ivykopal commited on
Commit
2f52420
·
verified ·
1 Parent(s): 7fc2004

Upload tasktemplates.py

Browse files
Files changed (1) hide show
  1. tasktemplates.py +104 -0
tasktemplates.py ADDED
@@ -0,0 +1,104 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # download templates from the github repository
2
+ import yaml
3
+ import datasets
4
+
5
+
6
+ _PREFIX = 'https://raw.githubusercontent.com/ivanvykopal/tasktemplates/master/src/templates/'
7
+
8
+ _DATASETS = [
9
+ 'aeslc',
10
+ 'anli',
11
+ 'billsum',
12
+ 'boolq',
13
+ 'c4',
14
+ 'cb',
15
+ 'cnn_dailymail',
16
+ 'cola',
17
+ 'common_gen',
18
+ 'copa',
19
+ 'cosmos_qa',
20
+ 'cxc',
21
+ 'doc_nli',
22
+ 'drop',
23
+ 'gigaword',
24
+ 'hellaswag',
25
+ 'hotpot_qa',
26
+ 'mnli',
27
+ 'mrpc',
28
+ 'mrqa',
29
+ 'multirc',
30
+ 'multi_news',
31
+ 'multi_nli',
32
+ 'newsqa',
33
+ 'newsroom',
34
+ 'nq_open',
35
+ 'piqa',
36
+ 'qnli',
37
+ 'qqp',
38
+ 'race',
39
+ 'record',
40
+ 'rte',
41
+ 'samsum',
42
+ 'search_qa',
43
+ 'snli',
44
+ 'social_i_qa',
45
+ 'squad',
46
+ 'sst2',
47
+ 'stsb',
48
+ 'wic',
49
+ 'wiki_auto',
50
+ 'wiki_lingua',
51
+ 'winogrande',
52
+ 'wnli',
53
+ 'wsc',
54
+ 'xsum'
55
+ ]
56
+
57
+
58
+ class TaskTemplatesConfig(datasets.BuilderConfig):
59
+ def __init__(self, data_url, **kwargs):
60
+ super(TaskTemplatesConfig, self).__init__(**kwargs)
61
+
62
+ self.data_url = data_url
63
+
64
+
65
+ class TaskTemplates(datasets.GeneratorBasedBuilder):
66
+ VERSION = datasets.Version("1.0.0")
67
+ BUILDER_CONFIGS = [
68
+ TaskTemplatesConfig(
69
+ name=dataset,
70
+ version=datasets.Version("1.0.0"),
71
+ data_url=f"{_PREFIX}{dataset}/templates.yaml"
72
+ ) for dataset in _DATASETS
73
+ ]
74
+ BUILDER_CONFIG_CLASS = TaskTemplatesConfig
75
+
76
+ def _info(self):
77
+ return datasets.DatasetInfo(
78
+ description="Task templates for various NLP tasks",
79
+ features=datasets.Features(
80
+ {
81
+ "task": datasets.Value("string"),
82
+ "templates": datasets.Sequence(datasets.Value("string"))
83
+ }
84
+ ),
85
+ )
86
+
87
+ def _split_generators(self, dl_manager):
88
+ return [
89
+ datasets.SplitGenerator(
90
+ name=datasets.Split.TRAIN,
91
+ gen_kwargs={"data_url": self.config.data_url}
92
+ )
93
+ ]
94
+
95
+ def _generate_examples(self, data_url):
96
+ with open(data_url, 'r') as f:
97
+ data = yaml.safe_load(f)
98
+
99
+ for task, templates in data.items():
100
+ for template in templates:
101
+ yield task, {
102
+ "task": task,
103
+ "templates": template
104
+ }