Hi All,

CTC as a loss function quite often used in sequence to sequence models isn't currently implemented and installing binding to warp-ctc is quite hacky (and warp-ctc isn't developed anymore). I'd like to add it natively to pytorch, probably based on the implementation of clstm.

My question is first of if that is even desired and second where to include the code? Is torch/lib/THNN/generic the right location?