A spreadsheet-agent benchmark of financial-modeling tasks. Each task provides an
Excel input workbook (and, where applicable, a reference output workbook plus a
grading rubric). This dataset is published in Arrow-compatible Parquet so it can
be loaded directly with 🤗 datasets — no custom loading script required.
The original per-task files remain in the tasks/ tree; the Parquet subsets
under data/ are a row-assembled, Arrow-native view of those same files.… See the full description on the dataset page:
https://huggingface.co/datasets/Longitude-Labs/bluefin-release.
