Aya Evaluation Suite contains a total of 26,750 open-ended conversation-style prompts to evaluate multilingual open-ended generation quality.To strike a balance between language coverage and the quality that comes with human curation, we create an evaluation suite that includes:
human-curated examples in 7 languages (tur, eng, yor, arb, zho, por, tel) → aya-human-annotated.
machine-translations of handpicked examples into 101 languages → dolly-machine-translated.… See the full description on the dataset page:
https://huggingface.co/datasets/CohereLabs/aya_evaluation_suite.
