Please ignore. I misread the code.
I was reading the source code for the language model data loader and I noticed that the seq_len parameter is capped at 5 as shown here:
seq_len = max(5, int(np.random.normal(bptt, 5)))
Does that mean, regardless of how big our bptt is, the sequence that we feed to the RNN at a time is at most 5 words long? If so, what’s the rational behind it?
Thanks.