-
Notifications
You must be signed in to change notification settings - Fork 2
/
train_setfit.py
162 lines (134 loc) · 3.91 KB
/
train_setfit.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
import argparse
import logging
from datasets import Dataset
import pandas as pd
from setfit import SetFitModel, SetFitTrainer
from sklearn.linear_model import LogisticRegression
from sentence_transformers import SentenceTransformer, models, LoggingHandler
logging.basicConfig(format='%(asctime)s - %(message)s',
datefmt='%Y-%m-%d %H:%M:%S',
level=logging.INFO,
handlers=[LoggingHandler()])
parser = argparse.ArgumentParser(description="Sentence embedding fine-tuning (SetFit) for few-shot classification")
parser.add_argument(
"--model_name_or_path",
type=str,
default=None,
required=True,
help="The name or path of DAPT-ed Transformer model to use.",
)
parser.add_argument(
"--adapter_path",
type=str,
default=None,
help="Path of the SEPT adapter.",
)
parser.add_argument(
"--adapter_name",
type=str,
default=None,
help="Name of the SEPT adapter.",
)
parser.add_argument(
"--train_dataset_path",
type=str,
default=None,
required=True,
help="Path of a local train .csv file.",
)
parser.add_argument(
"--eval_dataset_path",
type=str,
default=None,
required=True,
help="Path of a local eval .csv file",
)
parser.add_argument(
"--batch_size",
type=int,
default=16,
help="Batch size for contrastive fine-tuning.",
)
parser.add_argument(
"--num_epochs",
type=int,
default=1,
help="Number of training epochs.",
)
parser.add_argument(
"--num_iterations",
type=int,
default=20,
help="The number of text pairs to generate for each sentence for the contrastive learning.",
)
parser.add_argument(
"--num_samples",
type=int,
default=8,
help="Number of labeled samples per class.",
)
parser.add_argument(
"--text_col",
type=str,
default='text',
help="Name of the text column in the dataset.",
)
parser.add_argument(
"--label_col",
type=str,
default='label',
help="Name of the label column in the dataset.",
)
parser.add_argument(
"--model_save_path",
type=str,
default=None,
help="Save path of the model",
)
args = parser.parse_args()
num_samples = args.num_samples
num_iterations = args.num_iterations
num_epochs = args.num_epochs
batch_size = args.batch_size
model_path = args.model_name_or_path
adapter_path = args.adapter_path
adapter_name = args.adapter_name
model_save_path = args.model_save_path
text_col = args.text_col
label_col = args.label_col
# Load datasets
df_train = pd.read_csv(args.train_dataset_path)
df_eval = pd.read_csv(args.eval_dataset_path)
train_dataset = Dataset.from_pandas(df_train)
eval_dataset = Dataset.from_pandas(df_eval)
# Create Sentence Transformer
if adapter_path:
word_embedding_model = models.Transformer(model_path)
# Load and activate adapter
word_embedding_model.auto_model.load_adapter(args.adapter_path)
word_embedding_model.auto_model.set_active_adapters(args.adapter_name)
# Turn on gradient update for adapter parameters
word_embedding_model.auto_model.train_adapter(args.adapter_name)
# Turn on gradient update for Transformer parameters
word_embedding_model.auto_model.freeze_model(False)
pooling_model = models.Pooling(word_embedding_model.get_word_embedding_dimension(), pooling_mode='mean')
model = SentenceTransformer(modules=[word_embedding_model, pooling_model])
else:
model = SentenceTransformer(model_path)
# SetFit
setfit_model = SetFitModel(model_body=model,
model_head=LogisticRegression())
trainer = SetFitTrainer(
model=setfit_model,
train_dataset=train_dataset,
eval_dataset=eval_dataset,
batch_size=batch_size,
num_epochs=num_epochs,
num_iterations=num_iterations,
column_mapping={text_col: "text", label_col: "label"},
)
trainer.train()
metrics = trainer.evaluate()
print(metrics)
if model_save_path:
setfit_model._save_pretrained(model_save_path)