Uncensored but used identity ideology fascist language intentionally without being requested.
If the llm is asked to write a german text, it does not use proper german language. It does use the so called "gender"-language which 98% of germans reject because its discriminating, fascist, and an insult for german history, culture and language.
In german the general word for all clerks is "Mitarbeiter". This includes male AND female clerks. In the fascist socialist artificial identy ideology "gender" variant they write instead "Mitarbeiter:innen" which is grammatically wrong and a try to control the way germans speak, discuss and think. Fascism in pure form. Your model uses this fascist form.
I strongly recommend fixing this asap. Otherwise your llm renders useless for germans and also makes its suspicous for other fields because if it uses fascist methods in language models it may using them in different aspects as well, which makes your llm untrustworthy from the beginning.
this is no disrespect of your work, but maybe you should select your advisors for german language more carefully.
Hey;
Please contact REKA directly ; link:
https://huggingface.co/RekaAI/reka-flash-3
This is beyond my control.
thanks;
Sounds a bit like the pushed gender neutral pronouns of LatinX which is rejected (and makes no sense in the language), yet pushed by ideologies for some reason regardless.
@simonlange1971 Such behavior can be easily controlled by instructions and sampling parameters. I personally had such scenarios, which I easily fixed only by tuning the sampling parameters.
In your case it's easier to apply instructions.