Datasets:

HakaiInstitute

/

mussel-seg-1024-1024

like 2

It does work for the test split, but not the validation or train splits. If there's something I should do on my end in terms of webdataset chunk size that would help, let me know.

Rows from parquet row groups are too big to be read: 400.03 MiB (max=286.10 MiB)

Some rows are above our current size limit. The issue is here: https://github.com/huggingface/dataset-viewer/issues/2215