Is this like for cleaning html data from say common crawl ?

30kb model is super impressive.

What’s the model architecture?