Datasets:
metadata
license: mit
task_categories:
- question-answering
pretty_name: TabMWP Selection
size_categories:
- n<1K
This dataset is a high-fidelity selection from the Tabular Math Word Problems (TabMWP) benchmark (Lu et al., 2023). TabMWP is a leading resource for evaluating mathematical reasoning over heterogeneous tabular and textual data. To address potential noise and ensure the highest standards of logical grounding, this curated version consists of 100 hand-verified examples. Each entry has been audited to confirm that the multi-step reasoning chains—including information look-up and numerical computation—are factually sound and correctly aligned with the provided tabular context. This subset serves as an essential "gold standard" for evaluating the precision of LLMs in complex, semi-structured mathematical tasks.