blob: c59744a4807f3d1566491e9df508446b34d0eeb1 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
|
"""Normalization layers for transfromers.
Copied from lucidrains:
https://github.com/lucidrains/x-transformers/blob/main/x_transformers/x_transformers.py
"""
from typing import Dict, Type
import torch
from torch import nn
from torch import Tensor
class ScaleNorm(nn.Module):
def __init__(self, normalized_shape: int, eps: float = 1.0e-5) -> None:
super().__init__()
self.scale = normalized_shape ** -0.5
self.eps = eps
self.g = nn.Parameter(torch.ones(1))
def forward(self, x: Tensor) -> Tensor:
"""Applies scale norm."""
norm = torch.norm(x, dim=-1, keepdim=True) * self.scale
return x / norm.clamp(min=self.eps) * self.g
class PreNorm(nn.Module):
def __init__(self, normalized_shape: int, fn: Type[nn.Module]) -> None:
super().__init__()
self.norm = nn.LayerNorm(normalized_shape)
self.fn = fn
def forward(self, x: Tensor, **kwargs: Dict) -> Tensor:
"""Applies pre norm."""
x = self.norm(x)
return self.fn(x, **kwargs)
|
