I’m working on a language modeling task, but this question can be generalized to any implementation with RNNs.
In batch learning, how does the fastai implementation (such as in MultiBatchRNN
) deal with variable length sentences? Additionally, how does the BPTT parameter come in to picture here?