-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathparabart.py
275 lines (199 loc) · 9.5 KB
/
parabart.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
import torch
import torch.nn as nn
import torch.nn.functional as F
from transformers.modeling_bart import (
PretrainedBartModel,
LayerNorm,
EncoderLayer,
DecoderLayer,
LearnedPositionalEmbedding,
_prepare_bart_decoder_inputs,
_make_linear_from_emb
)
class ParaBart(PretrainedBartModel):
def __init__(self, config):
super().__init__(config)
self.shared = nn.Embedding(config.vocab_size, config.d_model, config.pad_token_id)
self.encoder = ParaBartEncoder(config, self.shared)
self.decoder = ParaBartDecoder(config, self.shared)
self.linear = nn.Linear(config.d_model, config.vocab_size)
self.adversary = Discriminator(config)
self.init_weights()
def forward(
self,
input_ids,
decoder_input_ids,
attention_mask=None,
decoder_padding_mask=None,
encoder_outputs=None,
return_encoder_outputs=False,
):
if attention_mask is None:
attention_mask = input_ids == self.config.pad_token_id
if encoder_outputs is None:
encoder_outputs = self.encoder(input_ids, attention_mask=attention_mask)
if return_encoder_outputs:
return encoder_outputs
assert encoder_outputs is not None
assert decoder_input_ids is not None
decoder_input_ids = decoder_input_ids[:, :-1]
_, decoder_padding_mask, decoder_causal_mask = _prepare_bart_decoder_inputs(
self.config,
input_ids=None,
decoder_input_ids=decoder_input_ids,
decoder_padding_mask=decoder_padding_mask,
causal_mask_dtype=self.shared.weight.dtype,
)
attention_mask2 = torch.cat((torch.zeros(input_ids.shape[0], 1).bool().cuda(), attention_mask[:, self.config.max_sent_len+2:]), dim=1)
# decoder
decoder_outputs = self.decoder(
decoder_input_ids,
torch.cat((encoder_outputs[1], encoder_outputs[0][:, self.config.max_sent_len+2:]), dim=1),
decoder_padding_mask=decoder_padding_mask,
decoder_causal_mask=decoder_causal_mask,
encoder_attention_mask=attention_mask2,
)[0]
batch_size = decoder_outputs.shape[0]
outputs = self.linear(decoder_outputs.contiguous().view(-1, self.config.d_model))
outputs = outputs.view(batch_size, -1, self.config.vocab_size)
# discriminator
for p in self.adversary.parameters():
p.required_grad=False
adv_outputs = self.adversary(encoder_outputs[1])
return outputs, adv_outputs
def prepare_inputs_for_generation(self, decoder_input_ids, past, attention_mask, use_cache, **kwargs):
assert past is not None, "past has to be defined for encoder_outputs"
encoder_outputs = past[0]
return {
"input_ids": None, # encoder_outputs is defined. input_ids not needed
"encoder_outputs": encoder_outputs,
"decoder_input_ids": torch.cat((decoder_input_ids, torch.zeros((decoder_input_ids.shape[0], 1), dtype=torch.long).cuda()), 1),
"attention_mask": attention_mask,
}
def get_encoder(self):
return self.encoder
def get_output_embeddings(self):
return _make_linear_from_emb(self.shared)
def get_input_embeddings(self):
return self.shared
@staticmethod
def _reorder_cache(past, beam_idx):
enc_out = past[0][0]
new_enc_out = enc_out.index_select(0, beam_idx)
past = ((new_enc_out, ), )
return past
def forward_adv(
self,
input_token_ids,
attention_mask=None,
decoder_padding_mask=None
):
for p in self.adversary.parameters():
p.required_grad=True
sent_embeds = self.encoder.embed(input_token_ids, attention_mask=attention_mask).detach()
adv_outputs = self.adversary(sent_embeds)
return adv_outputs
class ParaBartEncoder(nn.Module):
def __init__(self, config, embed_tokens):
super().__init__()
self.config = config
self.dropout = config.dropout
self.embed_tokens = embed_tokens
self.embed_synt = nn.Embedding(77, config.d_model, config.pad_token_id)
self.embed_synt.weight.data.normal_(mean=0.0, std=config.init_std)
self.embed_synt.weight.data[config.pad_token_id].zero_()
self.embed_positions = LearnedPositionalEmbedding(
config.max_position_embeddings, config.d_model, config.pad_token_id, config.extra_pos_embeddings
