Datasets:
Convert dataset to Parquet
Convert dataset to Parquet.
Hi @Davlan π hope you're well! Since datasets 4.0 dropped loading scripts, afriqa-gold-passages no longer loads, which breaks the AfroBench AfriQA tasks in lm-evaluation-harness (45 tasks, see EleutherAI/lm-evaluation-harness#3171). This is the standard HF Parquet conversion: same configs, splits, columns and rows, just without the script. I checked every config and split against what the script yields and they're identical. One note: wol is left out, because the script declares a wol config but has no wol entry in _LANG_2_PIVOT/_LANG_2_SPLITS, so it never loaded (KeyError) even before 4.0.
Would you mind merging this when you have a moment, or pointing me to whoever looks after the masakhane org? There are matching PRs on masakhapos and masakhaner-x π