why does this target_pad need normalization here?

[Zhang-O]

LS-ACELoss/lsaceloss.py

Line 24 in 6286fed

targets_padded = F.normalize(targets_padded, p=1, dim=1)

[Zhang-O]

I had viewed your answer in another issue which you noted that both of methods worked well.

Another question is :
Is therr any need to ignore blank? Just as code here https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppocr/losses/ace_loss.py ,
where they ignored the blank through setting the "ignore_index=0".

Wish to receive your reply!!!