New argos model zh_ru and ru_zh

Hello,

This is a new model for the zh-ru and ru-zh language.

I trained the model with argos train.

The model is trained on cleaned text corpora https://opus.nlpl.eu/
corpus: bible-uedin_v1_ru-zh, CCMatrix_v1_ru-zh, ELRC-wikipedia_health_v1_ru-zh, EUbookshop_v2_ru-zh, infopankki_v1_ru-zh, LinguaTools-WikiTitles_v2014_ru-zh, MultiParaCrawl_v9b_ru-zh, MultiUN_v1_ru-zh, NeuLab-TedTalks_v1_ru-zh, News-Commentary_v16_ru-zh, PHP_v1_ru-zh, QED_v2.0a_ru-zh, Tanzil_v1_ru-zh, TED2020_v1_ru-zh, tico-19_v2020-10-28_ru-zh, Ubuntu_v14.10_ru-zh, UNPC_v1_0_ru-zh, WikiMatrix_v1_ru-zh, wikimedia_v20210402_ru-zh

The check was carried out on the newstest-corpus 2017-2019-ru-zh

Bleu score on 100 000 train step: 11.87

model
translate-ru_zh-1_7.argosmodel

and model
translate-zh_ru-1_7.argosmodel

2 Likes

Thanks for training these!

For now I’m not adding non English language pairs to the default package index to keep the install size small. People can download and use the packages from this thread though.

1 Like