Dataset · NEAR-AI
US Treasury Bulletin QA (OfficeQA Manifest)
NEAR-AI/officeqa
A set of 246 English questions for evaluating QA search systems on financial documents.
What's inside
A benchmark-adapted manifest of questions from the databricks/officeqa dataset based on US Treasury bulletins from 1939-2025. The full version contains 246 English text examples, and a smoke split of 8 questions (4 simple and 4 complex) is highlighted for rapid pipeline verification. The source documents themselves are not included in the repository.
What you can do with it
- Evaluating the quality of RAG systems and models in financial document QA tasks
- Rapid verification of benchmark pipeline performance on a mini-sample of 8 questions
- Tasks
- Question answering
- Languages
- English
- Size
- up to 1K rows
- Downloads this month
- 51.8K
- Likes
- 0
How to load it
from datasets import load_dataset
ds = load_dataset("NEAR-AI/officeqa")