This dataset keeps the original English prompt in text_en and adds a Korean translation in text_ko.
The translation source files are stored in translations/train_ko.txt and translations/test_ko.txt. To rebuild the Parquet files from those translations, run:
uv run python scripts/build_bilingual_dataset.py