)
self.layers = nn.ModuleList([EncoderLayer(config) for _ in range(config.encoder_layers)])
self.synt_layers = nn.ModuleList([EncoderLayer(config) for _ in range(1)])
self.layernorm_embedding = LayerNorm(config.d_model)
self.synt_layernorm_embedding = LayerNorm(config.d_model)
self.pooling = MeanPooling(config)
def forward(self, input_ids, attention_mask):
input_token_ids, input_synt_ids = torch.split(input_ids, [self.config.max_sent_len+2, self.config.max_synt_len+2], dim=1)
input_token_mask, input_synt_mask = torch.split(attention_mask, [self.config.max_sent_len+2, self.config.max_synt_len+2], dim=1)
x = self.forward_token(input_token_ids, input_token_mask)
y = self.forward_synt(input_synt_ids, input_synt_mask)
encoder_outputs = torch.cat((x,y), dim=1)
sent_embeds = self.pooling(x, input_token_ids)
return encoder_outputs, sent_embeds
def forward_token(self, input_token_ids, attention_mask):
if self.training:
drop_mask = torch.bernoulli(self.config.word_dropout*torch.ones(input_token_ids.shape)).bool().cuda()
input_token_ids = input_token_ids.masked_fill(drop_mask, 50264)
input_token_embeds = self.embed_tokens(input_token_ids) + self.embed_positions(input_token_ids)
x = self.layernorm_embedding(input_token_embeds)
x = F.dropout(x, p=self.dropout, training=self.training)
x = x.transpose(0, 1)
for encoder_layer in self.layers:
x, _ = encoder_layer(x, encoder_padding_mask=attention_mask)
x = x.transpose(0, 1)
return x
def forward_synt(self, input_synt_ids, attention_mask):
input_synt_embeds = self.embed_synt(input_synt_ids) + self.embed_positions(input_synt_ids)
y = self.synt_layernorm_embedding(input_synt_embeds)
y = F.dropout(y, p=self.dropout, training=self.training)
# B x T x C -> T x B x C
y = y.transpose(0, 1)
for encoder_synt_layer in self.synt_layers:
y, _ = encoder_synt_layer(y, encoder_padding_mask=attention_mask)
# T x B x C -> B x T x C
y = y.transpose(0, 1)
return y
def embed(self, input_token_ids, attention_mask=None, pool='mean'):
if attention_mask is None:
attention_mask = input_token_ids == self.config.pad_token_id
x = self.forward_token(input_token_ids, attention_mask)
sent_embeds = self.pooling(x, input_token_ids)
return sent_embeds
class MeanPooling(nn.Module):
def __init__(self, config):
super().__init__()
self.config = config
def forward(self, x, input_token_ids):
mask = input_token_ids != self.config.pad_token_id
mean_mask = mask.float()/mask.float().sum(1, keepdim=True)
x = (x*mean_mask.unsqueeze(2)).sum(1, keepdim=True)
return x
class ParaBartDecoder(nn.Module):
def __init__(self, config, embed_tokens):
super().__init__()
self.dropout = config.dropout
self.embed_tokens = embed_tokens
self.embed_positions = LearnedPositionalEmbedding(
config.max_position_embeddings, config.d_model, config.pad_token_id, config.extra_pos_embeddings
)
self.layers = nn.ModuleList([DecoderLayer(config) for _ in range(1)])
self.layernorm_embedding = LayerNorm(config.d_model)
def forward(
self,
decoder_input_ids,
encoder_hidden_states,
decoder_padding_mask,
decoder_causal_mask,
encoder_attention_mask
):
x = self.embed_tokens(decoder_input_ids) + self.embed_positions(decoder_input_ids)
x = self.layernorm_embedding(x)
x = F.dropout(x, p=self.dropout, training=self.training)
x = x.transpose(0, 1)
encoder_hidden_states = encoder_hidden_states.transpose(0, 1)
for idx, decoder_layer in enumerate(self.layers):
x, _, _ = decoder_layer(
x,
encoder_hidden_states,
encoder_attn_mask=encoder_attention_mask,
decoder_padding_mask=decoder_padding_mask,
causal_mask=decoder_causal_mask)
x = x.transpose(0, 1)
return x,
class Discriminator(nn.Module):
def __init__(self, config):
super().__init__()
self.sent_layernorm_embedding = LayerNorm(config.d_model, elementwise_affine=False)
self.adv = nn.Linear(config.d_model, 74)
def forward(self, sent_embeds):
x = self.sent_layernorm_embedding(sent_embeds).squeeze(1)
x = self.adv(x)
return x