Concerning our translation server using #LibreTranslate at https://translate.fedilab.app we are doing our best to keep it available for everyone (Fedilab but also small instances or personal projects).
Our position requires us to be more careful with abuses (especially bots).
That's why sometimes you can notice some downtimes, but we still work on appropriate actions with the goal that it remains a useful service for everyone and not only for Fedilab.
libretranslate
While #Fedilab embeds its own translation feature (default option uses our #LibreTranslate service), you can also use #Lingva or #DeepL if you have an API key.
But we have to add the ability to let you use your instance as an option to translate messages. That will be done.
Today, I spent a long time trying to understand why the in-app translator using #LibreTranslate was still being abused.
I took drastic measures, because legit translations were stuck in the queue behind the abusers (now they are quick). But these measures also kicked out people using the API without abusing it. So please DM us if you're affected.
Tom
The next #Fedilab release will be able to detect a mismatch between the written text and the selected language of the message. This uses the #LibreTranslate server, so it has no extra weight on the final APK size and doesn't use any Google stuff. It will be turned off by default.
Thanks to @RachelThornSub for the suggestion.
Running a #LibreTranslate server takes work behind the scenes. My latest optimization adds a parallel queue system to avoid killing the service when it gets abused, but it can introduce immediate 503 errors for some users when the service is under heavy load. I will work on a smooth balance to stop bot abuse and avoid Fedilab/Holos users being restricted.
The only service being abused on my server is the translation one with #LibreTranslate, so I am going to introduce an API key that will discard a lot of bots.
I am a little confused by the reactions to my last suggestion. It makes me wonder if I should also stop #LibreTranslate in #Fedilab, because it also uses AI. But it was already trained. When I use it, I only run it, I do not train it. I would like to hear your views before I decide anything.
We can keep talking here calmly, I am really open to it.