question about code ‘sentence = klines_pos + enc_output[:,:,0,:].transpose(1,2)’

[atomishcv]

that is a good job! i have a question about code ‘sentence = klines_pos + enc_output[:,:,0,:].transpose(1,2)’
in the line_transformer.py's keylineencode. after the transformer encode the 'des', it out the dimension (1, 642,22,256). add the kline_poss(1, 642, 256). why you just take the third dimension, not like the maxpooling. i know the 22(21 + 1) represent the point on line.