search

mozilla / translations

The code, training pipeline, and models that power Firefox Translations
https://mozilla.github.io/translations/
Mozilla Public License 2.0
154 stars 33 forks source link

In config generation, switch to two stage training when the mono data is too small #632

Open gregtatum opened 5 months ago

gregtatum commented 5 months ago

From: https://github.com/mozilla/firefox-translations-training/pull/620#discussion_r1612332379

But I found another use case when we don't want ["one-stage" teacher training when using a pre-trained backtranslations model]. If the amount of mono-trg data is too small (for example for en-lt) we still want to use two-stage. We don't want to loop over 5M back-translated sentences.