Could random unicode noise potentially generate invalid UTF-8

[XapaJIaMnu]

I've seen models generate invalid UTF-8. I think that potentially for some groups of symbols, there could be order constraints (IE diacritics? Korean constructed blocks?

I'm pretty sure this is the case but haven't tested it. Someone could confirm? Also, potential fix?

OpusTrainer/src/opustrainer/modifiers/placeholders.py

Line 30 in 452c4ea

def get_random_unicode_words(min_length: int=2, max_length: int=10, max_words: int=3) -> List[str]: