espnet2.enh.layers.ncsnpp_utils.layers.contract_inner

source

espnet2.enh.layers.ncsnpp_utils.layers.contract_inner(x, y)

Compute the inner product of two tensors along the last axis of x and the first axis of y.

This function utilizes Einstein summation convention to compute the inner product, effectively performing a tensordot operation with the specified axes.

• Parameters:
  • x (torch.Tensor) – The first input tensor of shape (N, …, M).
  • y (torch.Tensor) – The second input tensor of shape (N, …, P).
• Returns: The result of the inner product of x and y, with shape (N, …, P).
• Return type: torch.Tensor

Examples

>>> x = torch.randn(2, 3, 4)
>>> y = torch.randn(2, 4, 5)
>>> result = contract_inner(x, y)
>>> result.shape
torch.Size([2, 3, 5])

NOTE

• The function assumes that the last dimension of x and the first dimension of y are compatible for the inner product.