Datasets:

jamescalam

/

unsplash-25k-photos

like 45

To fix the preview, the dataset must work in streaming mode. It's not working right now because new_url + "/file" is not the correct way to point to a file in an online ZIP file.
We use chained URLs in streaming mode, they look like this: zip://path/to/file::https://foo.bar/data.zip

We extended functions like os.path.join to support chained URLs, therefore you can just use os.path.join instead. It joins the inner path inside the ZIP archive instead of appending to the URL.