| --- |
| license: openrail++ |
| task_categories: |
| - table-question-answering |
| --- |
| |
| **BiomedSQL** |
|
|
| [GitHub](https://github.com/NIH-CARD/biomedsql) |
|
|
| [Paper](https://huggingface.co/papers/2505.20321) |
|
|
| **Dataset Summary**<br> |
| BiomedSQL is a text-to-SQL benchmark designed to evaluate Large Language Models (LLMs) on scientific tabular reasoning tasks. It consists of curated question-SQL query-answer triples covering a variety of biomedical and SQL reasoning types. The benchmark challenges models to apply implicit |
| scientific criteria rather than simply translating syntax. |
|
|
| **Repository Organization** |
| - **benchmark_data**: contains the question-SQL query-answer triples. |
| - Please note that you will have to replace the ```{project_id}``` and ```{dataset_name}``` placeholder strings in the ```benchmark_query``` column with the names of the GCP project and BigQuery dataset that you create if you wish to rerun them. Our GitHub code automatically handles this entity replacement through ```.env``` configurations when running experiment code.<br> |
| - **db_data**: contains the parquet files needed to recreate the tables in the evaluation database. |
