How to make a self-made dataset for rnn

I’m making my own dataset for rnn
I don’t know where the internal state (time) of GRU starts
The size is 1 batch = 100,256

In pytorch’s GRU, which do you start from right (deta [batch] [0]) or left (deta [batch] [100])?

RNNs expect an input in the shape [seq_len, batch_size, features] in the default setup.
The temporal dimension is thus in dim0 and the first time step would be at index 0.