A team of researchers in Saudi Arabia has compiled and released a dataset to teach large language models to better understand and follow instructions in Arabic. The Culturally Relevant Instruction Dataset For Arabic (CIDAR) contains 10,000 instruction and output pairs representing the Arab region. The researchers hope the dataset will enhance how effective language models work in the cultural context of Arabic language instructions.
Understanding Language Prompts in Arabic
previous post
