token split issue

[HITzhongyu]

Hi，
thanks for your pretrain model
I wonder if I use all the seqence at same length such as 70, the token has the same split?
And I try to use the pretrain model as feature extract, the output of the pretrain whether has the same length?

[jaouiwassim]

Could you reformulate the question please ? I didn't get what you mean by "split"