opus-mt-es-ro-ct2-int8

OPUS-MT Spanish to Romanian (CTranslate2 int8)

This model is a CTranslate2 conversion of Helsinki-NLP/opus-mt-es-ro with int8 quantization for efficient CPU inference.

Languages

• Source language (code): Spanish (es)
• Target language (code): Romanian (ro)

Model details

Usage

With CTranslate2

import ctranslate2
import sentencepiece as spm

model_path = "mijuanlo/opus-mt-es-ro-ct2-int8"
sp_source = spm.SentencePieceProcessor(f"{model_path}/source.spm")
sp_target = spm.SentencePieceProcessor(f"{model_path}/target.spm")

translator = ctranslate2.Translator(model_path)

def translate(text: str) -> str:
    input_ids = sp_source.encode(text, out_type=str)
    results = translator.translate_batch([input_ids], beam_size=4)
    output_ids = results[0].hypotheses[0]
    return sp_target.decode(output_ids)

print(translate("¡Hola, mundo!"))

Load from Hugging Face Hub

from huggingface_hub import snapshot_download
import ctranslate2
import sentencepiece as spm

model_id = "mijuanlo/opus-mt-es-ro-ct2-int8"
model_path = snapshot_download(repo_id=model_id)
translator = ctranslate2.Translator(model_path)
sp_source = spm.SentencePieceProcessor(f"{model_path}/source.spm")
sp_target = spm.SentencePieceProcessor(f"{model_path}/target.spm")

Conversion details

ct2-transformers-converter \
    --model Helsinki-NLP/opus-mt-es-ro \
    --output_dir opus-mt-es-ro-ct2-int8 \
    --quantization int8 \
    --force

License

MIT. The original model from Helsinki-NLP is also MIT.

Citation

@inproceedings{tiedemann-2020-tatoeba,
    title = {The Tatoeba Translation Challenge – Realistic Data Sets for Low Resource and Multilingual MT},
    author = {Tiedemann, J{\"o}rg},
    booktitle = {Proceedings of the Fifth Conference on Machine Translation (WMT)},
    year = {2020},
    publisher = {Association for Computational Linguistics},
}