Why can't handwritten and after 4 lines?

[Amricreate]

I actually created an OCR model for detecting Grantha scripts rendered in Noto Sans Grantha. This model can predict printed characters in multiple lines.

While I was doing the same for handwritten form of the same script. Initially I thought that training can be done for single line images, but I could train up to 4-line images without any segmentation fault. I actually synthesized those images by creating a blank and pasting each corresponding image saved locally line by line based on the text script created.

But after four lines, I created 5-line images which didn't work as well.

With paddings (only between each line):

I did the same with paddings to increase the space between characters in adjacent lines. Instead, it produced this error:

TESSDATA_PREFIX=../tesseract/tessdata make training MODEL_NAME=fourline START_MODEL=tam PSM=6 RANDOM_SEED=1234 TESSDATA=../tesseract/tessdata
You are using make version: 4.3
unicharset_extractor --output_unicharset "data/fourline/my.unicharset" --norm_mode 2 "data/fourline/all-gt"
Extracting unicharset from plain text file data/fourline/all-gt
Wrote unicharset file data/fourline/my.unicharset
merge_unicharsets data/tam/fourline.lstm-unicharset data/fourline/my.unicharset "data/fourline/unicharset"
Loaded unicharset of size 99 from file data/tam/fourline.lstm-unicharset
Loaded unicharset of size 37 from file data/fourline/my.unicharset
Wrote unicharset file data/fourline/unicharset.
tesseract "data/fourline-ground-truth/fourline_4.tif" data/fourline-ground-truth/fourline_4 --psm 6 lstm.train
tesseract "data/fourline-ground-truth/fourline_3.tif" data/fourline-ground-truth/fourline_3 --psm 6 lstm.train
tesseract "data/fourline-ground-truth/fourline_2.tif" data/fourline-ground-truth/fourline_2 --psm 6 lstm.train
tesseract "data/fourline-ground-truth/fourline_1.tif" data/fourline-ground-truth/fourline_1 --psm 6 lstm.train
tesseract "data/fourline-ground-truth/fourline_0.tif" data/fourline-ground-truth/fourline_0 --psm 6 lstm.train
python3 shuffle.py 1234 "data/fourline/all-lstmf"
python3 generate_eval_train.py data/fourline/all-lstmf 0.90
combine_lang_model
--input_unicharset data/fourline/unicharset
--script_dir data/langdata
--numbers data/fourline/fourline.numbers
--puncs data/fourline/fourline.punc
--words data/fourline/fourline.wordlist
--output_dir data

--lang fourline
Failed to read data from: data/fourline/fourline.wordlist
Failed to read data from: data/fourline/fourline.punc
Failed to read data from: data/fourline/fourline.numbers
Loaded unicharset of size 133 from file data/fourline/unicharset
Setting unichar properties
Setting script properties
Failed to load script unicharset from:data/langdata/Grantha.unicharset
Warning: properties incomplete for index 11 = ்
Warning: properties incomplete for index 16 = ்‌
Warning: properties incomplete for index 93 = ₹
Warning: properties incomplete for index 99 = 𑌳
Warning: properties incomplete for index 100 = 𑌝
Warning: properties incomplete for index 101 = 𑌶
Warning: properties incomplete for index 102 = 𑌥
Warning: properties incomplete for index 103 = 𑌖
Warning: properties incomplete for index 104 = 𑌤
Warning: properties incomplete for index 105 = 𑌙
Warning: properties incomplete for index 106 = 𑌮
Warning: properties incomplete for index 107 = 𑌯
Warning: properties incomplete for index 108 = 𑌰
Warning: properties incomplete for index 109 = 𑌪
Warning: properties incomplete for index 110 = 𑌸
Warning: properties incomplete for index 111 = 𑌵
Warning: properties incomplete for index 112 = 𑌜
Warning: properties incomplete for index 113 = 𑌘
Warning: properties incomplete for index 114 = 𑌛
Warning: properties incomplete for index 115 = 𑌣
Warning: properties incomplete for index 116 = 𑌡
Warning: properties incomplete for index 117 = 𑌬
Warning: properties incomplete for index 118 = 𑌗
Warning: properties incomplete for index 119 = 𑌦
Warning: properties incomplete for index 120 = 𑌢
Warning: properties incomplete for index 121 = 𑌲
Warning: properties incomplete for index 122 = 𑌞
Warning: properties incomplete for index 123 = 𑌠
Warning: properties incomplete for index 124 = 𑌭
Warning: properties incomplete for index 125 = 𑌷
Warning: properties incomplete for index 126 = 𑌟
Warning: properties incomplete for index 127 = 𑌚
Warning: properties incomplete for index 128 = 𑌕
Warning: properties incomplete for index 129 = 𑌹
Warning: properties incomplete for index 130 = 𑌨
Warning: properties incomplete for index 131 = 𑌫
Warning: properties incomplete for index 132 = 𑌧
Config file is optional, continuing...
Failed to read data from: data/langdata/fourline/fourline.config
Null char=2
Created data/fourline/fourline.traineddata
lstmtraining
--debug_interval 0
--traineddata data/fourline/fourline.traineddata
--old_traineddata ../tesseract/tessdata/tam.traineddata
--continue_from data/tam/fourline.lstm
--learning_rate 0.0001
--model_output data/fourline/checkpoints/fourline
--train_listfile data/fourline/list.train
--eval_listfile data/fourline/list.eval
--max_iterations 10000
--target_error_rate 0.01
2>&1 | tee -a data/fourline/training.log
Loaded file data/tam/fourline.lstm, unpacking...
Warning: LSTMTrainer deserialized an LSTMRecognizer!
Code range changed from 99 to 132!
Num (Extended) outputs,weights in Series:
1,36,0,1:1, 0
Num (Extended) outputs,weights in Series:
C3,3:9, 0
Ft16:16, 160
Total weights = 160
[C3,3Ft16]:16, 160
Mp3,3:16, 0
TxyLfys48:48, 12480
Lfx96:96, 55680
RxLrx96:96, 74112
Lfx192:192, 221952
Fc132:132, 25476
Total weights = 389860
Previous null char=2 mapped to 131
Continuing from data/tam/fourline.lstm
Compute CTC targets failed for data/fourline-ground-truth/fourline_0.lstmf!
Compute CTC targets failed for data/fourline-ground-truth/fourline_3.lstmf!
Compute CTC targets failed for data/fourline-ground-truth/fourline_2.lstmf!
Compute CTC targets failed for data/fourline-ground-truth/fourline_0.lstmf!
Compute CTC targets failed for data/fourline-ground-truth/fourline_3.lstmf!
Compute CTC targets failed for data/fourline-ground-truth/fourline_2.lstmf!
Compute CTC targets failed for data/fourline-ground-truth/fourline_0.lstmf!
Compute CTC targets failed for data/fourline-ground-truth/fourline_3.lstmf!
Compute CTC targets failed for data/fourline-ground-truth/fourline_2.lstmf!
Compute CTC targets failed for data/fourline-ground-truth/fourline_0.lstmf!
Compute CTC targets failed for data/fourline-ground-truth/fourline_3.lstmf!
.........

Two questions.

1. If Tesseract couldn't perform OCR on multi line handwritten images, how did it succeed in 4 lines?
2. Even if the handwritten characters are arranged exactly in the same form as printed images, why did't it work.

I would really appreciate your help.
I could provide you further details if needed.