Some characters are missing from the current model

[DesBw]

I know this is not really about AddisOCR. But, I am hoping that you have the knowhow to finetune the original training to improve it. I have been trying to fine tune the traineddata. But, I really struggled a lot because I have limited experience with the technology.

As you wrote in the Medium article, the accuracy of the current model is pretty high. I am also surprised to find the accurate to be very good. But, it could have been even better if some characters were not missing from the original training.
It cannot recognize the letter ሡ for example. It gets removed or replaced by a similar character in every line it appears.

እያሡ for example comes out as እያ in the best model and as እያሥ in the fast model.

Do you know how to fine tune it to add the missing characters?