Datasets:

Modalities:
Image
Text
Formats:
parquet
Languages:
English
ArXiv:
Libraries:
Datasets
Dask
License:

For WebSight v0.2, what would you like to see being improved or integrated?

#7
by VictorSanh - opened

No such thing as a stupid request!

Have you seen https://htmx.org/ ?

We have iterated on the requests and v0.2 is out!

A few noticeable improvements:

  • 💨 From traditional CSS to Tailwind CSS. Tailwind is CSS directly embedded in the HTML attribute class and is much more compact
  • 🚛 2M pairs of synthetic HTML webpages with their associated rendered screenshot, along with the prompt generated by an LLM to create that webpage
  • 🖼️ Much more visually appealing pages with the integration of real images

For v0.3 I think I'd like to see something better suited for our low-context local models.

What I find to be most useful is a more granular format. React component / Function level.

  • Identify each interactive element on the page and their user action + reaction. (Mouse click -> Displays dialog)
  • Categorize and label each navigational element on the page. (Pagination, header, ...)
  • Image of component
  • HTML/Markup/JSX of the component
  • Description of the component

As HTML can be quite large, and scraping/accessible navigation systems can't consume entire HTML on consumer context-levels too well, it would be beneficial to identify each component and its function within the pages context. They end up needing to truncate the markup and it makes a mess.

Sign up or log in to comment