data_loader.py

from datasets import DatasetInfo, GeneratorBasedBuilder, SplitGenerator, Split, Value

class SinhalaPerplexity(GeneratorBasedBuilder):
    def _info(self):
        return DatasetInfo(
            description="Sinhala Unicode and Romanized sentence pairs for perplexity evaluation.",
            features={
                "sinhala_unicode": Value("string"),
                "sinhala_romanized": Value("string"),
            },
            supervised_keys=None,
        )

    def _split_generators(self, dl_manager):
        data_path = dl_manager.download("data/diverse_sentences.csv")
        return [
            SplitGenerator(
                name=Split("data"),  # 👈 Your custom split name
                gen_kwargs={"filepath": data_path},
            )
        ]

    def _generate_examples(self, filepath):
        import csv
        with open(filepath, encoding="utf-8") as f:
            reader = csv.DictReader(f)
            for idx, row in enumerate(reader):
                yield idx, {
                    "sinhala_unicode": row["unicode_sentence"],
                    "sinhala_romanized": row["romanized_sentence"],
                }