holylovenia
commited on
Commit
·
8bc5179
1
Parent(s):
9839b57
Upload indo4b_plus.py with huggingface_hub
Browse files- indo4b_plus.py +180 -0
indo4b_plus.py
ADDED
@@ -0,0 +1,180 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# coding=utf-8
|
2 |
+
# Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor.
|
3 |
+
#
|
4 |
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
+
# you may not use this file except in compliance with the License.
|
6 |
+
# You may obtain a copy of the License at
|
7 |
+
#
|
8 |
+
# http://www.apache.org/licenses/LICENSE-2.0
|
9 |
+
#
|
10 |
+
# Unless required by applicable law or agreed to in writing, software
|
11 |
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
12 |
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
13 |
+
# See the License for the specific language governing permissions and
|
14 |
+
# limitations under the License.
|
15 |
+
|
16 |
+
from posixpath import split
|
17 |
+
from typing import Dict, List, Tuple
|
18 |
+
|
19 |
+
import datasets
|
20 |
+
|
21 |
+
from nusacrowd.utils import schemas
|
22 |
+
from nusacrowd.utils.configs import NusantaraConfig
|
23 |
+
from nusacrowd.utils.constants import (DEFAULT_NUSANTARA_VIEW_NAME,
|
24 |
+
DEFAULT_SOURCE_VIEW_NAME, Tasks)
|
25 |
+
|
26 |
+
_DATASETNAME = "indo4b_plus"
|
27 |
+
_SOURCE_VIEW_NAME = DEFAULT_SOURCE_VIEW_NAME
|
28 |
+
_UNIFIED_VIEW_NAME = DEFAULT_NUSANTARA_VIEW_NAME
|
29 |
+
|
30 |
+
_LOCAL = False
|
31 |
+
_LANGUAGES = ["ind", "sun", "jav"] # We follow ISO639-3 language code (https://iso639-3.sil.org/code_tables/639/data)
|
32 |
+
|
33 |
+
_CITATION = """\
|
34 |
+
@inproceedings{cahyawijaya-etal-2021-indonlg,
|
35 |
+
title = "{I}ndo{NLG}: Benchmark and Resources for Evaluating {I}ndonesian Natural Language Generation",
|
36 |
+
author = "Cahyawijaya, Samuel and
|
37 |
+
Winata, Genta Indra and
|
38 |
+
Wilie, Bryan and
|
39 |
+
Vincentio, Karissa and
|
40 |
+
Li, Xiaohong and
|
41 |
+
Kuncoro, Adhiguna and
|
42 |
+
Ruder, Sebastian and
|
43 |
+
Lim, Zhi Yuan and
|
44 |
+
Bahar, Syafri and
|
45 |
+
Khodra, Masayu and
|
46 |
+
Purwarianti, Ayu and
|
47 |
+
Fung, Pascale",
|
48 |
+
booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing",
|
49 |
+
month = nov,
|
50 |
+
year = "2021",
|
51 |
+
address = "Online and Punta Cana, Dominican Republic",
|
52 |
+
publisher = "Association for Computational Linguistics",
|
53 |
+
url = "https://aclanthology.org/2021.emnlp-main.699",
|
54 |
+
doi = "10.18653/v1/2021.emnlp-main.699",
|
55 |
+
pages = "8875--8898",
|
56 |
+
abstract = "Natural language generation (NLG) benchmarks provide an important avenue to measure progress
|
57 |
+
and develop better NLG systems. Unfortunately, the lack of publicly available NLG benchmarks for low-resource
|
58 |
+
languages poses a challenging barrier for building NLG systems that work well for languages with limited
|
59 |
+
amounts of data. Here we introduce IndoNLG, the first benchmark to measure natural language generation (NLG)
|
60 |
+
progress in three low-resource{---}yet widely spoken{---}languages of Indonesia: Indonesian, Javanese, and Sundanese.
|
61 |
+
Altogether, these languages are spoken by more than 100 million native speakers, and hence constitute an important
|
62 |
+
use case of NLG systems today. Concretely, IndoNLG covers six tasks: summarization, question answering, chit-chat,
|
63 |
+
and three different pairs of machine translation (MT) tasks. We collate a clean pretraining corpus of Indonesian,
|
64 |
+
Sundanese, and Javanese datasets, Indo4B-Plus, which is used to pretrain our models: IndoBART and IndoGPT.
|
65 |
+
We show that IndoBART and IndoGPT achieve competitive performance on all tasks{---}despite using only one-fifth
|
66 |
+
the parameters of a larger multilingual model, mBART-large (Liu et al., 2020). This finding emphasizes
|
67 |
+
the importance of pretraining on closely related, localized languages to achieve more efficient learning and faster inference
|
68 |
+
at very low-resource languages like Javanese and Sundanese.",
|
69 |
+
}
|
70 |
+
"""
|
71 |
+
|
72 |
+
_DESCRIPTION = """\
|
73 |
+
Indo4B-Plus is an extension of Indo4B, a large-scale Indonesian self-supervised pre-training corpus.
|
74 |
+
Indo4B-Plus extend Indo4B by adding two low-resource Indonesian local languages to the corpus, i.e., Sundanese and Javanese.
