29 lines
843 B
Python
29 lines
843 B
Python
|
# Copyright (c) Meta Platforms, Inc. and affiliates.
|
||
|
|
# All rights reserved.
|
||
|
|
#
|
||
|
|
# This source code is licensed under the license found in the
|
||
|
|
# LICENSE file in the root directory of this source tree.
|
||
|
|
|
||
|
|
"""Normalization modules."""
|
||
|
|
|
||
|
|
import typing as tp
|
||
|
|
|
||
|
|
import einops
|
||
|
|
import torch
|
||
|
|
from torch import nn
|
||
|
|
|
||
|
|
|
||
|
|
class ConvLayerNorm(nn.LayerNorm):
|
||
|
|
"""
|
||
|
|
Convolution-friendly LayerNorm that moves channels to last dimensions
|
||
|
|
before running the normalization and moves them back to original position right after.
|
||
|
|
"""
|
||
|
|
def __init__(self, normalized_shape: tp.Union[int, tp.List[int], torch.Size], **kwargs):
|
||
|
|
super().__init__(normalized_shape, **kwargs)
|
||
|
|
|
||
|
|
def forward(self, x):
|
||
|
|
x = einops.rearrange(x, 'b ... t -> b t ...')
|
||
|
|
x = super().forward(x)
|
||
|
|
x = einops.rearrange(x, 'b t ... -> b ... t')
|
||
|
|
return
|