Support packed sequence in our RNN implementation
Opened this issue · 0 comments
tobmi1 commented
Right now our RNN implementation (especially the dilated RNN) supports only tensor inputs. Ideally, it would also work with packed sequences like the vanilla torch RNNs