本文整理汇总了Python中torch.Tensor.chunk方法的典型用法代码示例。如果您正苦于以下问题：Python Tensor.chunk方法的具体用法？Python Tensor.chunk怎么用？Python Tensor.chunk使用的例子？那么恭喜您, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类torch.Tensor的用法示例。
在下文中一共展示了Tensor.chunk方法的1个代码示例，这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞，您的评价将有助于系统推荐出更棒的Python代码示例。
示例1: _compute_loss
# 需要导入模块: from torch import Tensor [as 别名]
# 或者: from torch.Tensor import chunk [as 别名]
def _compute_loss(self,
lm_embeddings: torch.Tensor,
token_embeddings: torch.Tensor,
forward_targets: torch.Tensor,
backward_targets: torch.Tensor = None) -> Tuple[torch.Tensor, torch.Tensor]:
# If bidirectional, lm_embeddings is shape (batch_size, timesteps, dim * 2)
# If unidirectional, lm_embeddings is shape (batch_size, timesteps, dim)
# forward_targets, backward_targets (None in the unidirectional case) are
# shape (batch_size, timesteps) masked with 0
if self._bidirectional:
forward_embeddings, backward_embeddings = lm_embeddings.chunk(2, -1)
backward_loss = self._loss_helper(1, backward_embeddings, backward_targets, token_embeddings)
else:
forward_embeddings = lm_embeddings
backward_loss = None
forward_loss = self._loss_helper(0, forward_embeddings, forward_targets, token_embeddings)
return forward_loss, backward_loss