Dataset · McGill-NLP
Real website navigation with multi-turn dialogs
McGill-NLP/WebLINX-full
Multimodal web interaction recordings formatted as dialogs for training web agents.
What's inside
Contains text dialogs, images, and video reflecting English navigation across real websites. Dataset size is between 10k and 100k records. The card notes this is the full version of the McGill-NLP dataset, not the main project page.
What you can do with it
- Training autonomous web agents to execute user instructions in the browser
- Fine-tuning multimodal models to understand web page interfaces via screenshots and video
- Evaluating AI multi-turn conversational ability during information retrieval on websites
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 406.4K
- Likes
- 8
Models trained on it
- McGill-NLP/fuyu-8b-weblinxText generation 102
- McGill-NLP/Sheared-LLaMA-2.7B-weblinxText generation 47
- McGill-NLP/flan-t5-xl-weblinxText generation 46
- McGill-NLP/Sheared-LLaMA-1.3B-weblinxText generation 39
- McGill-NLP/MindAct-xl-weblinxText generation 37
- McGill-NLP/MindAct-base-weblinxText generation 25
- McGill-NLP/flan-t5-base-weblinxText generation 22
- McGill-NLP/pix2act-base-weblinxText generation 22
How to load it
from datasets import load_dataset
ds = load_dataset("McGill-NLP/WebLINX-full")