Dataset · ByteDance-Seed
Wide Information Retrieval Benchmark for AI Agents
ByteDance-Seed/WideSearch
Compact English and Chinese text tasks from ByteDance to evaluate agents' ability to gather scattered facts.
What's inside
A compact text benchmark (fewer than 1,000 examples) in English and Chinese from the ByteDance-Seed team. Unlike single-fact retrieval tests, these tasks require LLM agents to collect a large volume of scattered but accessible information.
What you can do with it
- Evaluating retrieval and research LLM agents
- Comparing planning and data aggregation strategies during web search
- Languages
- Chinese, English
- Size
- up to 1K rows
- Downloads this month
- 24K
- Likes
- 47
How to load it
from datasets import load_dataset
ds = load_dataset("ByteDance-Seed/WideSearch")