-
Notifications
You must be signed in to change notification settings - Fork 21
/
transformer.py
304 lines (212 loc) · 8.71 KB
/
transformer.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
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
import math
import torch
import torch.nn as nn
import torch.nn.functional as f
class PositionalEncoding(nn.Module):
"""
https://pytorch.org/tutorials/beginner/transformer_tutorial.html
"""
def __init__(self, model_size, vocab_size=5000, dropout=0.1):
super().__init__()
self.dropout = nn.Dropout(p=dropout)
pe = torch.zeros(vocab_size, model_size)
position = torch.arange(0, vocab_size, dtype=torch.float).unsqueeze(1)
div_term = torch.exp(
torch.arange(0, model_size, 2).float()
* (-math.log(10000.0) / model_size)
)
pe[:, 0::2] = torch.sin(position * div_term)
pe[:, 1::2] = torch.cos(position * div_term)
pe = pe.unsqueeze(0)
self.register_buffer("pe", pe)
def forward(self, x):
# Compute positional encoding
pos_enc = self.pe[:, : x.size(1), :]
# Add positional encoding to input vector (typicall a word embedding vector)
x = x + pos_enc
# Here only return the positional encoding to inspect it (not needed in practice)
return self.dropout(x), pos_enc
class Attention(nn.Module):
### Implements Scaled Dot Product Attention
def __init__(self):
super().__init__()
def forward(self, Q, K, V, mask=None, dropout=None):
# All shapes: (batch_size, seq_len, hidden_size)
# Perform Q*K^T (* is the dot product here)
# We have to use torch.matmul since we work with batches!
out = torch.matmul(Q, K.transpose(1, 2)) # => shape: (B, L, L)
# Divide by scaling factor
out = out / (Q.shape[-1] ** 0.5)
#if mask is not None:
# NOT IMPLEMENTED YET
# Push throught softmax layer
out = f.softmax(out, dim=-1)
# Optional: Dropout
if dropout is not None:
out = nn.Dropout(out, dropout)
# Multiply with values V
out = torch.matmul(out, V)
return out
class AttentionHead(nn.Module):
def __init__(self, model_size, qkv_size):
super().__init__()
self.Wq = nn.Linear(model_size, qkv_size)
self.Wk = nn.Linear(model_size, qkv_size)
self.Wv = nn.Linear(model_size, qkv_size)
self.attention = Attention()
self._init_parameters()
def _init_parameters(self):
nn.init.xavier_uniform_(self.Wq.weight)
nn.init.xavier_uniform_(self.Wk.weight)
nn.init.xavier_uniform_(self.Wv.weight)
def forward(self, query, key, value):
return self.attention(self.Wq(query), self.Wk(key), self.Wv(value))
class MultiHeadAttention(nn.Module):
def __init__(self, model_size, num_heads):
super().__init__()
if model_size % num_heads != 0:
raise Exception("The model size must be divisible by the number of heads!")
