Create IndoMMLU.py
Browse files- IndoMMLU.py +171 -0
IndoMMLU.py
ADDED
@@ -0,0 +1,171 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import csv
|
2 |
+
import datasets
|
3 |
+
|
4 |
+
_CITATION = """\
|
5 |
+
@inproceedings{koto-etal-2023-indommlu,
|
6 |
+
title = "Large Language Models Only Pass Primary School Exams in {I}ndonesia: A Comprehensive Test on {I}ndo{MMLU}",
|
7 |
+
author = "Fajri Koto and Nurul Aisyah and Haonan Li and Timothy Baldwin",
|
8 |
+
booktitle = "Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP)",
|
9 |
+
month = December,
|
10 |
+
year = "2023",
|
11 |
+
address = "Singapore",
|
12 |
+
publisher = "Association for Computational Linguistics",
|
13 |
+
}"""
|
14 |
+
|
15 |
+
|
16 |
+
subject2english = {
|
17 |
+
'Sejarah': 'History',
|
18 |
+
'Geografi': 'Geography',
|
19 |
+
'Bahasa Lampung': 'Lampungic',
|
20 |
+
'IPS': 'Social science',
|
21 |
+
'Bahasa Bali': 'Balinese',
|
22 |
+
'Bahasa Makassar': 'Makassarese',
|
23 |
+
'Bahasa Banjar': 'Banjarese',
|
24 |
+
'Kimia': 'Chemistry',
|
25 |
+
'Biologi': 'Biology',
|
26 |
+
'IPA': 'Science',
|
27 |
+
'Agama Kristen': 'Christian religion',
|
28 |
+
'Kesenian': 'Art',
|
29 |
+
'Agama Islam': 'Islam religion',
|
30 |
+
'Agama Hindu': 'Hindu religion',
|
31 |
+
'Bahasa Madura': 'Madurese',
|
32 |
+
'Penjaskes': 'Sport',
|
33 |
+
'Bahasa Indonesia': 'Indonesian language',
|
34 |
+
'Fisika': 'Physics',
|
35 |
+
'Budaya Alam Minangkabau': 'Minangkabau culture',
|
36 |
+
'Bahasa Dayak Ngaju': 'Dayak language',
|
37 |
+
'Sosiologi': 'Sociology',
|
38 |
+
'Ekonomi': 'Economy',
|
39 |
+
'Bahasa Sunda': 'Sundanese',
|
40 |
+
'Bahasa Jawa': 'Javanese',
|
41 |
+
'PPKN': 'Civic education',
|
42 |
+
}
|
43 |
+
|
44 |
+
subject2group = {
|
45 |
+
'Sejarah': 'Humanities',
|
46 |
+
'Geografi': 'Social science',
|
47 |
+
'Bahasa Lampung': 'Local languages and cultures',
|
48 |
+
'IPS': 'Social science',
|
49 |
+
'Bahasa Bali': 'Local languages and cultures',
|
50 |
+
'Bahasa Makassar': 'Local languages and cultures',
|
51 |
+
'Bahasa Banjar': 'Local languages and cultures',
|
52 |
+
'Kimia': 'STEM',
|
53 |
+
'Biologi': 'STEM',
|
54 |
+
'IPA': 'STEM',
|
55 |
+
'Agama Kristen': 'Humanities',
|
56 |
+
'Kesenian': 'Humanities',
|
57 |
+
'Agama Islam': 'Humanities',
|
58 |
+
'Agama Hindu': 'Humanities',
|
59 |
+
'Bahasa Madura': 'Local languages and cultures',
|
60 |
+
'Penjaskes': 'Humanities',
|
61 |
+
'Bahasa Indonesia': 'Indonesian language',
|
62 |
+
'Fisika': 'STEM',
|
63 |
+
'Budaya Alam Minangkabau': 'Local languages and cultures',
|
64 |
+
'Bahasa Dayak Ngaju': 'Local languages and cultures',
|
65 |
+
'Sosiologi': 'Social science',
|
66 |
+
'Ekonomi': 'Social science',
|
67 |
+
'Bahasa Sunda': 'Local languages and cultures',
|
68 |
+
'Bahasa Jawa': 'Local languages and cultures',
|
69 |
+
'PPKN': 'Social science',
|
70 |
+
}
|
71 |
+
|
72 |
+
special_case = ['SD-SMP-SMA', 'SD-SMP']
|
73 |
+
level_mapper = {
|
74 |
+
'SMA': 'SMA',
|
75 |
+
'Seleksi PTN': 'University entrance test',
|
76 |
+
'SD': 'SD',
|
77 |
+
'SMP': 'SMP',
|
78 |
+
'Kelas I SD': 'SD',
|
79 |
+
'Kelas X SMA': 'SMA',
|
80 |
+
