The LeoLM/HellaSwag_de dataset, but removed rows where the number of endings differs between English and German.
The fix applied:
import datasets as hfds
ds = hfds.load_dataset("LeoLM/HellaSwag_de")
ds = ds.filter(lambda row: len(row["endings"]) == len(row["endings_de"]))
ds.push_to_hub("maxidl/LeoLM-HellaSwag_de-fixed")
This removes 7 examples from the validation set.
