I understood in part1 leasson 6 , mini batch is bptt*batch_size and Jeremy says, bptt is approximate. In the particular example it is around 80 but it varies from mini batch to mini batch.
I am curious to know, how do you feed each mini batch to lstm for training?
According to my understanding, during 1st mini batch , if I assume bptt =78, then , unrolled lstm will have 78 lstm units .
Now during 2nd mini batch if bptt =70, then unrolled lstm will have 70 units of lstm ? Is this correct?
Considering batch_size is fixed, if bptt varies, then no of records in each mini batch varies?
Follow up question
Bptt is replacenent of seq length (rnn)?
Is seq_len and bptt same ? By seq_len , I mean no of lstm units (unrolled) in a layer.