Dataset · databricks
Databricks benchmark for complex financial documents
databricks/officeqa
Q&A based on historical US Treasury bulletins to evaluate AI agents' ability to analyze tables, charts, and text.
What's inside
This benchmark by Databricks contains fewer than a thousand English Q&A pairs linked to real documents. Sourced from historical US Department of the Treasury bulletins from 1939 to 2025, it is packed with complex financial tables, diagrams, and text. The dataset is designed to end-to-end test how models search for facts, use tools, and build multi-step reasoning.
What you can do with it
- Evaluating the quality of RAG systems and AI agents working with complex reports
- Testing multi-step reasoning skills and data extraction from tables and charts
- Tasks
- Question answering, Text generation, Text retrieval
- Languages
- English
- Size
- up to 1K rows
- Downloads this month
- 35.2K
- Likes
- 42
- Access
- on request on Hugging Face
How to load it
from datasets import load_dataset
ds = load_dataset("databricks/officeqa")