Linguini is a benchmark to measure a language model’s linguistic reasoning skills without relying on pre-existing language-specific knowledge, based on the International Linguistic Olympiad problems. - View it on GitHub