|
75 |
+
Indo4B-Plus adds 82,582,025 words (∼2.07%) of Sundanese sentences and 331,041,877 words (∼8.29%) of Javanese
|
76 |
+
"""
|
77 |
+
|
78 |
+
_HOMEPAGE = "https://github.com/IndoNLP/indonlu"
|
79 |
+
|
80 |
+
_LICENSE = "CC0"
|
81 |
+
|
82 |
+
_LANGUAGES_MAP = {
|
83 |
+
"ind": "id",
|
84 |
+
"jav": "jv",
|
85 |
+
"sun": "su",
|
86 |
+
}
|
87 |
+
|
88 |
+
_URLS = {
|
89 |
+
"indo4b": "https://storage.googleapis.com/babert-pretraining/IndoNLG_finals/IndoNLG_ALL_new_dataset_preprocessed_uncased.txt.zip",
|
90 |
+
}
|
91 |
+
|
92 |
+
_SUPPORTED_TASKS = [Tasks.SELF_SUPERVISED_PRETRAINING]
|
93 |
+
|
94 |
+
_SOURCE_VERSION = "1.0.0"
|
95 |
+
|
96 |
+
_NUSANTARA_VERSION = "1.0.0"
|
97 |
+
|
98 |
+
class Indo4BPlus(datasets.GeneratorBasedBuilder):
|
99 |
+
"""Indo4B-Plus is a large-scale Indonesian self-supervised pre-training corpus consists
|
100 |
+
of around 4B words, covering three languages, i.e., Indonesian, Sundanese, and Javanese."""
|
101 |
+
|
102 |
+
DEFAULT_CONFIG_NAME = "indo4b_plus_source"
|
103 |
+
|
104 |
+
|
105 |
+
BUILDER_CONFIGS = [
|
106 |
+
NusantaraConfig(
|
107 |
+
name="indo4b_plus_source",
|
108 |
+
version=_SOURCE_VERSION,
|
109 |
+
description="Indo4B-Plus source schema",
|
110 |
+
schema="source",
|
111 |
+
subset_id="indo4b_plus",
|
112 |
+
),
|
113 |
+
NusantaraConfig(
|
114 |
+
name="indo4b_plus_nusantara_ssp",
|
115 |
+
version=_NUSANTARA_VERSION,
|
116 |
+
description="Indo4B-Plus Nusantara schema",
|
117 |
+
schema="nusantara_ssp",
|
118 |
+
subset_id="indo4b_plus",
|
119 |
+
),
|
120 |
+
]
|
121 |
+
|
122 |
+
def _info(self) -> datasets.DatasetInfo:
|
123 |
+
if self.config.schema == "source":
|
124 |
+
features = datasets.Features(
|
125 |
+
{
|
126 |
+
"id": datasets.Value("string"),
|
127 |
+
"text": datasets.Value("string"),
|
128 |
+
}
|
129 |
+
)
|
130 |
+
elif self.config.schema == "nusantara_ssp":
|
131 |
+
features = schemas.self_supervised_pretraining.features
|
132 |
+
|
133 |
+
return datasets.DatasetInfo(
|
134 |
+
description=_DESCRIPTION,
|
135 |
+
features=features,
|
136 |
+
homepage=_HOMEPAGE,
|
137 |
+
license=_LICENSE,
|
138 |
+
citation=_CITATION,
|
139 |
+
)
|
140 |
+
|
141 |
+
def _split_generators(self, dl_manager) -> List[datasets.SplitGenerator]:
|
142 |
+
"""Returns SplitGenerators."""
|
143 |
+
|
144 |
+
url = _URLS["indo4b"]
|
145 |
+
path = dl_manager.download_and_extract(url) + "/IndoNLG_ALL_new_dataset_preprocessed_uncased.txt"
|
146 |
+
|
147 |
+
return [
|
148 |
+
datasets.SplitGenerator(
|
149 |
+
name=datasets.Split.TRAIN,
|
150 |
+
gen_kwargs={
|
151 |
+
"filepath": path,
|
152 |
+
"split": "train",
|
153 |
+
},
|
154 |
+
),
|
155 |
+
]
|
156 |
+
|
157 |
+
def _generate_examples(self, filepath, split: str) -> Tuple[int, Dict]:
|
158 |
+
"""Yields examples as (key, example) tuples."""
|
159 |
+
|
160 |
+
with open(filepath, encoding="utf-8") as f:
|
161 |
+
if self.config.schema == "source":
|
162 |
+
for counter, row in enumerate(f):
|
163 |
+
if row.strip() != "":
|
164 |
+
yield (
|
165 |
+
counter,
|
166 |
+
{
|
167 |
+
"id": str(counter),
|
168 |
+
"text": row.strip(),
|
169 |
+
},
|
170 |
+
)
|
171 |
+
elif self.config.schema == "nusantara_ssp":
|
172 |
+
for counter, row in enumerate(f):
|
173 |
+
if row.strip() != "":
|
174 |
+
yield (
|
175 |
+
counter,
|
176 |
+
{
|
177 |
+
"id": str(counter),
|
178 |
+
"text": row.strip(),
|
179 |
+
},
|
180 |
+
)
|