File size: 1,949 Bytes
8f4c463 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 |
import datasets
_LANGUAGE_PAIRS = {
"en_de": "data/en_de.parquet",
"en_es": "data/en_es.parquet",
"en_fr": "data/en_fr.parquet",
"en_it": "data/en_it.parquet",
"en_ko": "data/en_ko.parquet",
"en_nl": "data/en_nl.parquet",
"en_pt": "data/en_pt.parquet",
"en_ru": "data/en_ru.parquet",
"en_zh": "data/en_zh.parquet",
}
VERSION = datasets.Version("1.0.0")
class Spite(datasets.GeneratorBasedBuilder):
VERSION = VERSION
BUILDER_CONFIGS = [
datasets.BuilderConfig(
name=lang_pair,
version=VERSION,
description=f"Training data for language pair: {lang_pair}",
)
for lang_pair in _LANGUAGE_PAIRS
]
DEFAULT_CONFIG_NAME = "en_de"
def _info(self):
return datasets.DatasetInfo(
description="Pseudolabeled speech translation dataset",
features=datasets.Features({
"src": datasets.Value("string"),
"mt": datasets.Value("string"),
'cometqe_22': datasets.Value("float64"),
'xcomet_xl': datasets.Value("float64"),
'blaser2_src': datasets.Value("float64"),
'audio_length': datasets.Value("float64"),
"example_id": datasets.Value("string"),
"index": datasets.Value("int64")
# add more columns if needed
}),
supervised_keys=None,
)
def _split_generators(self, dl_manager):
parquet_path = dl_manager.download(_LANGUAGE_PAIRS[self.config.name])
return [
datasets.SplitGenerator(
name=datasets.Split.TRAIN,
gen_kwargs={"filepath": parquet_path},
)
]
def _generate_examples(self, filepath):
import pyarrow.parquet as pq
table = pq.read_table(filepath)
for idx, row in enumerate(table.to_pylist()):
yield idx, row
|