xma.utils.debugging

print_gradient(x: Tensor, name: str) → Tensor

print gradient in backward (use only for debugging)

Parameters:

• x (torch.Tensor) – input tensor

• name (str) – additional metadata for the tensor (typically the name of the tensor)

Returns:

output tensor same as input tensor

Return type:

torch.Tensor