Datasets:

Languages:
Vietnamese
ArXiv:
License:
holylovenia commited on
Commit
3a475e9
1 Parent(s): 33f0368

Upload vigetext.py with huggingface_hub

Browse files
Files changed (1) hide show
  1. vigetext.py +151 -0
vigetext.py ADDED
@@ -0,0 +1,151 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from pathlib import Path
2
+
3
+ import datasets
4
+ import pandas as pd
5
+
6
+ from seacrowd.utils import schemas
7
+ from seacrowd.utils.configs import SEACrowdConfig
8
+ from seacrowd.utils.constants import Licenses, Tasks
9
+
10
+ _CITATION = """
11
+ @inproceedings{10.1145/3628797.3628837,
12
+ author = {Nguyen, Duc-Vu and Nguyen, Quoc-Nam},
13
+ title = {Evaluating the Symbol Binding Ability of Large Language Models for Multiple-Choice Questions in Vietnamese General Education},
14
+ year = {2023},
15
+ isbn = {9798400708916},
16
+ publisher = {Association for Computing Machinery},
17
+ address = {New York, NY, USA},
18
+ url = {https://doi.org/10.1145/3628797.3628837},
19
+ doi = {10.1145/3628797.3628837},
20
+ booktitle = {Proceedings of the 12th International Symposium on Information and Communication Technology},
21
+ pages = {379–386},
22
+ numpages = {8},
23
+ keywords = {Analysis of Language Models, Multiple Choice Symbol Binding, Multiple Choice Question Answering, Language Modeling},
24
+ location = {<conf-loc>, <city>Ho Chi Minh</city>, <country>Vietnam</country>, </conf-loc>},
25
+ series = {SOICT '23}
26
+ }
27
+ """
28
+
29
+ _DATASETNAME = "vigetext"
30
+
31
+ _DESCRIPTION = """
32
+ The high-quality dataset with structured guidelines for typing LaTeX formulas in Mathematics, Physics, Chemistry, and
33
+ Biology. Objective was to cover the entire scope of the Vietnamese General Education Examination spanning from 2017 to 2023.
34
+ This comprehensive approach included the challenging examinations of the years 2017 and 2018, which have been significant
35
+ for nearly all Vietnamese students in recent years. It is important to highlight that the exact and unquestionably correct
36
+ answers have been exclusively obtained from the Vietnamese Ministry of Education.
37
+ """
38
+
39
+ _HOMEPAGE = "https://huggingface.co/datasets/uitnlp/ViGEText_17to23"
40
+
41
+ _LANGUAGES = ["vie"]
42
+
43
+ _LICENSE = Licenses.UNKNOWN.value
44
+
45
+ _LOCAL = False
46
+
47
+ _URLS = {
48
+ _DATASETNAME: {
49
+ "train": "https://huggingface.co/datasets/uitnlp/ViGEText_17to23/resolve/main/data/train-00000-of-00001.parquet",
50
+ "validation": "https://huggingface.co/datasets/uitnlp/ViGEText_17to23/resolve/main/data/validation-00000-of-00001.parquet",
51
+ "test": "https://huggingface.co/datasets/uitnlp/ViGEText_17to23/resolve/main/data/test-00000-of-00001.parquet",
52
+ }
53
+ }
54
+
55
+ _SUPPORTED_TASKS = [Tasks.QUESTION_ANSWERING]
56
+
57
+ _SOURCE_VERSION = "1.0.0"
58
+
59
+ _SEACROWD_VERSION = "2024.06.20"
60
+
61
+
62
+ class VigetextDataset(datasets.GeneratorBasedBuilder):
63
+ """Vigetext is a dataset for evaluating the Symbol Binding Ability of Large Language Models for Multiple-Choice Questions in Vietnamese General Education."""
64
+
65
+ SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
66
+ SEACROWD_VERSION = datasets.Version(_SEACROWD_VERSION)
67
+
68
+ BUILDER_CONFIGS = [
69
+ SEACrowdConfig(
70
+ name=f"{_DATASETNAME}_source",
71
+ version=SOURCE_VERSION,
72
+ description=f"{_DATASETNAME} source schema",
73
+ schema="source",
74
+ subset_id=_DATASETNAME,
75
+ ),
76
+ SEACrowdConfig(
77
+ name=f"{_DATASETNAME}_seacrowd_qa",
78
+ version=SEACROWD_VERSION,
79
+ description=f"{_DATASETNAME} SEACrowd schema",
80
+ schema="seacrowd_qa",
81
+ subset_id=_DATASETNAME,
82
+ ),
83
+ ]
84
+
85
+ DEFAULT_CONFIG_NAME = f"{_DATASETNAME}_source"
86
+
87
+ def _info(self) -> datasets.DatasetInfo:
88
+ if self.config.schema == "source":
89
+ features = datasets.Features(
90
+ {
91
+ "id": datasets.Value("string"),
92
+ "input": datasets.Value("string"),
93
+ "target": datasets.Value("string"),
94
+ }
95
+ )
96
+
97
+ else:
98
+ features = schemas.qa_features
99
+
100
+ return datasets.DatasetInfo(
101
+ description=_DESCRIPTION,
102
+ features=features,
103
+ homepage=_HOMEPAGE,
104
+ license=_LICENSE,
105
+ citation=_CITATION,
106
+ )
107
+
108
+ def _split_generators(self, dl_manager: datasets.DownloadManager) -> list[datasets.SplitGenerator]:
109
+ """Returns SplitGenerators."""
110
+ urls = _URLS[_DATASETNAME]
111
+ data_dir = dl_manager.download_and_extract(urls)
112
+
113
+ return [
114
+ datasets.SplitGenerator(
115
+ name=datasets.Split.TRAIN,
116
+ gen_kwargs={"filepath": data_dir, "split": "train"},
117
+ ),
118
+ datasets.SplitGenerator(
119
+ name=datasets.Split.VALIDATION,
120
+ gen_kwargs={"filepath": data_dir, "split": "validation"},
121
+ ),
122
+ datasets.SplitGenerator(
123
+ name=datasets.Split.TEST,
124
+ gen_kwargs={"filepath": data_dir, "split": "test"},
125
+ ),
126
+ ]
127
+
128
+ def _generate_examples(self, filepath: Path, split: str) -> tuple[int, dict]:
129
+ df = pd.read_parquet(filepath[split])
130
+ data = df.to_dict(orient="records")
131
+ for i, item in enumerate(data):
132
+ if self.config.schema == "source":
133
+ yield i, {
134
+ "id": item["id"],
135
+ "input": item["input"],
136
+ "target": item["target"],
137
+ }
138
+ else:
139
+ question_and_options = item["input"].split("\n")
140
+ answer_map = {opt[0]: opt[2:].strip() for opt in question_and_options[1:]}
141
+ yield i, {
142
+ "id": str(i),
143
+ "question_id": item["id"],
144
+ "document_id": "",
145
+ "question": question_and_options[0],
146
+ "type": "multiple_choice",
147
+ "choices": [opt[2:].strip() for opt in question_and_options[1:]], # remove A., B., ... in the options
148
+ "context": "",
149
+ "answer": [answer_map[item["target"]]],
150
+ "meta": {}
151
+ }