Datasets:

biglam

/

brill_iconclass

like 7

I had this setup using this approach (https://github.com/bigscience-workshop/lam/issues/30#issuecomment-1181701303) but found it was quite slow - if you have a suggestion for a better approach that would be awesome :)

The main goal is to enable the viewer here. This dataset is relatively small and can be easily downloaded in the non-streaming mode, so I think it's OK to ignore the streaming performance (btw, the current implementation delegates the ZIP extraction logic to fsspec in the streaming mode, which is better than doing this manually as it's done in the linked snippet)

Awesome, thanks for this 🤗