-
Notifications
You must be signed in to change notification settings - Fork 2
/
model.py
182 lines (142 loc) · 6.42 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
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
from torch import nn
from typing import AnyStr, List
from transformers import AutoConfig
from transformers import AutoModel
class AutoTransformerMultiTask(nn.Module):
"""
Implements a transformer with multiple classifier heads for multi-task training
"""
def __init__(self, transformer_model: AnyStr, task_num_labels: List):
super(AutoTransformerMultiTask, self).__init__()
config = AutoConfig.from_pretrained(transformer_model)
self.config = config
self.xformer = AutoModel.from_pretrained(transformer_model, config=config)
# Pooling layers
self.pooling = nn.ModuleList([nn.Linear(config.hidden_size, config.hidden_size) for _ in task_num_labels])
self.act = nn.Tanh()
# Create the classifier heads
self.task_classifiers = nn.ModuleList([nn.Linear(config.hidden_size, n_labels) for n_labels in task_num_labels])
self.task_num_labels = task_num_labels
self.dropout = nn.Dropout(config.hidden_dropout_prob)
def forward(
self,
input_ids=None,
attention_mask=None,
labels=None,
task_num=0,
lam=1.0,
logits_mask=None
):
outputs = self.xformer(
input_ids,
attention_mask=attention_mask
)
sequence_output = outputs[0]
if len(sequence_output.shape) == 3:
sequence_output = sequence_output[:,0]
assert sequence_output.shape[0] == input_ids.shape[0]
assert sequence_output.shape[1] == self.config.hidden_size
pooled_output = self.pooling[task_num](sequence_output)
pooled_output = self.dropout(self.act(pooled_output))
logits = self.task_classifiers[task_num](pooled_output)
outputs = (logits,)
if labels is not None:
loss_fct = nn.CrossEntropyLoss()
loss = lam * loss_fct(logits.view(-1, self.task_num_labels[task_num]), labels.view(-1))
outputs = (loss,) + outputs
return outputs
class AutoTransformerForSentenceSequenceModeling(nn.Module):
"""
Implements a transformer which performs sequence classification on a sequence of sentences
"""
def __init__(self, transformer_model: AnyStr, num_labels: int = 2, sep_token_id: int = 2):
super(AutoTransformerForSentenceSequenceModeling, self).__init__()
config = AutoConfig.from_pretrained(transformer_model)
self.config = config
self.xformer = AutoModel.from_pretrained(transformer_model, config=config)
# Pooling layers
self.pooling = nn.Linear(config.hidden_size, config.hidden_size)
self.act = nn.Tanh()
# Create the classifier heads
self.classifier = nn.Linear(config.hidden_size, num_labels)
self.num_labels = num_labels
self.dropout = nn.Dropout(config.hidden_dropout_prob)
self.sep_token_id = sep_token_id
def forward(
self,
input_ids=None,
attention_mask=None,
labels=None,
lam=1.0
):
outputs = self.xformer(
input_ids,
attention_mask=attention_mask,
)
# Gather all of the SEP hidden states VERIFY THIS IS CORRECT!
hidden_states = outputs[0].reshape(-1, self.config.hidden_size)
locs = (input_ids == self.sep_token_id).view(-1)
#(n * seq_len x d) -> (n * sep_len x d)
sequence_output = hidden_states[locs]
assert sequence_output.shape[0] == sum(locs)
assert sequence_output.shape[1] == self.config.hidden_size
pooled_output = self.pooling(sequence_output)
pooled_output = self.dropout(self.act(pooled_output))
logits = self.classifier(pooled_output)
outputs = (logits,)
if labels is not None:
loss_fct = nn.CrossEntropyLoss()
loss = lam * loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
outputs = (loss,) + outputs
return outputs
class AutoTransformerForSentenceSequenceModelingMultiTask(nn.Module):
"""
Implements a transformer which performs sequence classification on a sequence of sentences
"""
def __init__(self, transformer_model: AnyStr, task_num_labels: List, sep_token_id: int = 2):
super(AutoTransformerForSentenceSequenceModelingMultiTask, self).__init__()
config = AutoConfig.from_pretrained(transformer_model)
self.config = config
self.xformer = AutoModel.from_pretrained(transformer_model, config=config)
# Pooling layers
self.pooling = nn.ModuleList([nn.Linear(config.hidden_size, config.hidden_size) for _ in task_num_labels])
self.act = nn.Tanh()
# Create the classifier heads
self.task_classifiers = nn.ModuleList([nn.Linear(config.hidden_size, n_labels) for n_labels in task_num_labels])
self.task_num_labels = task_num_labels
self.dropout = nn.Dropout(config.hidden_dropout_prob)
self.sep_token_id = sep_token_id
def forward(
self,
input_ids=None,
attention_mask=None,
labels=None,
logits_mask=None,
task_num=0,
lam=1.0
):
outputs = self.xformer(
input_ids,
attention_mask=attention_mask,
)
# Gather all of the SEP hidden states VERIFY THIS IS CORRECT!
#hidden_states = outputs[0].reshape(-1, self.config.hidden_size)
# locs = (input_ids == self.sep_token_id).view(-1)
# # (n * seq_len x d) -> (n * sep_len x d)
# sequence_output = hidden_states[locs]
if logits_mask is None:
sequence_output = outputs[0][:,0,:].reshape(-1, self.config.hidden_size)
else:
sequence_output = outputs[0][logits_mask == 1].reshape(-1, self.config.hidden_size)
assert sequence_output.shape[0] == logits_mask.sum().item()
assert sequence_output.shape[1] == self.config.hidden_size
pooled_output = self.pooling[task_num](sequence_output)
pooled_output = self.dropout(self.act(pooled_output))
logits = self.task_classifiers[task_num](pooled_output)
outputs = (logits,)
if labels is not None:
assert sequence_output.shape[0] == labels.shape[0]
loss_fct = nn.CrossEntropyLoss()
loss = lam * loss_fct(logits.view(-1, self.task_num_labels[task_num]), labels.view(-1))
outputs = (loss,) + outputs
return outputs