Dataset · UnipatAI
Wet lab experiment outcome prediction benchmark
UnipatAI/SPU
100 scientific tasks from recent papers to test AI's ability to rank candidates by real experiment outcomes.
What's inside
The dataset contains 100 public English tasks compiled from 85 scientific papers. Each task defines candidates, materials, conditions, and measurement protocols for a real wet lab experiment, requiring models to rank candidates by measured outcome. The data includes text and images.
What you can do with it
- Evaluating multimodal LLMs and scientific agents on experiment planning tasks
- Testing models' ability to predict physicochemical and biological outcomes without conducting experiments
- Languages
- English
- Size
- up to 1K rows
- Downloads this month
- 1.7K
- Likes
- 2
How to load it
from datasets import load_dataset
ds = load_dataset("UnipatAI/SPU")