Create datasets-for-simcse

datasets-for-simcse

@@ -0,0 +1,63 @@
+# -*- coding: utf-8 -*-
+
+import os
+import json
+import datasets
+
+_DESCRIPTION = """datasets-for-simcse"""
+
+_CITATION = ''
+
+GITHUB_HOME = ''
+
+class DatasetsForSimCSEConfig(datasets.BuilderConfig):
+
+    def __init__(self, features, data_url, citation, url, label_classes=(0, 1), **kwargs):
+        super().__init__(version=datasets.Version("1.0.0"), **kwargs)
+        self.features = features
+        self.label_classes = label_classes
+        self.data_url = data_url
+        self.citation = citation
+        self.url = url
+
+
+class DatasetsForSimCSE(datasets.GeneratorBasedBuilder):
+    """The Natural Language Inference Chinese(NLI_zh) Corpus."""
+
+    BUILDER_CONFIGS = [
+        DatasetsForSimCSEConfig(
+            name="nli_for_simcse",
+            description=_DESCRIPTION,
+            features=datasets.Features({"sent0": datasets.Value("string"),
+                                          "sent1": datasets.Value("string"),
+                                          "hard_neg": datasets.Value("string")}),
+            data_url='https://huggingface.co/datasets/princeton-nlp/datasets-for-simcse/resolve/main/nli_for_simcse.csv',
+            citation=_CITATION,
+            url=GITHUB_HOME,
+        )
+    ]
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=self.config.description,
+            features=self.config.features,
+            homepage=self.config.url,
+            citation=self.config.citation,
+        )
+
+    def _split_generators(self, dl_manager):
+        filepath = dl_manager.download(self.config.data_url)
+        return [datasets.SplitGenerator(
+                        name='nli_for_simcse',
+                        gen_kwargs={
+                            "filepath": filepath,
+                        })]
+
+    def _generate_examples(self, filepath):
+        """This function returns the examples in the raw (text) form."""
+        with open(filepath, 'r', encoding="utf-8") as f:
+            f.readline()
+            for idx, row in enumerate(f.readlines()):
+                sent0, sent1, hard_neg = row.split(',')
+                context = {'sent0': sent0, 'sent1': sent1, 'hard_neg': hard_neg}
+                yield idx, context