espnet.nets.pytorch_backend.transformer.mask.subsequent_mask

source

espnet.nets.pytorch_backend.transformer.mask.subsequent_mask(size, device='cpu', dtype=torch.bool)

Create mask for subsequent steps (size, size).

• Parameters:
  • size (int) – size of mask
  • device (str) – “cpu” or “cuda” or torch.Tensor.device
  • dtype (torch.dtype) – result dtype
• Return type: torch.Tensor

>>> subsequent_mask(3)
[[1, 0, 0],
 [1, 1, 0],
 [1, 1, 1]]