Create jhumaneval.py

jhumaneval.py

@@ -0,0 +1,63 @@
+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