Upload muse.py with huggingface_hub
Browse files
muse.py
ADDED
@@ -0,0 +1,197 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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 |
+
"""
|
17 |
+
Contains 110 large-scale ground-truth bilingual dictionaries created and released by Meta using an internal translation tool.
|
18 |
+
The dictionaries account for polysemy. The data comprises of a train and test split of 5000 and 1500 unique source words, as well as a larger set of up to 100k pairs.
|
19 |
+
It comprises of Europeans languages in every direction, and SEA languages to and from English.
|
20 |
+
"""
|
21 |
+
|
22 |
+
from pathlib import Path
|
23 |
+
from typing import Dict, List, Tuple
|
24 |
+
|
25 |
+
import datasets
|
26 |
+
|
27 |
+
from seacrowd.utils import schemas
|
28 |
+
from seacrowd.utils.configs import SEACrowdConfig
|
29 |
+
from seacrowd.utils.constants import Licenses, Tasks
|
30 |
+
|
31 |
+
_CITATION = """\
|
32 |
+
@inproceedings{lample2018word,
|
33 |
+
title={Word translation without parallel data},
|
34 |
+
author={Lample, Guillaume and Conneau, Alexis and Ranzato, Marc'Aurelio and Denoyer, Ludovic and J{\'e}gou, Herv{\'e}},
|
35 |
+
booktitle={International Conference on Learning Representations},
|
36 |
+
year={2018}}
|
37 |
+
}
|
38 |
+
"""
|
39 |
+
|
40 |
+
_DATASETNAME = "muse"
|
41 |
+
|
42 |
+
_DESCRIPTION = """\
|
43 |
+
Contains 110 large-scale ground-truth bilingual dictionaries created and released by Meta using an internal translation tool.
|
44 |
+
The dictionaries account for polysemy. The data comprises of a train and test split of 5000 and 1500 unique source words, as well as a larger set of up to 100k pairs.
|
45 |
+
It comprises of Europeans languages in every direction, and SEA languages to and from English.
|
46 |
+
"""
|
47 |
+
|
48 |
+
_HOMEPAGE = "https://github.com/facebookresearch/MUSE#ground-truth-bilingual-dictionaries"
|
49 |
+
|
50 |
+
_LANGUAGES = ["tgl", "ind", "zlm", "tha", "vie"]
|
51 |
+
|
52 |
+
_LICENSE = Licenses.CC_BY_NC_ND_4_0.value
|
53 |
+
|
54 |
+
_LOCAL = False
|
55 |
+
|
56 |
+
_TRAIN_URL_TEMPLATE = "https://dl.fbaipublicfiles.com/arrival/dictionaries/{src}-{tgt}.0-5000.txt"
|
57 |
+
_TEST_URL_TEMPLATE = "https://dl.fbaipublicfiles.com/arrival/dictionaries/{src}-{tgt}.5000-6500.txt"
|
58 |
+
|
59 |
+
_SUPPORTED_TASKS = [Tasks.MACHINE_TRANSLATION]
|
60 |
+
|
61 |
+
_SOURCE_VERSION = "1.0.0"
|
62 |
+
_SEACROWD_VERSION = "2024.06.20"
|
63 |
+
|
64 |
+
configs = {
|
65 |
+
"tgl": ["eng"],
|
66 |
+
"ind": ["eng"],
|
67 |
+
"zlm": ["eng"],
|
68 |
+
"tha": ["eng"],
|
69 |
+
"vie": ["eng"],
|
70 |
+
"eng": ["tha", "vie", "tgl", "zlm", "ind"],
|
71 |
+
}
|
72 |
+
|
73 |
+
langid_dict = {
|
74 |
+
"eng": "en",
|
75 |
+
"tgl": "tl",
|
76 |
+
"ind": "id",
|
77 |
+
"zlm": "ms",
|
78 |
+
"tha": "th",
|
79 |
+
"vie": "vi",
|
80 |
+
}
|
81 |
+
|
82 |
+
|
83 |
+
class MUSEDataset(datasets.GeneratorBasedBuilder):
|
84 |
+
"""Large-scale ground-truth bilingual dictionaries"""
|
85 |
+
|
86 |
+
SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
|
87 |
+
SEACROWD_VERSION = datasets.Version(_SEACROWD_VERSION)
|
88 |
+
|
89 |
+
BUILDER_CONFIGS = (
|
90 |
+
[
|
91 |
+
SEACrowdConfig(
|
92 |
+
name=f"{_DATASETNAME}_source",
|
93 |
+
version=SOURCE_VERSION,
|
94 |
+
description=f"{_DATASETNAME} source schema",
|
95 |
+
schema="source",
|
96 |
+
subset_id=f"{_DATASETNAME}_tgl_eng",
|
97 |
+
),
|
98 |
+
SEACrowdConfig(
|
99 |
+
name=f"{_DATASETNAME}_seacrowd_t2t",
