Splitting tokenization part to reduce time

[yusufcakmakk]

Hi all,

I have a suggestion for the tokenization part in here. When we start the process, and we have different block sizes, it calculates all tokens again and then grouping them. Instead, the tokenization step can be used once and then grouped based on block size.

What do you think about this improvement?

[iMountTai]

This can be achieved by modifying here.