Is it possible to make `Conv1d` accept `PackedSequence` as input?

We know Conv1d input shape is (batch_size, channels, length), the channels act as input features of input sequence in my case.

My input sequences have different lengths. I know I can use nn.utils.rnn.pack_sequence for padding sequences to same lengths, but how to make Conv1d compatible with it like RNN?

Because I don’t want to the padding zeros which may effect training.