multilanguage train

Hi, how does detection in two languages work in general? Earlier I tried to detect using easyocr with the languages ru and en, in principle it works, but so far it does not detect well, so I decided to annotate my dataset and retrain the model.

but I ran into a problem, the fact is that the texts that I need to detect look like this: P6ОЖ00, 9OОЖ08, 3QОЖ2383 and so on, the meaning is like this (the first two can either be numbers or English letters, the second two are Cyrillic and then the numbers go). I can't teach, because I get an error related to encoding. Now I don't know how to use a dataset of 4000 texts to train a model.