'Kelas XI SMA': 'SMA',
|
81 |
+
'Kelas XII SMA': 'SMA',
|
82 |
+
'V SD': 'SD',
|
83 |
+
'VI SD': 'SD',
|
84 |
+
'VII SMP': 'SMP',
|
85 |
+
'VIII SMP ': 'SMP',
|
86 |
+
'IX SMP': 'SMP',
|
87 |
+
'Kelas III SD':'SD',
|
88 |
+
'Kelas IV SD': 'SD',
|
89 |
+
'Kelas II SD': 'SD'
|
90 |
+
}
|
91 |
+
|
92 |
+
def fix_level(level, kelas):
|
93 |
+
#Fixing Level
|
94 |
+
if level in special_case:
|
95 |
+
kelas = float(kelas)
|
96 |
+
if kelas >=1 and kelas <= 6:
|
97 |
+
level = 'SD'
|
98 |
+
elif kelas >=7 and kelas <= 9:
|
99 |
+
level = 'SMP'
|
100 |
+
elif kelas >=10:
|
101 |
+
level = 'SMA'
|
102 |
+
else:
|
103 |
+
print(level)
|
104 |
+
fixed_level = level_mapper[level]
|
105 |
+
|
106 |
+
#Fixing class
|
107 |
+
fixed_kelas = -1
|
108 |
+
if kelas.strip() in ['PTN', '2023-10-12 00:00:00']:
|
109 |
+
fixed_kelas = 13
|
110 |
+
elif kelas == '4,5,6':
|
111 |
+
fixed_kelas = 6
|
112 |
+
else:
|
113 |
+
fixed_kelas = int(kelas.strip())
|
114 |
+
|
115 |
+
# sanity check over the level and kelas
|
116 |
+
return fixed_level, fixed_kelas
|
117 |
+
|
118 |
+
|
119 |
+
_URL = {
|
120 |
+
'test': "https://huggingface.co/datasets/indolem/IndoMMLU/resolve/main/test.csv",
|
121 |
+
}
|
122 |
+
|
123 |
+
class IndoMMLUConfig(datasets.BuilderConfig):
|
124 |
+
"""IndoMMLUConfig for IndoMMLU"""
|
125 |
+
|
126 |
+
def __init__(self, **kwargs):
|
127 |
+
"""BuilderConfig for IndoStoryCloze.
|
128 |
+
**kwargs: keyword arguments forwarded to super.
|
129 |
+
"""
|
130 |
+
# Version history:
|
131 |
+
# 1.0.0: Release version
|
132 |
+
super().__init__(version=datasets.Version("1.0.0"), **kwargs)
|
133 |
+
self.features = ['subject', 'group', 'level', 'class', 'question', 'options', 'answer', 'is_for_fewshot']
|
134 |
+
|
135 |
+
|
136 |
+
class IndoMMLU(datasets.GeneratorBasedBuilder):
|
137 |
+
"""The IndoMMLU Datasets."""
|
138 |
+
|
139 |
+
BUILDER_CONFIGS = [IndoMMLUConfig()]
|
140 |
+
|
141 |
+
def _info(self):
|
142 |
+
features = {feature: datasets.Value("string") for feature in self.config.features}
|
143 |
+
|
144 |
+
return datasets.DatasetInfo(
|
145 |
+
description='IndoMMLU',
|
146 |
+
features=datasets.Features(features),
|
147 |
+
homepage='https://github.com/fajri91/IndoMMLU',
|
148 |
+
citation=_CITATION
|
149 |
+
)
|
150 |
+
|
151 |
+
def _split_generators(self, dl_manager):
|
152 |
+
downloaded_file = dl_manager.download_and_extract(_URLs)
|
153 |
+
|
154 |
+
return [
|
155 |
+
datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"data_file": downloaded_file['test']}),
|
156 |
+
]
|
157 |
+
|
158 |
+
def _generate_examples(self, data_file):
|
159 |
+
data = csv.DictReader(open(data_file, newline=''))
|
160 |
+
for i, row in enumerate(data):
|
161 |
+
fixed_level, fixed_kelas = fix_level(row['level'], row['kelas'])
|
162 |
+
yield i, {
|
163 |
+
"subject": subject2english[row['subject']],
|
164 |
+
"group": subject2group[row['subject']],
|
165 |
+
"level": fixed_level,
|
166 |
+
"class": fixed_class,
|
167 |
+
"question": row['soal'],
|
168 |
+
"options": row['jawaban'].split('\n'),
|
169 |
+
"answer": row['kunci'],
|
170 |
+
"is_for_fewshot": row['is_for_fewshot']
|
171 |
+
}
|