For multilingual general task instruction data, we incorporate ALPACA dataset (Taori et al., 2023), which consists of 52k English questions and corresponding response, and we obtain its foreign version with in-house translation engine. The six languages are Arabic (Ar), Greek (El), Hindi (Hi), Turkish (Tr), Vietnamese (Vi), Chinese (Zh).
Dataloader name:
multilingual_alpaca/multilingual_alpaca.py
DataCatalogue: http://seacrowd.github.io/seacrowd-catalogue/card.html?multilingual_alpaca