Repository navigation
Expand file tree
/
Copy pathdata_utils.py
More file actions
130 lines (117 loc) · 4.36 KB
/
Copy pathdata_utils.py
File metadata and controls
130 lines (117 loc) · 4.36 KB
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
from datasets import load_dataset
from transformers import (
AutoTokenizer
)
def load_and_preprocess_data(task, args):
if "mnli" in task:
dataset = load_dataset("glue", "mnli")
else:
dataset = load_dataset("glue", task)
tokenizer = AutoTokenizer.from_pretrained(args.model)
def tokenize_function(examples):
# Handle different input formats
if "premise" in examples and "hypothesis" in examples:
# MNLI and similar tasks
return tokenizer(
examples["premise"],
examples["hypothesis"],
truncation=True,
padding="max_length",
max_length=args.max_seq_length,
)
elif "question" in examples and "sentence" in examples:
# QNLI and similar tasks
return tokenizer(
examples["question"],
examples["sentence"],
truncation=True,
padding="max_length",
max_length=args.max_seq_length,
)
elif "sentence1" in examples and "sentence2" in examples:
# MRPC, STS-B
return tokenizer(
examples["sentence1"],
examples["sentence2"],
truncation=True,
padding="max_length",
max_length=args.max_seq_length,
)
elif "question1" in examples and "question2" in examples:
# QQP
return tokenizer(
examples["question1"],
examples["question2"],
truncation=True,
padding="max_length",
max_length=args.max_seq_length,
)
elif "sentence" in examples:
# CoLA, SST-2
return tokenizer(
examples["sentence"],
truncation=True,
padding="max_length",
max_length=args.max_seq_length,
)
else:
raise ValueError(f"Unexpected format for task {task}")
tokenized_datasets = dataset.map(tokenize_function, batched=True)
if task == "cola":
tokenized_datasets = tokenized_datasets.remove_columns(["sentence", "idx"])
elif task == "sst2":
tokenized_datasets = tokenized_datasets.remove_columns(["sentence", "idx"])
elif task == "mrpc":
tokenized_datasets = tokenized_datasets.remove_columns(
["sentence1", "sentence2", "idx"]
)
elif task == "qqp":
tokenized_datasets = tokenized_datasets.remove_columns(
["question1", "question2", "idx"]
)
elif task == "stsb":
tokenized_datasets = tokenized_datasets.remove_columns(
["sentence1", "sentence2", "idx"]
)
elif task == "qnli":
tokenized_datasets = tokenized_datasets.remove_columns(
["question", "sentence", "idx"]
)
elif task == "rte":
tokenized_datasets = tokenized_datasets.remove_columns(
["sentence1", "sentence2", "idx"]
)
elif task == "wnli":
tokenized_datasets = tokenized_datasets.remove_columns(
["sentence1", "sentence2", "idx"]
)
elif task == "mnli_matched" or task == "mnli_mismatched" or task == "mnli":
tokenized_datasets = tokenized_datasets.remove_columns(
["premise", "hypothesis", "idx"]
)
else:
raise ValueError(f"Unexpected task {task}")
tokenized_datasets = tokenized_datasets.rename_column("label", "labels")
tokenized_datasets.set_format("torch")
if (
task == "cola"
or task == "sst2"
or task == "mrpc"
or task == "qqp"
or task == "stsb"
or task == "qnli"
or task == "rte"
or task == "wnli"
):
train_dataset = tokenized_datasets["train"]
val_dataset = tokenized_datasets["validation"]
test_dataset = tokenized_datasets["test"]
elif task == "mnli_matched":
train_dataset = tokenized_datasets["train"]
val_dataset = tokenized_datasets["validation_matched"]
test_dataset = tokenized_datasets["test_matched"]
elif task == "mnli_mismatched":
train_dataset = tokenized_datasets["train"]
val_dataset = tokenized_datasets["validation_mismatched"]
test_dataset = tokenized_datasets["test_mismatched"]
return train_dataset, val_dataset, test_dataset