Dataset · google
Multi-step information search benchmark
google/deepsearchqa
900 prompts for evaluating agents on complex multi-step information search tasks across 17 domains.
What's inside
A set of 900 English prompts for evaluating agents in complex multi-step information search tasks. Covers 17 different subject domains; this is a factuality benchmark from Google DeepMind. Includes links to the Kaggle leaderboard, technical report, and evaluation starter code.
What you can do with it
- Evaluating search agents on multi-step questions
- Checking the factuality of RAG systems
- Comparing models on the DeepSearchQA benchmark
- Tasks
- Question answering
- Languages
- English
- Size
- up to 1K rows
- Downloads this month
- 25.1K
- Likes
- 137
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("google/deepsearchqa")