# Define sizes of Q/K/V based on model size and number of heads
self.qkv_size = model_size // num_heads
self.heads = nn.ModuleList(
[AttentionHead(model_size, self.qkv_size) for _ in range(num_heads)]
)
# Linear layer to "unify" all heads into one
self.Wo = nn.Linear(model_size, model_size)
# Initalize parameters of output layer
self._init_parameters()
def _init_parameters(self):
nn.init.xavier_uniform_(self.Wo.weight)
def forward(self, query, key, value):
# Push Q, K, V through all the Attention Heads
out_heads = tuple([ attention_head(query, key, value) for attention_head in self.heads ])
# Concatenate the outputs of all Attention Heads
out = torch.cat(out_heads, dim=-1)
# Push concatenated outputs through last layers => output size is model_size again
return self.Wo(out)
class FeedForward(nn.Module):
def __init__(self, model_size, hidden_size=2048):
super().__init__()
# Define basic Feed Forward Network as proposed in the original paper
self.net = nn.Sequential(
nn.Linear(model_size, hidden_size),
nn.ReLU(),
nn.Linear(hidden_size, model_size),
)
def forward(self, X):
return self.net(X)
class TransformerEncoderLayer(nn.Module):
def __init__(self, model_size, num_heads, ff_hidden_size, dropout):
super().__init__()
# MultiHeadAttention block
self.mha1 = MultiHeadAttention(model_size, num_heads)
self.dropout1 = nn.Dropout(dropout)
self.norm1 = nn.LayerNorm(model_size)
# FeedForward block
self.ff = FeedForward(model_size, ff_hidden_size)
self.dropout2 = nn.Dropout(dropout)
self.norm2 = nn.LayerNorm(model_size)
def forward(self, source):
# MultiHeadAttentionBlock
out1 = self.mha1(source, source, source)
out1 = self.dropout1(out1)
out1 = self.norm1(out1 + source)
# FeedForward block
out2 = self.ff(out1)
out2 = self.dropout2(out2)
out2 = self.norm2(out2 + out1)
# Return final output
return out2
class TransformerEncoder(nn.Module):
def __init__(self,
num_layers=6,
model_size=512,
num_heads=8,
ff_hidden_size=2048,
dropout= 0.1):
super().__init__()
# Define num_layers (N) encoder layers
self.layers = nn.ModuleList(
[ TransformerEncoderLayer(model_size, num_heads, ff_hidden_size, dropout) for _ in range(num_layers) ]
)
def forward(self, source):
for l in self.layers:
source = l(source)
return source
##
## Decoder
##
class TransformerDecoderLayer(nn.Module):
def __init__(self, model_size, num_heads, ff_hidden_size, dropout):
super().__init__()
# 1st MultiHeadAttention block (decoder input only)
self.mha1 = MultiHeadAttention(model_size, num_heads)
self.dropout1 = nn.Dropout(dropout)
self.norm1 = nn.LayerNorm(model_size)
# 2nd MultiHeadAttention block (encoder & decoder)
self.mha2 = MultiHeadAttention(model_size, num_heads)
self.dropout2 = nn.Dropout(dropout)
self.norm2 = nn.LayerNorm(model_size)
self.ff = FeedForward(model_size, ff_hidden_size)
self.dropout3 = nn.Dropout(dropout)
self.norm3 = nn.LayerNorm(model_size)
def forward(self, target, memory):
# 1st MultiHeadAttention block
out1 = self.mha1(target, target, target)
out1 = self.dropout1(out1)
out1 = self.norm1(out1 + target)
# 2nd MultiHeadAttention block
out2 = self.mha2(out1, memory, memory)
out2 = self.dropout2(out2)
out2 = self.norm2(out2 + out1)
# FeedForward block
out3 = self.ff(out2)
out3 = self.dropout3(out3)
out3 = self.norm3(out3 + out2)
# Return final output
return out3
class TransformerDecoder(nn.Module):
def __init__(self,
num_layers=6,
model_size=512,
num_heads=8,
ff_hidden_size=2048,
dropout= 0.1):
super().__init__()
# Define num_layers (N) decoder layers
self.layers = nn.ModuleList(
[ TransformerDecoderLayer(model_size, num_heads, ff_hidden_size, dropout) for _ in range(num_layers) ]
)
def forward(self, target, memory):
# Push through each decoder layer
for l in self.layers:
target = l(target, memory)
return target
class Transformer(nn.Module):
def __init__(self,
num_encoder_layers=6,
num_decoder_layers=6,
model_size=512,
num_heads=8,
ff_hidden_size=2048,
dropout= 0.1):
super().__init__()
# Definer encoder
self.encoder = TransformerEncoder(
num_layers=num_encoder_layers,
model_size=model_size,
num_heads=num_heads,
ff_hidden_size=ff_hidden_size,
dropout=dropout
)
#Define decoder
self.decoder = TransformerDecoder(
num_layers=num_decoder_layers,
model_size=model_size,
num_heads=num_heads,
ff_hidden_size=ff_hidden_size,
dropout=dropout
)
def forward(self, source, target):
memory = self.encoder(source)
return self.decoder(target, memory)