-
Notifications
You must be signed in to change notification settings - Fork 0
/
layer.py
224 lines (200 loc) · 8.14 KB
/
layer.py
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
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
from functools import partial
import torch
from torch import nn
from torch.nn.modules.transformer import (
_get_activation_fn,
Module,
Tensor,
Optional,
MultiheadAttention,
Linear,
Dropout,
LayerNorm,
)
from torch.utils.checkpoint import checkpoint
class TransformerEncoderLayer(Module):
r"""TransformerEncoderLayer is made up of self-attn and feedforward network.
This standard encoder layer is based on the paper "Attention Is All You Need".
Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez,
Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in
Neural Information Processing Systems, pages 6000-6010. Users may modify or implement
in a different way during application.
Args:
d_model: the number of expected features in the input (required).
nhead: the number of heads in the multiheadattention models (required).
dim_feedforward: the dimension of the feedforward network model (default=2048).
dropout: the dropout value (default=0.1).
activation: the activation function of intermediate layer, relu or gelu (default=relu).
layer_norm_eps: the eps value in layer normalization components (default=1e-5).
batch_first: If ``True``, then the input and output tensors are provided
as (batch, seq, feature). Default: ``False``.
Examples::
>>> encoder_layer = nn.TransformerEncoderLayer(d_model=512, nhead=8)
>>> src = torch.rand(10, 32, 512)
>>> out = encoder_layer(src)
Alternatively, when ``batch_first`` is ``True``:
>>> encoder_layer = nn.TransformerEncoderLayer(d_model=512, nhead=8, batch_first=True)
>>> src = torch.rand(32, 10, 512)
>>> out = encoder_layer(src)
"""
__constants__ = ["batch_first"]
def __init__(
self,
d_model,
nhead,
dim_feedforward=2048,
dropout=0.1,
activation="relu",
layer_norm_eps=1e-5,
batch_first=False,
pre_norm=False,
device=None,
dtype=None,
recompute_attn=False,
save_trainingset_representations=False,
) -> None:
factory_kwargs = {"device": device, "dtype": dtype}
super().__init__()
self.self_attn = MultiheadAttention(
d_model, nhead, dropout=dropout, batch_first=batch_first, **factory_kwargs
)
# Implementation of Feedforward model
self.linear1 = Linear(d_model, dim_feedforward, **factory_kwargs)
self.dropout = Dropout(dropout)
self.linear2 = Linear(dim_feedforward, d_model, **factory_kwargs)
self.norm1 = LayerNorm(d_model, eps=layer_norm_eps, **factory_kwargs)
self.norm2 = LayerNorm(d_model, eps=layer_norm_eps, **factory_kwargs)
self.dropout1 = Dropout(dropout)
self.dropout2 = Dropout(dropout)
self.pre_norm = pre_norm
self.recompute_attn = recompute_attn
self.save_trainingset_representations = save_trainingset_representations
self.saved_src_to_attend_to = None
self.activation = _get_activation_fn(activation)
def __setstate__(self, state):
if "activation" not in state:
state["activation"] = F.relu
super().__setstate__(state)
self.__dict__.setdefault("save_trainingset_representations", False)
def forward(
self,
src: Tensor,
src_mask: Optional[Tensor] = None,
src_key_padding_mask: Optional[Tensor] = None,
) -> Tensor:
r"""Pass the input through the encoder layer.
Args:
src: the sequence to the encoder layer (required).
src_mask: the mask for the src sequence (optional).
src_key_padding_mask: the mask for the src keys per batch (optional).
Shape:
see the docs in Transformer class.
"""
if self.save_trainingset_representations:
assert (
isinstance(src_mask, int) and not self.training
), "save_trainingset_representations is only supported in eval mode and requires src_mask to be an int"
if self.pre_norm:
src_ = self.norm1(src)
else:
src_ = src
if isinstance(src_mask, tuple):
# global attention setup
assert not self.self_attn.batch_first
assert src_key_padding_mask is None
global_src_mask, trainset_src_mask, valset_src_mask = src_mask
num_global_tokens = global_src_mask.shape[0]
num_train_tokens = trainset_src_mask.shape[0]
global_tokens_src = src_[:num_global_tokens]
train_tokens_src = src_[
num_global_tokens : num_global_tokens + num_train_tokens
]
global_and_train_tokens_src = src_[: num_global_tokens + num_train_tokens]
eval_tokens_src = src_[num_global_tokens + num_train_tokens :]
attn = (
partial(checkpoint, self.self_attn)
if self.recompute_attn
else self.self_attn
)
global_tokens_src2 = attn(
global_tokens_src,
global_and_train_tokens_src,
global_and_train_tokens_src,
None,
True,
global_src_mask,
)[0]
train_tokens_src2 = attn(
train_tokens_src,
global_tokens_src,
global_tokens_src,
None,
True,
trainset_src_mask,
)[0]
eval_tokens_src2 = attn(
eval_tokens_src, src_, src_, None, True, valset_src_mask
)[0]
src2 = torch.cat(
[global_tokens_src2, train_tokens_src2, eval_tokens_src2], dim=0
)
elif isinstance(src_mask, int):
assert src_key_padding_mask is None
single_eval_position = src_mask
src_to_attend_to = src_[:single_eval_position]
if self.save_trainingset_representations:
if (
single_eval_position == src_.shape[0]
or single_eval_position is None
):
self.saved_src_to_attend_to = src_to_attend_to
elif single_eval_position == 0:
if self.saved_src_to_attend_to is None:
raise ValueError(
"First save the trainingset representations by passing in a src_mask of None or the length of the src"
)
src_to_attend_to = self.saved_src_to_attend_to
else:
raise ValueError(
"save_trainingset_representations only supports single_eval_position == 0 or single_eval_position == src.shape[0]"
)
src_left = self.self_attn(
src_[:single_eval_position],
src_[:single_eval_position],
src_[:single_eval_position],
)[0]
src_right = self.self_attn(
src_[single_eval_position:], src_to_attend_to, src_to_attend_to
)[0]
src2 = torch.cat([src_left, src_right], dim=0)
else:
if self.recompute_attn:
src2 = checkpoint(
self.self_attn,
src_,
src_,
src_,
src_key_padding_mask,
True,
src_mask,
)[0]
else:
src2 = self.self_attn(
src_,
src_,
src_,
attn_mask=src_mask,
key_padding_mask=src_key_padding_mask,
)[0]
src = src + self.dropout1(src2)
if not self.pre_norm:
src = self.norm1(src)
if self.pre_norm:
src_ = self.norm2(src)
else:
src_ = src
src2 = self.linear2(self.dropout(self.activation(self.linear1(src_))))
src = src + self.dropout2(src2)
if not self.pre_norm:
src = self.norm2(src)
return src