-
Notifications
You must be signed in to change notification settings - Fork 3
/
model.py
163 lines (124 loc) · 5.48 KB
/
model.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
import math
import torch
import torch.nn.functional as F
from torch import nn
class PositionalEncoding(nn.Module):
def __init__(self, d_model, dropout=0.1, max_len=5000):
super(PositionalEncoding, self).__init__()
self.dropout = nn.Dropout(p=dropout)
position = torch.arange(max_len).unsqueeze(1)
div_term = torch.exp(torch.arange(
0, d_model, 2).float() * (-math.log(10000.0) / d_model))
pe = torch.zeros(max_len, 1, d_model)
pe[:, 0, 0::2] = torch.sin(position * div_term)
pe[:, 0, 1::2] = torch.cos(position * div_term)
# pe = pe.unsqueeze(0).transpose(0, 1)
pe = pe.transpose(0, 1)
self.register_buffer('pe', pe)
def forward(self, x):
x = x + self.pe[:, :x.size(1)]
return self.dropout(x)
class LearnedPositionEncoding(nn.Embedding):
def __init__(self, d_model, dropout=0.1, max_len=100):
super().__init__(max_len, d_model)
self.dropout = nn.Dropout(p=dropout)
def forward(self, x):
weight = self.weight.data.unsqueeze(0)
x = x + weight
return self.dropout(x)
class Adapter(nn.Module):
def __init__(self, in_dim, out_dim, hidden_dim=64):
super().__init__()
self.linear1 = nn.Linear(in_dim, hidden_dim)
self.linear2 = nn.Linear(hidden_dim, out_dim)
self.gelu = nn.GELU()
def forward(self, x):
z = self.gelu(self.linear1(x))
z = self.linear2(z)
return x+z
class TransformerEncoderLayer(nn.Module):
__constants__ = ['batch_first']
def __init__(self, d_model, nhead, adapter_size=64, dim_feedforward=3072, dropout=0.1, activation="relu",
layer_norm_eps=1e-5, batch_first=False,
device=None, dtype=None) -> None:
super().__init__()
factory_kwargs = {'device': device, 'dtype': dtype}
self.self_attn = nn.MultiheadAttention(d_model, nhead, dropout=dropout, batch_first=batch_first,
**factory_kwargs)
self.dropout1 = nn.Dropout(dropout)
self.adapter1 = Adapter(d_model, d_model, hidden_dim=adapter_size)
self.norm1 = nn.LayerNorm(d_model, eps=layer_norm_eps)
self.linear1 = nn.Linear(d_model, dim_feedforward)
self.activation = nn.GELU()
self.linear2 = nn.Linear(dim_feedforward, d_model)
self.dropout = nn.Dropout(dropout)
self.dropout2 = nn.Dropout(dropout)
self.adapter2 = Adapter(d_model, d_model, hidden_dim=adapter_size)
self.norm2 = nn.LayerNorm(d_model, eps=layer_norm_eps)
def __setstate__(self, state):
if 'activation' not in state:
state['activation'] = F.relu
super(TransformerEncoderLayer, self).__setstate__(state)
def forward(self, src, src_mask=None, src_key_padding_mask=None):
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.
"""
src2 = self.self_attn(src, src, src, attn_mask=src_mask,
key_padding_mask=src_key_padding_mask)[0]
src2 = self.dropout1(src2)
src2 = self.adapter1(src2)
src = self.norm1(src + src2)
src2 = self.linear2(self.dropout(self.activation(self.linear1(src))))
src2 = self.dropout2(src2)
src2 = self.adapter2(src2)
src = self.norm2(src + src2)
return src
def activate_adapter(self):
tune_layers = [self.adapter1, self.adapter2, self.norm1, self.norm2]
for layer in tune_layers:
for param in layer.parameters():
param.requires_grad = True
class Model(nn.Module):
def __init__(self, mode, num_layers=4, adapter_size=64, dim=768, window_size=100, nhead=8, dim_feedforward=3072, dropout=0.1):
super(Model, self).__init__()
if mode == 'adapter':
encoder_layer = TransformerEncoderLayer(
dim, nhead, adapter_size=adapter_size, dim_feedforward=dim_feedforward, dropout=dropout, batch_first=True)
else:
encoder_layer = nn.TransformerEncoderLayer(
dim, nhead, dim_feedforward, dropout, batch_first=True)
self.trans_encder = nn.TransformerEncoder(
encoder_layer=encoder_layer, num_layers=num_layers)
self.pos_encoder1 = PositionalEncoding(d_model=768)
# self.pos_encoder2 = LearnedPositionEncoding(
# d_model=768, max_len=window_size)
self.fc1 = nn.Linear(dim * window_size, 2)
def forward(self, x):
B, _, _ = x.size()
# x = x*math.sqrt(self.dim)
x = self.pos_encoder1(x)
# x = self.pos_encoder2(x)
x = self.trans_encder(x) # mask默认None
x = x.contiguous().view(B, -1)
x = self.fc1(x)
return x
def train_adapter(self):
for param in self.parameters():
param.requires_grad = False
for layer in self.trans_encder.layers:
layer.activate_adapter()
for param in self.fc1.parameters():
param.requires_grad = True
def train_classifier(self):
for param in self.parameters():
param.requires_grad = False
for param in self.fc1.parameters():
param.requires_grad = True
if __name__ == '__main__':
model = Model('adapter')
pass