--- language: - be - cs - pl - ru - uk - zle - zlw tags: - translation - opus-mt-tc license: cc-by-4.0 model-index: - name: opus-mt-tc-big-zle-zlw results: - task: name: Translation rus-ces type: translation args: rus-ces dataset: name: flores101-devtest type: flores_101 args: rus ces devtest metrics: - name: BLEU type: bleu value: 23.1 - task: name: Translation ukr-ces type: translation args: ukr-ces dataset: name: flores101-devtest type: flores_101 args: ukr ces devtest metrics: - name: BLEU type: bleu value: 25.1 - task: name: Translation bel-pol type: translation args: bel-pol dataset: name: tatoeba-test-v2021-08-07 type: tatoeba_mt args: bel-pol metrics: - name: BLEU type: bleu value: 47.1 - task: name: Translation rus-ces type: translation args: rus-ces dataset: name: tatoeba-test-v2021-08-07 type: tatoeba_mt args: rus-ces metrics: - name: BLEU type: bleu value: 53.4 - task: name: Translation rus-pol type: translation args: rus-pol dataset: name: tatoeba-test-v2021-08-07 type: tatoeba_mt args: rus-pol metrics: - name: BLEU type: bleu value: 53.7 - task: name: Translation ukr-ces type: translation args: ukr-ces dataset: name: tatoeba-test-v2021-08-07 type: tatoeba_mt args: ukr-ces metrics: - name: BLEU type: bleu value: 58.0 - task: name: Translation ukr-pol type: translation args: ukr-pol dataset: name: tatoeba-test-v2021-08-07 type: tatoeba_mt args: ukr-pol metrics: - name: BLEU type: bleu value: 57.0 - task: name: Translation rus-ces type: translation args: rus-ces dataset: name: newstest2013 type: wmt-2013-news args: rus-ces metrics: - name: BLEU type: bleu value: 26.0 --- # opus-mt-tc-big-zle-zlw Neural machine translation model for translating from East Slavic languages (zle) to West Slavic languages (zlw). This model is part of the [OPUS-MT project](https://github.com/Helsinki-NLP/Opus-MT), an effort to make neural machine translation models widely available and accessible for many languages in the world. All models are originally trained using the amazing framework of [Marian NMT](https://marian-nmt.github.io/), an efficient NMT implementation written in pure C++. The models have been converted to pyTorch using the transformers library by huggingface. Training data is taken from [OPUS](https://opus.nlpl.eu/) and training pipelines use the procedures of [OPUS-MT-train](https://github.com/Helsinki-NLP/Opus-MT-train). * Publications: [OPUS-MT – Building open translation services for the World](https://aclanthology.org/2020.eamt-1.61/) and [The Tatoeba Translation Challenge – Realistic Data Sets for Low Resource and Multilingual MT](https://aclanthology.org/2020.wmt-1.139/) (Please, cite if you use this model.) ``` @inproceedings{tiedemann-thottingal-2020-opus, title = "{OPUS}-{MT} {--} Building open translation services for the World", author = {Tiedemann, J{\"o}rg and Thottingal, Santhosh}, booktitle = "Proceedings of the 22nd Annual Conference of the European Association for Machine Translation", month = nov, year = "2020", address = "Lisboa, Portugal", publisher = "European Association for Machine Translation", url = "https://aclanthology.org/2020.eamt-1.61", pages = "479--480", } @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", month = nov, year = "2020", address = "Online", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2020.wmt-1.139", pages = "1174--1182", } ``` ## Model info * Release: 2022-03-23 * source language(s): bel rus ukr * target language(s): ces pol * valid target language labels: >>ces<< >>pol<< * model: transformer-big * data: opusTCv20210807+bt ([source](https://github.com/Helsinki-NLP/Tatoeba-Challenge)) * tokenization: SentencePiece (spm32k,spm32k) * original model: [opusTCv20210807+bt_transformer-big_2022-03-23.zip](https://object.pouta.csc.fi/Tatoeba-MT-models/zle-zlw/opusTCv20210807+bt_transformer-big_2022-03-23.zip) * more information released models: [OPUS-MT zle-zlw README](https://github.com/Helsinki-NLP/Tatoeba-Challenge/tree/master/models/zle-zlw/README.md) * more information about the model: [MarianMT](https://huggingface.co/docs/transformers/model_doc/marian) This is a multilingual translation model with multiple target languages. A sentence initial language token is required in the form of `>>id<<` (id = valid target language ID), e.g. `>>ces<<` ## Usage A short example code: ```python from transformers import MarianMTModel, MarianTokenizer src_text = [ ">>pol<< Это метафора.", ">>pol<< Что вы делали?" ] model_name = "pytorch-models/opus-mt-tc-big-zle-zlw" tokenizer = MarianTokenizer.from_pretrained(model_name) model = MarianMTModel.from_pretrained(model_name) translated = model.generate(**tokenizer(src_text, return_tensors="pt", padding=True)) for t in translated: print( tokenizer.decode(t, skip_special_tokens=True) ) # expected output: # To metafora. # Co robiliście? ``` You can also use OPUS-MT models with the transformers pipelines, for example: ```python from transformers import pipeline pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-tc-big-zle-zlw") print(pipe(">>pol<< Это метафора.")) # expected output: To metafora. ``` ## Benchmarks * test set translations: [opusTCv20210807+bt_transformer-big_2022-03-23.test.txt](https://object.pouta.csc.fi/Tatoeba-MT-models/zle-zlw/opusTCv20210807+bt_transformer-big_2022-03-23.test.txt) * test set scores: [opusTCv20210807+bt_transformer-big_2022-03-23.eval.txt](https://object.pouta.csc.fi/Tatoeba-MT-models/zle-zlw/opusTCv20210807+bt_transformer-big_2022-03-23.eval.txt) * benchmark results: [benchmark_results.txt](benchmark_results.txt) * benchmark output: [benchmark_translations.zip](benchmark_translations.zip) | langpair | testset | chr-F | BLEU | #sent | #words | |----------|---------|-------|-------|-------|--------| | bel-pol | tatoeba-test-v2021-08-07 | 0.65517 | 47.1 | 287 | 1706 | | rus-ces | tatoeba-test-v2021-08-07 | 0.69695 | 53.4 | 2934 | 16831 | | rus-pol | tatoeba-test-v2021-08-07 | 0.72176 | 53.7 | 3543 | 21505 | | ukr-ces | tatoeba-test-v2021-08-07 | 0.73149 | 58.0 | 1787 | 8550 | | ukr-pol | tatoeba-test-v2021-08-07 | 0.74649 | 57.0 | 2519 | 13201 | | bel-ces | flores101-devtest | 0.41248 | 11.1 | 1012 | 22101 | | bel-pol | flores101-devtest | 0.42240 | 10.2 | 1012 | 22520 | | rus-ces | flores101-devtest | 0.50971 | 23.1 | 1012 | 22101 | | rus-pol | flores101-devtest | 0.48672 | 18.4 | 1012 | 22520 | | ukr-ces | flores101-devtest | 0.52482 | 25.1 | 1012 | 22101 | | ukr-pol | flores101-devtest | 0.48790 | 18.8 | 1012 | 22520 | | rus-ces | newstest2012 | 0.45834 | 18.8 | 3003 | 65456 | | rus-ces | newstest2013 | 0.52364 | 26.0 | 3000 | 57250 | ## Acknowledgements The work is supported by the [European Language Grid](https://www.european-language-grid.eu/) as [pilot project 2866](https://live.european-language-grid.eu/catalogue/#/resource/projects/2866), by the [FoTran project](https://www.helsinki.fi/en/researchgroups/natural-language-understanding-with-cross-lingual-grounding), funded by the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 771113), and the [MeMAD project](https://memad.eu/), funded by the European Union’s Horizon 2020 Research and Innovation Programme under grant agreement No 780069. We are also grateful for the generous computational resources and IT infrastructure provided by [CSC -- IT Center for Science](https://www.csc.fi/), Finland. ## Model conversion info * transformers version: 4.16.2 * OPUS-MT git hash: 1bdabf7 * port time: Thu Mar 24 00:50:29 EET 2022 * port machine: LM0-400-22516.local