|
100 |
+
version=SEACROWD_VERSION,
|
101 |
+
description=f"{_DATASETNAME} SEACrowd schema",
|
102 |
+
schema="seacrowd_t2t",
|
103 |
+
subset_id=f"{_DATASETNAME}_tgl_eng",
|
104 |
+
),
|
105 |
+
]
|
106 |
+
+ [
|
107 |
+
SEACrowdConfig(
|
108 |
+
name=f"{_DATASETNAME}_{src_lang}_{tgt_lang}_source",
|
109 |
+
version=datasets.Version(_SOURCE_VERSION),
|
110 |
+
description=f"{_DATASETNAME} source schema",
|
111 |
+
schema="source",
|
112 |
+
subset_id=f"{_DATASETNAME}_{src_lang}_{tgt_lang}",
|
113 |
+
)
|
114 |
+
for src_lang in configs
|
115 |
+
for tgt_lang in configs[src_lang]
|
116 |
+
]
|
117 |
+
+ [
|
118 |
+
SEACrowdConfig(
|
119 |
+
name=f"{_DATASETNAME}_{src_lang}_{tgt_lang}_seacrowd_t2t",
|
120 |
+
version=datasets.Version(_SOURCE_VERSION),
|
121 |
+
description=f"{_DATASETNAME} SEACrowd schema",
|
122 |
+
schema="seacrowd_t2t",
|
123 |
+
subset_id=f"{_DATASETNAME}_{src_lang}_{tgt_lang}",
|
124 |
+
)
|
125 |
+
for src_lang in configs
|
126 |
+
for tgt_lang in configs[src_lang]
|
127 |
+
]
|
128 |
+
)
|
129 |
+
|
130 |
+
DEFAULT_CONFIG_NAME = f"{_DATASETNAME}_source"
|
131 |
+
|
132 |
+
def _info(self) -> datasets.DatasetInfo:
|
133 |
+
|
134 |
+
if self.config.schema == "source":
|
135 |
+
features = datasets.Features(
|
136 |
+
{
|
137 |
+
"id": datasets.Value("string"),
|
138 |
+
"src_text": datasets.Value("string"),
|
139 |
+
"tgt_text": datasets.Value("string"),
|
140 |
+
}
|
141 |
+
)
|
142 |
+
|
143 |
+
elif self.config.schema == "seacrowd_t2t":
|
144 |
+
features = schemas.text2text_features
|
145 |
+
|
146 |
+
return datasets.DatasetInfo(
|
147 |
+
description=_DESCRIPTION,
|
148 |
+
features=features,
|
149 |
+
homepage=_HOMEPAGE,
|
150 |
+
license=_LICENSE,
|
151 |
+
citation=_CITATION,
|
152 |
+
)
|
153 |
+
|
154 |
+
def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
|
155 |
+
|
156 |
+
_, src_lang, tgt_lang = self.config.subset_id.split("_")
|
157 |
+
train_url = _TRAIN_URL_TEMPLATE.format(src=langid_dict[src_lang], tgt=langid_dict[tgt_lang])
|
158 |
+
test_url = _TEST_URL_TEMPLATE.format(src=langid_dict[src_lang], tgt=langid_dict[tgt_lang])
|
159 |
+
|
160 |
+
train_file = dl_manager.download_and_extract(train_url)
|
161 |
+
test_file = dl_manager.download_and_extract(test_url)
|
162 |
+
|
163 |
+
return [
|
164 |
+
datasets.SplitGenerator(
|
165 |
+
name=datasets.Split.TRAIN,
|
166 |
+
gen_kwargs={
|
167 |
+
"src_lang": src_lang,
|
168 |
+
"tgt_lang": tgt_lang,
|
169 |
+
"filepath": train_file,
|
170 |
+
},
|
171 |
+
),
|
172 |
+
datasets.SplitGenerator(
|
173 |
+
name=datasets.Split.VALIDATION,
|
174 |
+
gen_kwargs={
|
175 |
+
"src_lang": src_lang,
|
176 |
+
"tgt_lang": tgt_lang,
|
177 |
+
"filepath": test_file,
|
178 |
+
},
|
179 |
+
),
|
180 |
+
]
|
181 |
+
|
182 |
+
def _generate_examples(self, src_lang: str, tgt_lang: str, filepath: Path) -> Tuple[int, Dict]:
|
183 |
+
if self.config.schema == "source":
|
184 |
+
for row_id, line in enumerate(open(filepath)):
|
185 |
+
src_text, tgt_text = line.strip().split("\t")
|
186 |
+
yield row_id, {"id": row_id, "src_text": src_text, "tgt_text": tgt_text}
|
187 |
+
|
188 |
+
elif self.config.schema == "seacrowd_t2t":
|
189 |
+
for row_id, line in enumerate(open(filepath)):
|
190 |
+
src_text, tgt_text = line.strip().split("\t")
|
191 |
+
yield row_id, {
|
192 |
+
"id": row_id,
|
193 |
+
"text_1": src_text,
|
194 |
+
"text_2": tgt_text,
|
195 |
+
"text_1_name": src_lang,
|
196 |
+
"text_2_name": tgt_lang,
|
197 |
+
}
|