|
import json |
|
import datasets |
|
|
|
|
|
_DESCRIPTION = """\ |
|
This is a Japanese translated version of HumanEval, an evaluation harness for the HumanEval problem solving dataset described in the paper "Evaluating Large Language Models Trained on Code". |
|
""" |
|
|
|
_URL = "https://raw.githubusercontent.com/KuramitsuLab/jhuman-eval/main/data/jhuman-eval.jsonl.gz" |
|
|
|
_LICENSE = "MIT" |
|
|
|
|
|
class JHumaneval(datasets.GeneratorBasedBuilder): |
|
VERSION = datasets.Version("0.1.0") |
|
|
|
BUILDER_CONFIGS = [ |
|
datasets.BuilderConfig( |
|
name="jhumaneval", |
|
version=VERSION, |
|
description=_DESCRIPTION, |
|
) |
|
] |
|
|
|
def _info(self): |
|
features = datasets.Features( |
|
{ |
|
"task_id": datasets.Value("string"), |
|
"prompt_en": datasets.Value("string"), |
|
"prompt": datasets.Value("string"), |
|
"entry_point": datasets.Value("string"), |
|
"canonical_solution": datasets.Value("string"), |
|
"test": datasets.Value("string"), |
|
} |
|
) |
|
|
|
return datasets.DatasetInfo( |
|
description=_DESCRIPTION, |
|
features=features, |
|
supervised_keys=None, |
|
license=_LICENSE, |
|
) |
|
|
|
def _split_generators(self, dl_manager): |
|
"""Returns SplitGenerators.""" |
|
data_dir = dl_manager.download_and_extract(_URL) |
|
return [ |
|
datasets.SplitGenerator( |
|
name=datasets.Split.TEST, |
|
gen_kwargs={ |
|
"filepath": data_dir, |
|
}, |
|
) |
|
] |
|
|
|
def _generate_examples(self, filepath): |
|
"""Yields examples.""" |
|
with open(filepath, encoding="utf-8") as file: |
|
data = [json.loads(line) for line in file] |
|
id_ = 0 |
|
for sample in data: |
|
yield id_, sample |
|
id_ += 1 |