i have answers:
+input label (create by preprocess.py ) are encoded to int array (index of char) in clude: a-z, <'> and ==> output dim(num classes) = num char(a-z,', space) + 1 (for black ctc) (ex: 29 in English)
but i see in vocab file, you haven add <SOS>,<EOS> and las_network.conf have add_label = 1...?
i dont understand input and output of las model, how it work? do i need to create data train again for las model or use same data with ds2 model?
and with las model, how many out classes - output_dim (if add orther char,...)? because i think with las model, it need other char: , ,... so num of ouput_dim will > num of ouput_dim ds2 model.
Looking forward to receiving your reply soon
Thanks!