We recently released Belebele, a first-of-its-kind multilingual reading comprehension dataset. It's parallel for 122 language variants, enabling direct comparison of how well models understand different languages. Dataset

→ View original post on X — @aiatmeta