Dataset · allenai
Benchmark of highly complex user prompts for LLMs
allenai/WildBench
A selection of real, complex English prompts from WildChat for evaluating large language models.
What's inside
The dataset is a selected subset of real user dialogues from the WildChat corpus, structured as a benchmark by the Allen Institute for AI (AI2). It contains between 1 and 10 thousand English examples in tabular and text formats under the CC BY license.
What you can do with it
- Comparing and evaluating LLMs on real, non-template user tasks
- Analyzing the weaknesses of language models when handling complex instructions
- Tasks
- Text generation
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 2K
- Likes
- 41
How to load it
from datasets import load_dataset
ds = load_dataset("allenai/WildBench")