About multi-layer forward pass

[daleigehhh]

Hi, @KimUyen
I am working on a local map prediction problem and I want to use Bi-ConvLSTM to solve my problem.
In your implementation ConvLSTM part, line 238:

## LSTM forward direction input_fw = input_tensor for layer_num in range(self.layer_num): h, c = hidden_states[layer_num] output_inner = [] for t in range(seq_len): h, c = self.cells_fw[layer_num](input_tensor=input_fw[:, t, :, :, :], cur_state=[h, c]) output_inner.append(h) layer_output = torch.stack(output_inner, dim=1) layer_outputs_fw.append(layer_output) layer_outputs = torch.stack(layer_outputs_fw, dim=1) last_state_fw = [h, c]
why the input of every layer is the same, I thought it should be the hidden state of the last layer, am I right?