Training Tesseract with new fonts accuracy

I'm trying to train Tesseract with 10 new Farsi fonts and for each font I have 150 pages of tiff file. I started training for 1000000 iteration but still get accuracy~88%. Does anyone here have any experience with Tesseract training parameters like adam beta, momentum and network specs? I want to change them to have better train accuracy. This is my sh codes for training new fonts:

rm -rf trainingv2/output/*
OMP_THREAD_LIMIT=24 lstmtraining \
    --continue_from latest/tesseract/tessdata/fas.lstm \
    --model_output trainingv2/output/Bfonts \
    --traineddata latest/tesseract/tessdata/fas.traineddata \
    --train_listfile trainingv2/trainforB/fas.training_files.txt \
    --max_image_MB 35000\
    --randomly_rotate true\
    --debug_interval 10\
    --max_iterations 1000000\