Time Delayed NN implemented in PyTorch.
kernels = [(1, 25), (2, 50), (3, 75), (4, 100), (5, 125), (6, 150)]
tdnn = TDNN(kernels, input_embedding_size)
# in is tensor with shape [batch_size, max_seq_len, max_word_len, input_embedding_size]
out = tdnn(in)