Thank you. I agree for nn.RNN, but the tutorial is implementing a RNN from scratch, using nn.Linear, and the latter has batch dimension first (contrary to what the tutorial says), no ?