shahrukhx01
commited on
Commit
•
4e98941
1
Parent(s):
254528a
add model files
Browse files- multitask_model.py +144 -0
- test.py +0 -0
multitask_model.py
ADDED
@@ -0,0 +1,144 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
Implementation borrowed from transformers package and extended to support multiple prediction heads:
|
3 |
+
|
4 |
+
https://github.com/huggingface/transformers/blob/master/src/transformers/models/bert/modeling_bert.py
|
5 |
+
"""
|
6 |
+
|
7 |
+
import torch
|
8 |
+
from torch import nn
|
9 |
+
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
|
10 |
+
|
11 |
+
import transformers
|
12 |
+
from transformers import BertTokenizer
|
13 |
+
from transformers import models
|
14 |
+
from transformers.modeling_outputs import SequenceClassifierOutput
|
15 |
+
|
16 |
+
from transformers.models.bert.configuration_bert import BertConfig
|
17 |
+
from transformers.models.bert.modeling_bert import (
|
18 |
+
BertPreTrainedModel,
|
19 |
+
BERT_INPUTS_DOCSTRING,
|
20 |
+
_TOKENIZER_FOR_DOC,
|
21 |
+
_CHECKPOINT_FOR_DOC,
|
22 |
+
_CONFIG_FOR_DOC,
|
23 |
+
BertModel,
|
24 |
+
)
|
25 |
+
|
26 |
+
from transformers.file_utils import (
|
27 |
+
add_code_sample_docstrings,
|
28 |
+
add_start_docstrings_to_model_forward,
|
29 |
+
)
|
30 |
+
|
31 |
+
|
32 |
+
class BertForSequenceClassification(BertPreTrainedModel):
|
33 |
+
def __init__(self, config, **kwargs):
|
34 |
+
super().__init__(transformers.PretrainedConfig())
|
35 |
+
self.num_labels = kwargs.get("task_labels_map", {})
|
36 |
+
self.config = config
|
37 |
+
|
38 |
+
self.bert = BertModel(config)
|
39 |
+
classifier_dropout = (
|
40 |
+
config.classifier_dropout
|
41 |
+
if config.classifier_dropout is not None
|
42 |
+
else config.hidden_dropout_prob
|
43 |
+
)
|
44 |
+
self.dropout = nn.Dropout(classifier_dropout)
|
45 |
+
## add task specific output heads
|
46 |
+
self.classifier1 = nn.Linear(
|
47 |
+
config.hidden_size, list(self.num_labels.values())[0]
|
48 |
+
)
|
49 |
+
self.classifier2 = nn.Linear(
|
50 |
+
config.hidden_size, list(self.num_labels.values())[1]
|
51 |
+
)
|
52 |
+
|
53 |
+
self.init_weights()
|
54 |
+
|
55 |
+
@add_start_docstrings_to_model_forward(
|
56 |
+
BERT_INPUTS_DOCSTRING.format("batch_size, sequence_length")
|
57 |
+
)
|
58 |
+
@add_code_sample_docstrings(
|
59 |
+
tokenizer_class=_TOKENIZER_FOR_DOC,
|
60 |
+
checkpoint=_CHECKPOINT_FOR_DOC,
|
61 |
+
output_type=SequenceClassifierOutput,
|
62 |
+
config_class=_CONFIG_FOR_DOC,
|
63 |
+
)
|
64 |
+
def forward(
|
65 |
+
self,
|
66 |
+
input_ids=None,
|
67 |
+
attention_mask=None,
|
68 |
+
token_type_ids=None,
|
69 |
+
position_ids=None,
|
70 |
+
head_mask=None,
|
71 |
+
inputs_embeds=None,
|
72 |
+
labels=None,
|
73 |
+
output_attentions=None,
|
74 |
+
output_hidden_states=None,
|
75 |
+
return_dict=None,
|
76 |
+
task_name=None,
|
77 |
+
):
|
78 |
+
r"""
|
79 |
+
labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`):
|
80 |
+
Labels for computing the sequence classification/regression loss. Indices should be in :obj:`[0, ...,
|
81 |
+
config.num_labels - 1]`. If :obj:`config.num_labels == 1` a regression loss is computed (Mean-Square loss),
|
82 |
+
If :obj:`config.num_labels > 1` a classification loss is computed (Cross-Entropy).
|
83 |
+
"""
|
84 |
+
return_dict = (
|
85 |
+
return_dict if return_dict is not None else self.config.use_return_dict
|
86 |
+
)
|
87 |
+
|
88 |
+
outputs = self.bert(
|
89 |
+
input_ids,
|
90 |
+
attention_mask=attention_mask,
|
91 |
+
token_type_ids=token_type_ids,
|
92 |
+
position_ids=position_ids,
|
93 |
+
head_mask=head_mask,
|
94 |
+
inputs_embeds=inputs_embeds,
|
95 |
+
output_attentions=output_attentions,
|
96 |
+
output_hidden_states=output_hidden_states,
|
97 |
+
return_dict=return_dict,
|
98 |
+
)
|
99 |
+
|
100 |
+
pooled_output = outputs[1]
|
101 |
+
|
102 |
+
pooled_output = self.dropout(pooled_output)
|
103 |
+
logits = None
|
104 |
+
if task_name == list(self.num_labels.keys())[0]:
|
105 |
+
logits = self.classifier1(pooled_output)
|
106 |
+
elif task_name == list(self.num_labels.keys())[1]:
|
107 |
+
logits = self.classifier2(pooled_output)
|
108 |
+
|
109 |
+
loss = None
|
110 |
+
if labels is not None:
|
111 |
+
if self.config.problem_type is None:
|
112 |
+
if self.num_labels[task_name] == 1:
|
113 |
+
self.config.problem_type = "regression"
|
114 |
+
elif self.num_labels[task_name] > 1 and (
|
115 |
+
labels.dtype == torch.long or labels.dtype == torch.int
|
116 |
+
):
|
117 |
+
self.config.problem_type = "single_label_classification"
|
118 |
+
else:
|
119 |
+
self.config.problem_type = "multi_label_classification"
|
120 |
+
|
121 |
+
if self.config.problem_type == "regression":
|
122 |
+
loss_fct = MSELoss()
|
123 |
+
if self.num_labels[task_name] == 1:
|
124 |
+
loss = loss_fct(logits.squeeze(), labels.squeeze())
|
125 |
+
else:
|
126 |
+
loss = loss_fct(logits, labels)
|
127 |
+
elif self.config.problem_type == "single_label_classification":
|
128 |
+
loss_fct = CrossEntropyLoss()
|
129 |
+
loss = loss_fct(
|
130 |
+
logits.view(-1, self.num_labels[task_name]), labels.view(-1)
|
131 |
+
)
|
132 |
+
elif self.config.problem_type == "multi_label_classification":
|
133 |
+
loss_fct = BCEWithLogitsLoss()
|
134 |
+
loss = loss_fct(logits, labels)
|
135 |
+
if not return_dict:
|
136 |
+
output = (logits,) + outputs[2:]
|
137 |
+
return ((loss,) + output) if loss is not None else output
|
138 |
+
|
139 |
+
return SequenceClassifierOutput(
|
140 |
+
loss=loss,
|
141 |
+
logits=logits,
|
142 |
+
hidden_states=outputs.hidden_states,
|
143 |
+
attentions=outputs.attentions,
|
144 |
+
)
|
test.py
DELETED
File without changes
|