Dataset · zjunlp
LongDS multi-step data analysis benchmark for AI agents
zjunlp/LongDS
The LongDS-Bench multimodal benchmark (up to 1,000 examples) tests how agents maintain context during long data analysis sessions.
What's inside
A dataset for the LongDS-Bench benchmark by the ZJUNLP team, containing fewer than 1,000 examples in English across text and image formats. Tasks simulate real-world multi-step data analytics where filters, metric definitions, assumptions, and intermediate tables change throughout the dialogue. The data storage format is not specified in the description.
What you can do with it
- Testing AI agents' ability to maintain complex context and branching in long analytical sessions
- Comparing multimodal LLMs in sequential table and chart analysis tasks
- Tasks
- Question answering, Text generation
- Languages
- English
- Size
- up to 1K rows
- Downloads this month
- 2.5K
- Likes
- 7
How to load it
from datasets import load_dataset
ds = load_dataset("zjunlp/LongDS")