Utilities for Transformer
10import torchSubsequent mask to mask out data from future (subsequent) time steps
13def subsequent_mask(seq_len):17 mask = torch.tril(torch.ones(seq_len, seq_len)).to(torch.bool).unsqueeze(-1)
18 return mask10import torch13def subsequent_mask(seq_len):17 mask = torch.tril(torch.ones(seq_len, seq_len)).to(torch.bool).unsqueeze(-1)
18 return mask