Dataset · Qwen
Website interaction trajectories
Qwen/WebWorldData
1.06M trajectories (state, action, next_state) from real websites for training browser world models.
What's inside
This dataset contains 1.06M interaction trajectories from the open web. Each trajectory is a sequence of (state, action, next_state) transitions, where states are represented as A11y trees extracted from real websites via Playwright. Covering English and Chinese, it serves as training data for the WebWorld model series.
What you can do with it
- Training WebWorld browser world models
- Evaluating web agents using A11y trees
- Fine-tuning LLMs to predict the next website state
- Tasks
- Text generation
- Languages
- English, Chinese
- Size
- 100K–1M rows
- Downloads this month
- 1.4K
- Likes
- 83
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("Qwen/WebWorldData")