Datasets:

danielz01

/

xView2

like 0

Seems like a bug on our side. Looking at it!

The internal error is:

libcommon.parquet_utils.TooBigRows: Rows from parquet row groups are too big to be read: 464.13 MiB (max=286.10 MiB)

The size in bytes of the contents of the rows is above our limit (more than 400MB of JSON to show one page of the dataset viewer in this case). We see the contents on the main dataset page (https://huggingface.co/datasets/danielz01/xView2) because we apply "truncation" to the rows to always show something, at least. On the contrary, on the dedicated viewer page (https://huggingface.co/datasets/danielz01/xView2/viewer/competition/train), we don't apply this truncation because we want the user to be able to dig into the data.

At least, we should make it clear in the UI.

I opened an issue here: https://github.com/huggingface/datasets-server/issues/2139

Thanks! That's very helpful!

The issue has been fixed by this pull request: https://github.com/huggingface/datasets-server/pull/2144