Datasets:

sbx

/

superlim-2

like 10

meta.error_span and meta.confusion_pairs are parsed as a String at the moment even though they should be parsed as dictionaries. The problem is that in half the cases in the data, the meta.error_span and meta.confusion_pairs are empty dictionaries. I have not found a way in the datasets typing system to define an empty or optional typing value, so we might have to implement a workaround.

Code in question: https://huggingface.co/datasets/sbx/superlim-2/blob/main/superlim-2.py#L205