30 lines
838 B
Python
30 lines
838 B
Python
|
import torch
|
||
|
|
|
||
|
|
|
||
|
|
class SoftmaxPower(torch.nn.Module):
|
||
|
|
dim: int | None
|
||
|
|
power: float
|
||
|
|
|
||
|
|
def __init__(self, power: float = 2.0, dim: int | None = None) -> None:
|
||
|
|
super().__init__()
|
||
|
|
self.dim = dim
|
||
|
|
self.power = power
|
||
|
|
|
||
|
|
def __setstate__(self, state):
|
||
|
|
super().__setstate__(state)
|
||
|
|
if not hasattr(self, "dim"):
|
||
|
|
self.dim = None
|
||
|
|
if not hasattr(self, "power"):
|
||
|
|
self.power = 2.0
|
||
|
|
|
||
|
|
def forward(self, input: torch.Tensor) -> torch.Tensor:
|
||
|
|
output: torch.Tensor = torch.abs(input).pow(self.power)
|
||
|
|
if self.dim is None:
|
||
|
|
output = output / output.sum()
|
||
|
|
else:
|
||
|
|
output = output / output.sum(dim=self.dim, keepdim=True)
|
||
|
|
return output
|
||
|
|
|
||
|
|
def extra_repr(self) -> str:
|
||
|
|
return f"dim={self.dim} ; power={self.power}"
|