@lynxpda Hi and congratulations for all your efforts! You made it!
Tell us what do you expect from us
Expect more people to join this thread in the coming days.
Friendly,
@lynxpda Hi and congratulations for all your efforts! You made it!
Tell us what do you expect from us
Expect more people to join this thread in the coming days.
Friendly,
This is probably as good as the model is going to get in the immediate future. To help improving the dataset long term you can collect more high quality Kabyle data to train future models with.
Great work; thank you
Hello LibreTranslate community,
One year and half later, I launched this little experimental space running LibreTranslate with suggestions enabled for Kabyle language.
The instance is running 3 experimental kabyle models : 2 of them I built.
Link : LibreTranslate Kabyle - a Hugging Face Space by axxam
The suggestions are stored automatically in this dataset (using HF token upload to commit) as .csv
and .json
, whenever a new suggestion is detected. The dataset can be filtered later on.
Dataset : axxam/libretranslate-suggestions at main
Using the LibreTranslate API of this instance, I can compare translations from the 3 kabyle models + query Google Translate via Mozhi frontend.
Link : Kabyle Translate - a Hugging Face Space by axxam
And to review and submit a « best translation » I was thinking about something like that using LibreTranslate API :
Link : athmane mokraoui [BoF] ⏚ꝃ⌁⁂: "Une idée comme ça pour les « suggestions révisées…" - mstdn
Thank you all for you advices, I learned and still learning a lot in this community forum. I hope we will build a better argosmodels that could be run locally and self-hosted in the future with the help of the community.
Note : I use language codes kab, kab_comp, kab_comp2, kab_all
as a trick to have the kabyle models working together.
Hi,
A little update :
For info, the Kabyle experimental LibreTranslate instance and related tools were migrated to a called « Imsidag Localizers Community » on HF (for now) instead of my own account.
2 other admins were added.
New link : Imsidag-community (Imsidag)
I added new trained model called Kabyle (OS)
as default one. It is called OS as I added « backtranslation » of about 58000 « OpenSubtitles » strings to the training pipeline made with LibreTranslate locally.
Bonus experimental TTS in Kabyle (not published, only PoC) : athmane mokraoui [BoF] ⏚ꝃ⌁⁂: "Un peu de traduction propulsée par LibreTranslate…" - mstdn
Have a nice day.
Once you have a best Kabyle model done I can upload it to the index