaMT vi → en (Vietnamese to English) Drop-in copy of [Xenova/opus-mt-vi-en](https://huggingface.co/Xenova/opus-mt-vi-en) at revision `541dfd72ec07` 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](https://github.com/AmerAlreyahi/OptimaMT): [all 51 pairs](https://huggingface.co/collections/OptimaMT/optimamt-lighter-opus-mt-onnx-models-6af228dc9cd91f015e56e869), named `OptimaMT/-`. **Used in production:** - [OpenCircle](https://github.com/AmerAlreyahi/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` | 465 → 255 MiB (-45 %) | 442 → 232 MiB (-47 %) | yes | | ONNX Runtime defaults | 251 → 198 MiB (-21 %) | 181 → 154 MiB (-15 %) | yes | FLORES-200 devtest (1,012 sentences): the same translations as the original, BLEU 24.31, chrF 51.95 (greedy). ONNX Runtime 1.31.1, CPU, 4 threads (Linux; Windows in a VM). Interactive results: [OptimaMT/results](https://huggingface.co/spaces/OptimaMT/results) · method: [REPORT.md](https://github.com/AmerAlreyahi/OptimaMT/blob/main/REPORT.md). ## Use ```js import { pipeline } from "@huggingface/transformers"; const translator = await pipeline("translation", "OptimaMT/vi-en", { dtype: "q8", session_options: { extra: { session: { disable_quant_qdq: "1" } } }, }); const [out] = await translator("Ngọn hải đăng đứng trên bờ biển."); console.log(out.translation_text); ``` Python ([translate.py](https://github.com/AmerAlreyahi/OptimaMT/blob/main/scripts/translate.py): onnxruntime, sentencepiece, numpy): ```bash python translate.py OptimaMT/vi-en "Ngọn hải đăng đứng trên bờ biển." ``` ## Files Only `onnx/decoder_model_merged_quantized.onnx` is rebuilt; every other file is the original, byte for byte. Vocabulary: 53,739 tokens. ## Licence and credits - Weights: [OPUS-MT](https://huggingface.co/Helsinki-NLP/opus-mt-vi-en) by the University of Helsinki, CC BY 4.0. - ONNX export: [Xenova/opus-mt-vi-en](https://huggingface.co/Xenova/opus-mt-vi-en) at revision `541dfd72ec07`. - Modified: the decoder's output projection was moved out of its two branches (graph change only). CC BY 4.0.