Hello.
I am closely following the project.
Mastered the argos translate model training procedure.
What knowledge do I need to have to train high-quality models? To get closer to the results of google and yandex translate.
1 Like
Great to hear you’ve successfully trained models!
The most important thing for training high quality models is currating the data, the more quality data the better. My strategy has been to primarily use the datasets available on Opus and Wiktionary.
You can experiment with increasing the training time, by default the training scripts train for 10,000 OpenNMT epochs but I’ve also had good results with 10,000-100,000 especially for languages with a lot of data.
The OpenNMT Forum is also a good resource to see what has worked for other people, and I can answer questions here on the LibreTranslate forum.
1 Like